apache tinkerpop logo


Developer Documentation

This document contains information for TinkerPop developers, contributors, and community members. It focuses on technical information and other internal processes related to the project.

jan de bray


Contributions via GitHub pull requests are gladly accepted from their original author. By submitting any copyrighted material via pull request, email, or other means you agree to license the material under the project’s open source license and warrant that you have the legal authority to do so.

Ways to Contribute

gremlin apache While the concept of an open source contribution can refer to doing development work on the code base, there are many other ways outside of coding to contribute to Apache TinkerPop. Participating on the various mailing lists, offering ideas, reporting bugs, writing documentation are all welcome contributions to the project that help improve the TinkerPop. This section of the document is designed to help provide some structure for potential contributors and to give them ideas for how they could get started becoming more involved in the TinkerPop community.

As a quick acknowledgment, this section was influenced by Apache Spark’s well organized "Contributing to Spark" document.

Mailing List Participation

TinkerPop has two mailing lists: gremlin-users and dev. Subscribing to and participating on one or both of these mailing lists is a great way to contribute to TinkerPop. Helping users with their problems by answering their questions on gremlin-users is a massive help to the community and an easy way for the community to become familiar with the contributor.

The dev@tinkerpop.apache.org mailing list is where all design discussion, early feature announcements, release discussions and other similar communication takes place. Having insight into the project at this level, will yield a lot of information about the day-to-day course of TinkerPop and provides a way to help shape the direction of the project.

Testing Releases

Proposed TinkerPop releases are not official until after they are evaluated by the community. The process of evaluation occurs on the dev@tinkerpop.apache.org mailing list where the community is invited to review release artifacts and vote if they should be released or not. Anyone is free to comment and provide a vote (+1/+0/-1) on a release. Note that votes of non-PMC members are considered non-binding, but are certainly considered in the review process.

Reviewing Changes

Virtually all changes to TinkerPop’s source code are performed via GitHub pull requests. When a pull request is issued it undergoes a review process prior to being merged to a release branch. Anyone is free to comment and provide a vote (+1/+0/-1) on a pull request. Note that votes of non-committers are considered non-binding, but are certainly considered in the review process.

Writing Documentation

Proposing a documentation change is quite similar to providing a source code change, which is described below in more detail. Note that there is a difference between project documentation and the TinkerPop web site. The project documentation refers to the documentation that is published per release like, tutorials, reference documentation, and other similar pages. This content is not updated after release and remains static bound to that version. The TinkerPop web site refers to the static home page and its related content, such as Understanding Gremlin, Data System Providers and other similar pages. This content is not bound to any particular version and can be published at any time where new content simply replaces old content.

For both types of documentation, changes can be submitted via pull request. For project documentation, TinkerPop has a robust documentation system that is based on asciidoc. The content can be found in the docs/src. Documentation is version specific, so consider the appropriate branch on which to submit the pull request so that the documentation is reflective of the version it is tied to. To view generated documentation locally, read more about environment configurations in the Documentation Environment and Contributor Documentation sections.

For web site changes, the process is largely the same except that the documentation system is HTML based instead of Asciidoc. The content can be found in the source control tree at docs/site. The web site is always published from the master branch as it is not bound to a version, so there is no need to submit a pull request to any other branches besides that one. If the change to this documentation involves changing provider listings in some way, first propose those changes on the dev@tinkerpop.apache.org mailing list and achieve consensus (wait time is 72 hours as usual). At that point, pull requests or direct changes can be submitted. Again, see the Documentation Environment section for more information on how to generate the site locally.

Reporting Bugs

Well formed bug reports (especially when accompanied by a pull request that fixes the problem) are welcome contributions that ultimately help improve the quality of TinkerPop. Of course, a bug report is only as good as its reproduction steps and the surrounding details of the issue. When a bug is encountered, it should be documented in JIRA, where it will likely trigger some discussion and ultimately a resolution.

TinkerPop Libraries

TinkerPop is more than just the core source code repository. It is a rich ecosystem of libraries developed by providers, who look to use or extend upon the TinkerPop interfaces. TinkerPop is highly focused on its interfaces and a small set of reference implementations of those interfaces as part of its core code base. As such, large and independent new functionality is often rejected from inclusion in TinkerPop itself, however, it is very much encouraged that such a project stand on its own as separate repository. Projects like this that meet the TinkerPop listing policy can be added to the TinkerPop web site and promoted there. Please see the current listing of such libraries on the TinkerPop home page to get an idea as to what kinds of libraries are available from the community. When in doubt about whether a library might be a good fit for hosting and ongoing maintenance in the TinkerPop project itself, please ask on the dev@tinkerpop.apache.org mailing list.

Considering Code Changes

Code changes can take a fair bit of effort from both the contributor doing the work, as well as the people who will be reviewing and testing that work. It is beneficial to all involved that some basic considerations be made by the contributor to ensure the best use is made of everyone’s time.

For those who are trying to find a place to start to contribute, consider looking at unresolved issues that have the "trivial" priority as these issues are specifically set aside as low-hanging fruit for newcomers.

Before proceeding, contributors should evaluate if the proposed change is likely to be relevant, new and actionable:

  • Is it clear that code must change? Proposing a JIRA issue and pull request is appropriate only when a clear problem or change has been identified. When in doubt, email dev@tinkerpop.apache.org first about the possible change.

  • Search the mailing list archives for related discussions. Often, the problem has been discussed before, with a resolution that doesn’t require a code change, or recording what kinds of changes will not be accepted as a resolution.

  • Search JIRA for existing issues.

  • Is the scope of the change matched to the contributor’s level of experience? Anyone is qualified to suggest a typo fix, but refactoring a core feature in serialization or OLAP requires much more understanding of TinkerPop. Some changes require building up experience first.

Before considering how to contribute code, it is useful to understand how code is reviewed, and why changes may be rejected. Simply put, changes that have many or large positives, and few negative effects or risks, are much more likely to be merged, and merged quickly. Risky and less valuable changes are very unlikely to be merged, and may be rejected outright rather than receive iterations of review.


  • Change has already been discussed and is known to committers

  • Fixes the root cause of a bug in existing functionality

  • Adds functionality or fixes a problem needed by a large number of users

  • Simple, targeted

  • Easily tested; has tests

  • Reduces complexity and lines of code

Negatives, Risks

  • Band-aids a symptom of a bug only

  • Introduces complex new functionality, especially an API that needs to be supported

  • Adds complexity that only helps a niche use case

  • Adds user-space functionality that does not need to be maintained in TinkerPop, but could be hosted externally and promoted in provider listings

  • Changes a public API or semantics thus introducing a breaking change

  • Adds large dependencies

  • Changes versions of existing dependencies

  • Adds a large amount of code

  • Makes lots of modifications in one "big bang" change

Contributing Code Changes

Generally, TinkerPop uses JIRA to track logical issues, including bugs and improvements, and uses GitHub pull requests to manage the review and merge of specific code changes. That is, JIRA issues are used to describe what should be fixed or changed, and high-level approaches, and pull requests describe how to implement that change in the project’s source code.

As a first step to making a contribution, consider JIRA:

  1. Find the existing JIRA ticket that the change pertains to.

    1. Do not create a new ticket if creating a change to address an existing issue in JIRA; add to the existing discussion and work instead.

    2. Look for existing pull requests that are linked from the ticket, to understand if someone is already working on the JIRA.

  2. If the change is new, then it usually needs a new ticket. However, trivial changes, where the change is virtually the same as the how it should change do not require a JIRA (e.g. "Fix typos in Foo javadoc").

  3. If required, create a new ticket:

    1. Provide a descriptive Title and a detailed Description. For bug reports, this should ideally include a short reproduction of the problem.

    2. Set required fields - these are detailed later in this document in the Issue Tracker Conventions section.

  4. If the change is a large change, consider inviting discussion on the issue at dev@tinkerpop.apache.org first before proceeding to implement the change.

Next, make changes and prepare a pull request:

  1. Fork and then clone the Apache TinkerPop GitHub repository if not already done.

  2. Make changes in the fork

    1. It is typically best to create a branch for the changes. Consider naming that branch after the JIRA issue number to easily track what that branch is for.

    2. Consider which release branch (e.g. master, 3.5-dev - consult the Branches Section for more information) to create the development branch from in the first place. In other words, is the change to be targeted at a specific TinkerPop version (e.g. a patch to an older version)? When in doubt, please ask on dev@tinkerpop.apache.org.

  3. Build the project and run tests.

    1. A simple build can be accomplished with maven: mvn clean install.

    2. Often, a "simple build" isn’t sufficient and integration tests are required: mvn clean install -DskipIntegrationTests=false -DincludeNeo4j.

    3. Docker can help simplify building and testing: docker/build.sh -t -i -n

    4. Please see the Building and Testing section for more building and testing options.

  4. Consider whether documentation or tests need to be added or updated as part of the change, and add them as needed.

    1. Nearly all changes should include a modification to the CHANGELOG.asciidoc file - one or more entries to help summarize the change.

    2. Some changes will require updates to the Upgrade Documentation. Updates to this documentation are usually reserved for major new features and breaking changes.

    3. Docker can help simplify documentation generation: docker/build.sh -d

    4. Please see the Building and Testing section for more documentation generation options.

  5. Open the pull request against the appropriate branch on the Apache TinkerPop repository.

    1. Target the pull request at the appropriate branch in TinkerPop’s repository

    2. Prefix the name of the pull request with the JIRA issue number (include a brief description after that).

    3. Include a link to the ticket in JIRA in the pull request description.

    4. Include a rough synopsis of how the changes were tested. This might be as simple as "Ran mvn clean install to success and performed manual testing in the Gremlin Console".

    5. Include other descriptive elements about the change if they are not already included in the JIRA ticket.

    6. Automated builds will occur with Travis. Please be sure that the pull request passes those builds and correct them if there are problems.

Once the pull request has been placed it will go into review:

  1. Other reviewers, including committers, may comment on the changes and suggest modifications. Changes can be added by simply pushing more commits to the same branch.

  2. Lively, polite, rapid technical debate is encouraged from everyone in the community. The outcome may be a rejection of the entire change.

  3. Reviewers can indicate that a change looks suitable for merging with by providing a "+1". Please see the Review then Commit process for more details.

  4. Sometimes, other changes will be merged which conflict with your pull request’s changes. The PR can’t be merged until the conflict is resolved. In these cases the PR must be rebased, conflicts resolved and the resulting changes force pushed back to the branch.

  5. Try to be responsive to the discussion rather than let days pass between replies.

  6. Reviewers may request additional pull requests (e.g. one for each release branch that it may be related to) if the changes brings extensive conflict between branches.

On successful review, the pull request will be merged to the main repository and the JIRA issue will be closed.


TinkerPop version numbers follow a format of EPOCH.MAJOR.MINOR.PATCH, where:

  • EPOCH version is incremented for epochs of thinking around TinkerPop and its relation to the graph ecosystem and is part of a complete release.

  • MAJOR version is incremented for incompatible changes and is part of a complete release.

  • MINOR version is incremented for backward compatible changes and is part of a complete release.

  • PATCH version is incremented for backward compatible bug fixes and typically only applied to client modules for partial release.

The above definitions refer to notions of "complete" and "partial" releases. A complete release is one that releases all TinkerPop components across all languages. A partial release is one that releases just a single module of a driver/client. A partial release is helpful in that it decouples the release of client/driver modules from what are typically server modules. In this way, for example, a small fix can be applied and released around .NET without having to release all the other components which may not have experienced any real change.

The number for EPOCH rarely changes and when it does, it typically represents a complete re-write of the software and thus massive changes can be expected. As a result of the definition for EPOCH, there is greater focus on the other three numbers when considering where changes will land:

  1. A client with a specific EPOCH.MAJOR.MINOR will be backward compatible with user application code written for EPOCH.MAJOR.MINOR.

  2. A client with specific EPOCH.MAJOR will be backward compatible with the EPOCH.MAJOR version of the server.

A complete release is always just EPOCH.MAJOR.MINOR where the PATCH is inferred as "0", thus 3.6.x line would initially release all components as 3.6.0. A PATCH number must start with "1" within a release because of how Nuget treats "0". If it was determined that gremlin-python needed a special release then it could be released independently of the other modules as

In short, the dev mailing list is open to discussion on where a change will land. The discussion just needs to have good reasoning and use the general principles above as a framework for coming to a consensus.

Development Environment

TinkerPop is fairly large body of code spread across many modules and covering multiple programming languages. Despite this complexity, it remains relatively straightforward a project to build. This following subsections explain how to configure a development environment for TinkerPop.


System Configuration

At a minimum, development of TinkerPop requires Java 8 but it is preferable to use Java 11 cross-compiled to Java 8 (the cross-compilation happens automatically as part of the build). Starting with 3.7.0, you can also build with Java 17. Note, however, that there are some issues with deep reflection so there are --add-opens JVM options included in the pom files to enable this to work. Maven (requiring a minimum of Maven 3.5.3+) is used as the common build system, which even controls the builds of non-JVM GLVs such as gremlin-python. Java and Maven are described as a "minimum" for a development environment, because they will only build JVM portions of TinkerPop and many integration tests will not fire with this simple setup. It is possible to get a clean and successful build with this minimum, but it will not be possible to build non-JVM aspects of the project and those will go untested.

To gain the ability to execute all aspects of the TinkerPop build system, other environmental configurations must be established. Those prerequisites are defined in the following subsections.

As of TinkerPop 3.5.5, environment configuration for Gremlin Language Variants can be optional, as Docker becomes a system requirement to build and test GLVs inside of Maven. Please make sure Docker is installed and running on your system. You will need to install both Docker Engine and Docker Compose, which are included in Docker Desktop.

Use Java 11 for documentation generation with bin/process-docs.sh and for other build features outside of the basic mvn clean install sort of function.
For those who intend to offer a contribution, building with a minimal configuration may not be sufficient when submitting a pull request. Consider setting up the full environment.
Consider using SDKMAN! to manage Java and Maven versions and environments.
For those using Windows, efforts have been made to keep the build OS independent, but, in practice, it is likely that TinkerPop’s build system will only allow for a minimum build at best.

Refer to Building On Windows section for more details.

Groovy Environment

Groovy is not used in the standard build, but when generating documentation it does require the loading of a Gremlin Console instance. The Gremlin Console is Groovy-based and the documentation bootstrapping loads TinkerPop plugins which requires proper configuration of Graph/Ivy dependency loaders as described in the Gremlin Applications Section of the Reference Documentation.

The base configuration described in that link may need to be modified if there is a desire to work with the Gremlin Console (for documentation generation or just general testing) in a way that utilizes SNAPSHOT releases in the Apache Snapshots Repository. In that case, the grapeConfig.xml will need to include a resolver for that repository and the basic Ivy configuration will look as follows:

  <settings defaultResolver="downloadGrapes"/>
    <chain name="downloadGrapes" returnFirst="true">
      <filesystem name="cachedGrapes">
        <ivy pattern="${user.home}/.groovy/grapes/[organisation]/[module]/ivy-[revision].xml"/>
        <artifact pattern="${user.home}/.groovy/grapes/[organisation]/[module]/[type]s/[artifact]-[revision](-[classifier]).[ext]"/>
      <ibiblio name="localm2" root="${user.home.url}/.m2/repository/" checkmodified="true" changingPattern=".*" changingMatcher="regexp" m2compatible="true"/>
      <ibiblio name="jcenter" root="https://jcenter.bintray.com/" m2compatible="true"/>
      <ibiblio name="ibiblio" m2compatible="true"/>
      <ibiblio name="apache-snapshots" root="http://repository.apache.org/snapshots/" m2compatible="true"/>

The above configuration is just a modification of the default. Perhaps the most lean common configuration might just be:

  <settings defaultResolver="downloadGrapes"/>
    <chain name="downloadGrapes">
      <ibiblio name="local" root="file:${user.home}/.m2/repository/" m2compatible="true"/>
      <ibiblio name="central" root="https://repo1.maven.org/maven2/" m2compatible="true"/>

In the above case, the configuration largely relies on the standard Maven builds to create a well cached .m2 directory. Under typical development circumstances, SNAPSHOT will find themselves deployed there locally and that is all that will be required for Grape to do its work.

As a final word, it is important to take note of the order used for these references as Grape will check them in the order they are specified and depending on that order, an artifact other than the one expected may be used which is typically an issue when working with SNAPSHOT dependencies.

Documentation Environment

The documentation generation process is not Maven-based and uses shell scripts to process the project’s asciidoc. The scripts should work on Mac and Linux. Javadocs should be built using Java 11.

We recommend performing documentation generation on Linux. For the scripts to work on Mac, you will need to install GNU versions of the utility programs via homebrew, e.g.grep, awk, sed, findutils, and diffutils.

To generate documentation, it is required that Hadoop 3.3.x is running in pseudo-distributed mode. Be sure to set the HADOOP_GREMLIN_LIBS environment variable as described in the reference documentation. It is also important to set the CLASSPATH to point at the directory containing the Hadoop configuration files, like mapred-site.xml.

The /etc/hadoop/yarn-site.xml file prefers this configuration over the one provided in the Hadoop documentation referenced above:


The /etc/hadoop/mapred-site.xml file prefers the following configuration:


Also note that awk version 4.0.1 is required for documentation generation. The YARN recipe also uses the zip program to create an archive so that needs to be installed, too, if you don’t have it already.

The Hadoop 3.3.x installation instructions call for installing pdsh but installing that seems to cause permission problems when executing sbin/start-dfs.sh. Skipping that prerequisite seems to solve the problem.

Documentation can be generated locally with:


Documentation is generated to the target/docs directory. It is also possible to generate documentation locally with Docker. docker/build.sh -d.

The installation of plugins sometimes fails in this step with the error: Error grabbing grapes - download failed. It often helps in this case to delete the directories for the dependencies that cannot be downloaded in the .m2 (~/.m2/) and in the grapes (~/.groovy/grapes/) cache. E.g., if the error is about asm#asm;3.2!asm.jar, then remove the asm/asm sub directory in both directories.
Unexpected failures with OLAP often point to a jar conflict that arises in scenarios where Hadoop or Spark dependencies (or other dependencies for that matter) are modified and conflict. It is not picked up by the enforcer plugin because the inconsistency arises through plugin installation in Gremlin Console at document generation time. Making adjustments to the various paths by way of the <manifestEntries> on the jar given the functionality provided by the DependencyGrabber class which allows you to manipulate (typically deleting conflicting files from /lib and /plugin) plugin loading will usually resolve it, though it could also be a more general environmental problem with Spark or Hadoop. The easiest way to see the error is to simply run the examples in the Gremlin Console which more plainly displays the error than the failure of the documentation generation process.

To generate the web site locally, there is no need for any of the above infrastructure. Site generation is a simple shell script:


The site will be generated to the target/site/home directory.

Python Environment

As of TinkerPop 3.2.2, the build optionally requires Python to build the gremlin-python module. If Python is not installed, TinkerPop will still build with Maven, but native Python tests and Java tests that require Python code will be skipped. Developers should also install pip and virtualenv (version 15.0.2 - older versions may cause build failures).

The build expects Python a python3 installation which should be 3.5.3 or better. Python also tests kerberos and therefore requires:

sudo apt install libkrb5-dev krb5-user

As of TinkerPop 3.5.5, gremlin-python uses Docker for all tests inside of Maven, and Python installation will not be required to run gremlin-python through Maven. Please make sure Docker is installed and running on your system.

Once the Python environment is established, the full building and testing of gremlin-python may commence. It can be done manually from the command line with:

mvn clean install -Pglv-python

which enables the "glv-python" Maven profile or in a more automated fashion simply add a .glv file to the root of the gremlin-python module which will signify to Maven that the environment is Python-ready. The .glv file need not have any contents and is ignored by Git. A standard mvn clean install will then build gremlin-python in full.

The .glv file in gremlin-python also activates the "console-integration-tests" Maven profile to run gremlin-console integration tests. Alternatively, this profile can be activated manually. Note that unlike gremlin-python the tests are actually integration tests and therefore must be actively switched on with -DskipIntegrationTests=false:

mvn clean install -pl gremlin-console -DskipIntegrationTests=false
For those who do not have a full Maven environment, please see this section for how Docker can be used to help run tests.
Consider use of pyenv to better manage Python versions and environments.

See the Release Environment section for more information on release manager configurations.

DotNet Environment

The build optionally requires .NET SDK (>=8.0) to work with the gremlin-dotnet module. If .NET SDK is not installed, TinkerPop will still build with Maven, but .NET projects will be skipped.

As of TinkerPop 3.5.5, gremlin-dotnet uses Docker for running all test projects inside of Maven, and .NET SDK will not be required to run gremlin-dotnet tests through Maven. Please make sure Docker is installed and running on your system.

gremlin-dotnet can be built and tested from the command line with:

mvn clean install -Pgremlin-dotnet

which enables the "gremlin-dotnet" Maven profile or in a more automated fashion simply add a .glv file to the src and test directories of the gremlin-dotnet module which will signify to Maven that the environment is .NET-ready. The .glv file need not have any contents and is ignored by Git. A standard mvn clean install will then build gremlin-dotnet in full.

In order to pack the Gremlin.Net.Template project, it is also necessary to install Mono. The template can still be built and tested without Mono but packing will be skipped. To pack the template (which will also download the NuGet CLI tool) the nuget property has to be set:

mvn clean install -Dnuget
For those who do not have a full Maven environment, please see this section for how Docker can be used to help run tests.

See the Release Environment section for more information on release manager configurations.

JavaScript Environment

When building gremlin-javascript, mvn command will include a local copy of Node.js runtime and npm inside your project using com.github.eirslett:frontend-maven-plugin plugin. This copy of the Node.js runtime will not affect any other existing Node.js runtime instances in your machine.

To run the development and build scripts of gremlint and its corresponding web page docs/gremlint, Node.js and npm have to be installed. When generating or publishing the TinkerPop website, the docs/gremlint web page has to be built. Consequently, the scripts bin/generate-home.sh and bin/publish-home.sh require that Node.js and npm are installed. Version 8.x or newer of npm is required. This is covered in more detail in the Site section.

As of TinkerPop 3.5.5, gremlin-javascript uses Docker for all tests inside of Maven. Please make sure Docker is installed and running on your system.

Beware of unexpected or unwanted changes on package-lock.json files when committing and merging.
For those who do not have a full Maven environment, please see this section for how Docker can be used to help run tests.
Consider using nvm to manage node.js versions and environments.

See the Release Environment section for more information on release manager configurations.

Go Environment

The build optionally requires Go (>=1.21) to work with the gremlin-go module. Creating an empty .glv file will enable running of tests inside of Maven. If .glv file does not exist, TinkerPop will still build with Maven, but Go projects will be skipped.

gremlin-go can be built and tested from the command line with:

mvn clean install -pl gremlin-go

Alternatively, after installing Go, gremlin-go can be built from the command line with:

go build

Docker allows you to test the driver without installing any dependencies. The following command can be used to run docker:

docker-compose up --exit-code-from gremlin-go-integration-tests

See the Release Environment section for more information on release manager configurations.

Docker Environment

The build optionally requires Docker to build Docker images of Gremlin Server and Gremlin Console. The Docker images can be built from the command line with:

mvn clean install -pl gremlin-server,gremlin-console -DdockerImages

which enables the "docker-images" Maven profile.

If confronted with "Permission denied" errors on Linux, it may be necessary to do the following:

sudo groupadd docker
sudo usermod -aG docker $USER
newgrp docker
sudo chmod 666 /var/run/docker.sock

As of TinkerPop 3.5.5, a docker image of the Gremlin Server will be built automatically with mvn clean install, which is use for GLV tests inside of Docker. To skip building this image, append the -DskipImageBuild flag to Maven commands.

Release Environment

This section is only useful to TinkerPop release managers and describes prerequisites related to deploying an official release of TinkerPop.

Maven needs to be configured to deploy maven artifacts. Apache LDAP credentials can be used for this. Release managers should encrypt their Apache LDAP password as described in the Apache Maven docs. The encrypted password can then be configured in the settings.xml as described in the section "Set up your development environment" of this Apache Infra article. This configuration will be used by mvn deploy.

For Python releases, uploading to pypi uses twine which is automatically installed by the build process in maven. Twine refers to HOME/.pypirc file for configuration on the pypi deploy environments and username and password combinations. The file typically looks like this:


username = <username>
password =

username = <username>
password =

The approach above assumes basic authentication and a blank password will in .pyirc will have Maven prompt for a password. That said, it is preferred that your account be secured with 2FA, in which case, basic authentication will no longer be allowed. Instead, an API token is required. It can be generated under your account settings and then the username in .pyirc is set to token and the password to the token value. Please see PyPI documentation for more details.

For .NET releases, install Mono. The release process is known to work with 6.12.0, so it is best to probably install that version. Release managers should probably also do an install of nuget 3.4.4 as it will help with environmental setup. To get an environment ready to deploy to NuGet, it is necessary to have a NuGet API key. First, create an account with nuget and request that a PMC member add your account to the Gremlin.Net and the Gremlin.Net.Template package in nuget so that you can deploy. Next, generate an API key for your account on the nuget website. The API key should be added to NuGet.Config with the following:

mono nuget.exe setApiKey [your-api-key]

This should update ~/.config/NuGet/NuGet.Config a file with an entry containing the encrypted API key. On mvn deploy, this file will be referenced on the automated nuget push.

To deploy JavaScript / TypeScript artifacts on the npm registry, the release manager must set the authentication information on the ~/.npmrc file. The easiest way to do that is to use the npm adduser command. This must be done only once, as the auth token doesn’t have an expiration date and it’s stored on your file system. If this account is newly created then request that a PMC member add your account to the "gremlin" package on npm.

Deploying Docker images to Docker Hub requires an account that is a member of the TinkerPop organization. So if you don’t already have an account on Docker Hub then create one and request that a PMC member adds your account to the TinkerPop organization. Afterwards, authentication information needs to be added to the ~/.docker/config.json file. This information can simply be added with the docker login command which will ask for credentials. This must be done only once. Finally, docker push can be used to push images to Docker Hub which will be done automatically on mvn deploy or it can be triggered manually with mvn docker:push.

Building and Testing

The following commands are a mix of Maven flags and shell scripts that handle different build operations

  • Build project: mvn clean install

    • Build a specific module (e.g. gremlin-server) within the project: mvn clean install -pl gremlin-server

    • Build without assertions for "iterator leaks" which are enabled by default: mvn clean install -DtestIteratorLeaks=false

    • Specify specific tests in a TinkerPop Suite to run with the GREMLIN_TESTS environment variable, along with the Maven project list argument, e.g.:

      export GREMLIN_TESTS='org.apache.tinkerpop.gremlin.process.traversal.step.map.PathTest$Traversals,org.apache.tinkerpop.gremlin.process.traversal.PathTest'
      mvn -Dmaven.javadoc.skip=true --projects tinkergraph-gremlin test
    • Clean the .groovy/grapes/org.apache.tinkerpop directory on build: mvn clean install -DcleanGrapes

    • Turn off "heavy" logging in the "process" tests: mvn clean install -DargLine="-DmuteTestLogs=true"

    • The test suite for neo4j-gremlin is disabled by default - to turn it on: mvn clean install -DincludeNeo4j

  • Regenerate toy graph data (only necessary given changes to IO classes): mvn clean install -Dio from tinkergraph-gremlin directory

    • If there are changes to the Gryo format, it may be necessary to generate the Grateful Dead dataset from GraphSON (see IoDataGenerationTest.shouldWriteGratefulDead)

  • Start Gremlin Server with Docker using the standard test configuration: docker/gremlin-server.sh

  • Check license headers are present: mvn apache-rat:check

  • Build AsciiDocs (see Documentation Environment): bin/process-docs.sh

    • Build AsciiDocs (but don’t evaluate code blocks): bin/process-docs.sh --dryRun

    • Build AsciiDocs (but don’t evaluate code blocks in specific files): bin/process-docs.sh --dryRun docs/src/reference/the-graph.asciidoc,docs/src/tutorial/getting-started,…​

    • Build AsciiDocs (but evaluate code blocks only in specific files): bin/process-docs.sh --fullRun docs/src/reference/the-graph.asciidoc,docs/src/tutorial/getting-started,…​

    • Process a single AsciiDoc file: docs/preprocessor/preprocess-file.sh `pwd`/gremlin-console/target/apache-tinkerpop-gremlin-console-*-standalone "" "*" `pwd`/docs/src/xyz.asciidoc

  • Build JavaDocs/JSDoc: mvn process-resources -Djavadoc

    • Javadoc to target/site/apidocs directory

    • JSDoc to the gremlin-javascript/src/main/javascript/gremlin-javascript/doc/ directory

  • Specify the seed used for Random in tests mvn clean install -DtestSeed - useful when a test fails, the seed will be printed in the build output so that the test can run with the same version of random (look for "TestHelper" logger in output)

  • Check for newer dependencies: mvn versions:display-dependency-updates or mvn versions:display-plugin-updates

  • Check the effective pom.xml: mvn -pl gremlin-python -Pglv-python help:effective-pom -Doutput=withProfilePom.xml

  • Deploy JavaDocs/AsciiDocs: bin/publish-docs.sh svn-username

  • Integration Tests: mvn verify -DskipIntegrationTests=false

    • Execute with the -DincludeNeo4j option to include neo4j-gremlin tests. As of 3.7.0, transactional tests on Gremlin Server, Gremlin Driver, and GLVs will run automatically with integration tests against the TinkerTransactionGraph.

    • Execute with the -DuseEpoll option to try to use Netty native transport (works on Linux, but will fallback to Java NIO on other OS).

  • Benchmarks: mvn verify -DskipBenchmarks=false

    • Reports are generated to the console and to gremlin-tools/gremlin-benchmark/target/reports/benchmark.

  • Test coverage report: mvn clean install -Dcoverage - note that the install is necessary because report aggregation is bound to that part of the lifecycle.

    • Reports are generated to gremlin-tools/gremlin-coverage/target/site.

  • cd site

    • Generate web site locally: bin/generate-home.sh

    • Publish web site: bin/publish-home.sh <username>

Building On Windows

The following steps must be taken in order to build TinkerPop on Windows:

  1. Install winutils for Hadoop

    • Download the latest version of winutils binaries for Hadoop. The binaries winutils.exe and hadoop.dll are required.

      • Note that these libraries require Microsoft Visual C Redistributable 2015-2022 to be installed. We've tested it on Windows 10 and Windows 11 with Microsoft Visual C Redistributable 2015-2022 14.32.31326.

    • Place contents of the bin folder on your local driver in the following folder structure:

      • e.g. hadoop-3.3.1/bin/winutils.exe

    • Set HADOOP_HOME to point to the hadoop-3.3.1 folder

    • Add %HADOOP_HOME%\bin to your PATH

  2. Run mvn clean install from root of tinkerpop

  3. Follow IDE specific steps if applicable:

You should now be able to work with TinkerPop on Windows.

Docker Integration

TinkerPop provides a shell script, that can start several build tasks within a Docker container. The required Docker images will be built automatically if they don’t exist yet. Thus the first invocation of the Docker script is expected to take some time.

The script can be found under PROJECT_HOME/docker/build.sh. The following tasks are currently supported:

  • run standard test suite

  • run integration tests

  • build Java docs

  • build user docs

A list of command line options is provided by docker/build.sh --help. The container will install, configure and start all required dependencies, such as Hadoop.

By default, this script will run every module in the project. However, if you are planning on working on just a small set of the modules (e.g. the GLVs) then you can use the script options to reduce the modules included by specifically selecting which modules you want. This behavior is currently supported for the non-Java GLVs and gremlin-console. This option will include only the selected modules as well as gremlin-server, gremlin-test, neo4j-gremlin and all their dependencies. This is the minimum set of modules required to build and test the GLVs.

Build and test gremlin-python and gremlin-go with minimal Gremlin Server dependencies
./docker/build.sh --tests --integration-tests --python --golang

Options can be passed to Docker by setting the TINKERPOP_DOCKER_OPTS environment variable. A speed boost can be gained at the expense of memory by using tmpfs and the special directory /usr/src/tinkermem.

Build in-memory
TINKERPOP_DOCKER_OPTS="--tmpfs /usr/src/tinkermem:exec,mode=0755,rw,noatime,size=2000m"
Disable IPv6 for Hadoop
TINKERPOP_DOCKER_OPTS="--sysctl net.ipv6.conf.all.disable_ipv6=1 --sysctl net.ipv6.conf.default.disable_ipv6=1"

A custom maven settings.xml can be supplied, for example, to point to a local proxy. Copy the settings.xml to the PROJECT_HOME/ directory. The Docker script will detect and copy it to the running container.

If the container is used to generate the user docs, it will start a web server and show the URL that is used to host the HTML docs.

After finishing all tasks, the script will immediately destroy the container.

Docker can also be helpful to developers who do not want to run tests from a Maven environment, which may be a bit opaque when dealing with test failures and largely unhelpful for debugging. This situation is typically case for developers doing work on Gremlin Language Variants (e.g. Python). To help alleviate this problem, developers can start a standalone Gremlin Server with its standard test configuration that is used in the standard Maven build.

Generally speaking, most developers will want to test their code against the latest build of Gremlin Server in the TinkerPop repository. To do that, first be sure to build a Docker image of the current code:

mvn clean install -DskipTests

Next, generate the a Docker image for Gremlin Server with:

mvn clean install -pl :gremlin-server -DdockerImages -DskipTests
If changes are made to the repository that need to be reflected in the Gremlin Server Docker image then the old image should be removed and then the above commands re-executed.

Finally, start the server with:


Starting Gremlin Server this way makes it possible to run Gremlin Language Variant tests without Maven (for example, directly from a debugger) which should greatly reduce development friction for these environments.

It is also possible to specify the exact version of Gremlin Server to run with the test configuration. This version should be an existing Docker image version and must be an explicit version that maps to an actual TinkerPop artifact:

docker/gremlin-server.sh 3.4.2

To be a bit more clear, the version can not be a Docker tag like "latest" because there is no such TinkerPop artifact that has been published with that version number.

Testing Sub-Modules with Docker

Currently the modules gremlin-go, gremlin-javascript, gremlin-dotnet, gremlin-python and gremlin-console can be tested through Docker.

Please make sure Docker is installed and running on your system. You will need to install both Docker Engine and Docker Compose, which are included in Docker Desktop.

The following environment variables used by Docker Compose will automatically be set when running through Maven.

The docker compose environment variable GREMLIN_SERVER specifies the Gremlin server docker image to use, i.e. an image with the tag tinkerpop/gremlin-server:$GREMLIN_SERVER, and is a required environment variable. This also requires the specified docker image to exist, either locally or in Docker Hub.

Running mvn clean install -pl gremlin-server -DskipTests -DskipIntegrationTests=true -Dci -am in the main tinkerpop directory will automatically build a local SNAPSHOT Gremlin server image. If your OS Platform cannot build a local SNAPSHOT Gremlin server through maven, it is recommended to use the latest released server version from Docker Hub (do not use GREMLIN_SERVER=latest, use actual version number, e.g. GREMLIN_SERVER=3.5.x or GREMLIN_SERVER=3.6.x).

The docker compose environment variable HOME specifies the user home directory for mounting volumes during test image set up. This variable is set by default in Unix/Linux, but will need to be set for Windows, for example, run $env:HOME=$env:USERPROFILE in PowerShell.

There are different ways to launch the test suite and set the GREMLIN_SERVER environment variable depending on your Platform:

  • Run Maven commands, e.g. mvn clean install inside of project folder e.g. tinkerpop/gremlin-go, or mvn clean install -pl gremlin-go inside of tinkerpop (platform-agnostic - recommended)

  • Add GREMLIN_SERVER=<server-image-version> and HOME=<user-home-directory> to an .env file inside project folder and run docker-compose up --exit-code-from gremlin-go-integration-tests (Platform-agnostic).

  • Run GREMLIN_SERVER=<server-image-version> docker-compose up --exit-code-from gremlin-go-integration-tests in Unix/Linux.

  • Run $env:GREMLIN_SERVER="<server-image-version>";$env:HOME=$env:USERPROFILE;docker-compose up --exit-code-from gremlin-go-integration-tests in Windows PowerShell.

You should see exit code 0 upon successful completion of the test suites. Run docker-compose down to remove the service containers (not needed if you executed Maven commands or run.sh), or docker-compose down --rmi all to remove the service containers while deleting all used images.

Note for running docker with MacOS on ARM processors: Docker’s performance is extremely poor on ARM Mac’s in its default configuration. It is recommended to enable both the "New Virtualization Framework" and "VirtioFS" under Docker Desktop Settings → Experimental Features.

Intellij Usage

Most core TinkerPop developers are using Intellij for their work so this section helps describe the mechanisms for best working with it as an IDE.


Installation and basic configuration of Intellij is beyond the scope of this writing and it is assumed that the TinkerPop GitHub repository has been cloned and the root of the repository is open in Intellij. From there, we can begin to look at configuration options specifically relevant to TinkerPop itself.

TinkerPop has a module called gremlin-shaded which contains shaded dependencies for some libraries that are widely used and tend to introduce conflicts. To ensure that Intellij properly interprets this module after importing the Maven pom.xml perform the following steps:

  1. Build gremlin-shaded from the command line with mvn clean install.

  2. Right-click on the gremlin-shaded module in the project viewer of Intellij and select "Remove module". If this menu option is not available (as is the case in newer versions of Intellij - first noticed in 13.1.5), then open the "Maven Projects" side panel, right click the gremlin-shaded module and select "Ignore Project".

  3. In the "Maven Projects" Tool window and click the tool button for "Reimport All Maven projects" (go to View | Tool Windows | Maven Projects on the main menu if this panel is not activated).

  4. At this point it should be possible to compile and run the tests within Intellij, but in the worst case, use File | Invalidate Caches/Restart to ensure that indices properly rebuild.

Note that it may be necessary to re-execute these steps if the gremlin-shaded pom.xml is ever updated.

You will initially see lots of errors related to the Gremlin lexer/parser. The gremlin-language module requires ANTLR processing. While this processing is configured to execute with Maven, it can also be setup to generate parser files within Intellij itself on command:

  1. Install the ANTLR4 Grammar Plugin for Intellij

  2. Right-click on the Gremlin.g4 file and "Configure ANTLR"

  3. Set "Output directory where all output is generated" to target/generated-sources/antlr4

  4. Set "Grammar file encoding" to utf-8

  5. Set "Package/namespace for the generated code" to org.apache.tinkerpop.gremlin.language.grammar

  6. Set "Language" to Java

  7. Set "Case transformation in the Preview window" should be "Leave as-is"

  8. The "generate parse tree listener" should be unchecked and the "generate parse tree visitor" should be checked.

With these settings it should be possible to right-click Gremlin.g4 and "Generate ANTLR Recognizer" which will place the generated code in where specified at target/generated-sources/antlr4. Be sure to right-click the antlr4 directory and "Mark directory as" "Generated Sources Root" which should allow Intellij to recognize it.

The gremlin-groovy module uses a Java annotation processor to help support DSLs. Annotation processing in Intellij should be set up by the Maven import, but if this is set up incorrectly you will see a cryptic error message when building: java: Compilation failed: internal java compiler error. To fix this, search for the Intellij setting Annotation Processors. Make sure that annotation processing is enabled for gremlin-groovy but disabled for gremlin-annotations. This should fix the internal java compile error. Next you may see compile errors complaining that the jsr223 classes __, `CredentialTraversal, CredentialTraversalSource, and DefaultCredentialTraversal cannot be found. Those classes are generated by annotations. To fix these errors, make sure mark the directory gremlin-groovy/target/generated-sources/annotations as "Generated Sources Root".

Developers working on the neo4j-gremlin module should enabled the include-neo4j Maven profile in Intellij. This will ensure that tests will properly execute within the IDE.


It is generally assumed that JVM-based debugging of TinkerPop code in Intellij is a relatively straightforward task for most developers, but it is worth pointing out a few important points related to it and to drill into some specifics for the non-JVM languages.


There are generally no complexities to running the debugger for any JVM-based test in the repository, but the following tips are helpful to know when doing so:

  • It is not possible to run the tests in gremlin-test without an Graph implementation. For example, it is not possible to just right-click org.apache.tinkerpop.gremlin.process.traversal.step.filter.CoinTest and then select Debug 'CoinTest'. Instead, running that test would require opening tinkergraph-gremlin and executing the TinkerGraphProcessStandardTest (as an example) which runs the full Gremlin test suite to include CoinTest.

  • To run just CoinTest, set a GREMLIN_TESTS environment variable with the fully qualified path name to the test configuration in Intellij. Note that the fully qualified test name is really org.apache.tinkerpop.gremlin.process.traversal.step.filter.CoinTest$Traversals.

  • Gherkin tests have a similar pattern in that they require a Graph implementation to execute them. Therefore, debugging entails going to tinkergraph-gremlin and running TinkerGraphFeatureTest in the debugger.

  • It is possible to filter the Gherkin tests by adding a system property to the debug configuration that specifies the tags to use or ignore. For example to just run the coin() tests: -Dcucumber.filter.tags="@StepCoin"


Debugging Python within this mainly JVM-based project structure requires a bit of configuration. The following steps will help get Intellij prepared for this task:

  1. Install the Python plugin from JetBrains which should provide PyCharm like functionality.

  2. Right-click the "tinkerpop" top-level module in Intellij’s project explorer and "Open Module Settings".

  3. Select "Platform Settings | SDKs" and then click the "+" to "Add Python SDK…​"

  4. Choose a "Virtualenv Environment" and a "New environment". Set the "Location" to "<project-root>/gremlin-python/src/main/python". Select a "Base interpreter" that matches the version required by gremlin-python and click "OK".

  5. Select "Project Settings | Modules" and then select "gremlin-python" from the listing. Change the "Module SDK" to the newly added Python SDK.

  6. Open a terminal to gremlin-python/scr/main/python and do venv/bin/pip3 install -e . to pull in all of the gremlin-python dependencies.

At this stage, it should be possible to run unit tests in Python:

  1. Pull down the drop down in the toolbar for "Run/Debug Configurations" and select "Edit Configurations…​"

  2. Click the "+" to "Python Tests | pytest" and browse to a test to supply the "Script path".

  3. Go to "Python interpreter" and select "Use SDK of module" and choose "gremlin-python" in the drop-down and click OK.

  4. The test should not appear in the "Run/Debug Configurations" and can be executed.

When the Debug button is pressed, Intellij may display a notification that using the debugger requires some additional downloads - confirm those installations as required.

Many of the tests in gremlin-python require Gremlin Server. They are effectively integration tests. TinkerPop makes it easy to debug these tests by providing a Docker based test server which is rigged up with all the configurations required for the tests to execute. Start this server with docker/gremlin-server.sh -n where the -n will enable Neo4j for transaction based tests.

For Committers

business gremlin

The guidelines that follow generally apply to those with commit access to the main repository, but those seeking to contribute will also find helpful information here on the development style and process for the project.

Initial Setup

Once the Apache TinkerPop PMC has sent an invitation to a contributor to become a new committer and that contributor has accepted that invitation and provided their iCLA to Apache, then there are some administrative steps that the new committer can expect to go through to complete the process so that they have access to Apache resources like the Git repository. While the information for completing the process can be found in a multitude of places the following listing provides a summary of what to do next:

  • Look for a welcome email from root_at_apache.org. This will contain your Apache user name

    • e.g. "Welcome to the Apache Software Foundation (ASF)!"

  • Visit https://id.apache.org/reset/enter and enter your user name

  • Look for a password reset email from root_at_apache.org. This will have been sent after you confirmed your user name, above.

    • e.g. "Password reset request for [username] from Apache ID"

  • Visit the link provided in the password reset email, and choose a new password. ASF asks you to choose a strong one. You will see a "Password change successful" page when this is done.

  • You will now have SVN access

  • Try sending yourself an email at your new [username]@apache.org email address. It should forward to your primary address.

  • Check your account details at username

  • Link your accounts using gitbox (https://gitbox.apache.org/setup/)

    • Link your Apache account

    • Link your GitHub account

      • You will be asked to "Authorize Apache M.A.T.T."

    • Obtain MFA status

      • Visit id.apache.org and set your GitHub ID to be invited to the org

      • Wait for an email to arrive from "support_at_github.com", and click "Join @apache". Accept the invitation on github.com. Your new MFA status will not immediately be reflected on github, but you will get a confirmation email from noreply_at_github.com. Later, the status will read "MFA ENABLED"

        • e.g. "[GitHub] @asf-gitbox has invited you to join the @apache organization"

      • You can find yourself by searching in the Apache Committers listing

  • Read the Apache Committer Guide and Apache Committer FAQ

  • Read through the other sections of this document - the Developer Documentation - for more details on project procedures and other administrative items.

  • If you have trouble committing, email dev@tinkerpop.apache.org


TinkerPop has a user mailing list and a dev mailing list. As a committer, it is a good idea to join both.

TinkerPop also has a Slack channel for more real-time communication about the project among contributors, though it must be kept in mind that project discussion and decisions must occur on the dev mailing list mentioned above.

Occasionally, online meetings via video conference are held. These meetings are schedule via the dev mailing list about a week before they are to occur to find a day and time that is available for those interested in attending. On the day of the meeting, the meeting organizer will create a Google Hangout (or similar video conferencing link). At that point, all who are interested can attend. Meeting minutes should be taken and added to the Meetings section of this document using the pattern already established.

Media Content Submission

Tinkerpop hosts Twitch streams for various knowledge sharing sessions and demos, as well as a YouTube channel to upload video content. Everyone in the community is welcome to submit contents to stream or upload, abiding the following process:

  1. Send a DISCUSS thread to the dev mailing list that describes the topic, the platform it goes to, who will present it and a date to do it.

  2. If the proposal comes from someone who isn’t an official TinkerPop committer, then they must obtain a committer "sponsor", who is responsible for and endorses the content. (This can be asked on the DISCUSS thread).

  3. The proposal must achieve lazy consensus at minimum.

  4. After obtaining approval from the DISCUSS thread, start planning and promoting your stream/video!

The process above applies to two content models: a class of content and individual content. A class of content refers to a content series which would encompass several individual streams or recordings, whereas individual content would be a single one-off stream or recording (which presumably does not fit into an established class). Proposing either a class of content or individual content both require a DISCUSS thread as written above, but once a class is established, content makers are free to make content under the model defined by that class without having a special DISCUSS thread for each recording or stream. You may, of course, use the dev list to organize a particular stream or recording for a particular content class.

The following bullets represent current content classes:

  • A Contributorcast is like a podcast and is held by contributors to the TinkerPop project. They may or may not be official committers, but they should have submitted merged pull requests of significance or have some recognized standing in Stackoverflow, Discord or just generally the TinkerPop Community. At least one official committer must be on the stream and those committers are in charge of content presented and those other contributors they invite to join. There is no particular format for this content and it is meant to be informal in its style but informational and more often than not about TinkerPop itself.

  • TinkerPop Wide streams are about the wider TinkerPop Community. They focus on third-party tools, libraries, graph system implementations and applications that are powered by TinkerPop, but might also stretch into the general graph world. An official committer must host the stream.

All media needs to achieve a certain level of consistency and quality with existing content. Content creators and their guests should abide by the community guidelines for Twitch, YouTube and the ASF when producing media.

Release Notes

There is a two-pronged approach to maintaining the change log and preparing the release notes.

  1. For work that is documented in JIRA, run the release notes report to include all of the tickets targeted for a specific release. This report can be included in the release announcement.

  2. The manual change log (CHANGELOG.asciidoc) can be used to highlight large changes, describe themes (e.g. "We focused on performance improvements") or to give voice to undocumented changes.

Given the dependence on the JIRA report for generating additions to the CHANGELOG.asciidoc, which uses the title of the issue as the line presented in the release note report, titles should be edited prior to release to be useful in that context. In other words, an issue title should be understandable as a change in the fewest words possible while still conveying the gist of the change.

Changes that break the public APIs should be marked with a "breaking" label and should be distinguished from other changes in the release notes.


TinkerPop has several release branches:

  • 3.0-dev - 3.0.x (no longer maintained)

  • 3.1-dev - 3.1.x (no longer maintained)

  • 3.2-dev - 3.2.x (no longer maintained)

  • 3.3-dev - 3.3.x (no longer maintained)

  • 3.4-dev - 3.4.x (no longer maintained)

  • 3.5-dev - 3.5.x (non-breaking bug fixes and enhancements)

  • 3.6-dev - 3.6.x (non-breaking bug fixes and enhancements)

  • master - 3.7.x (current development)

  • 4.0-dev - 4.0.x (future development)

The branch description above that reads "non-breaking bug fixes and enhancements" simply means that within that release line (i.e. patch version) changes should not alter existing behavior, introduce new APIs, change serialization formats, modify protocols, etc. In this way, users and providers have an easy way to know that within a minor release line, they can be assured that their upgrades will not introduce potential problems. A good rule of thumb is to consider whether a client of one version within a release line can interact properly with a server version within that same line. If so, it is likely an acceptable change within that branch.

Changes to earlier branches should merge forward toward master (e.g. 3.6-dev should merge to master). Please read more about this process in the Pull Requests section. Note that 4.0-dev is rebased on master and currently behaves as a fresh repository as all 3.x content was removed.

As described in versioning, it is possible to do a "partial" release which will utilize a four-digit version that starts with a "1" (e.g. The branching strategy for a partial release requires that a -dev branch be created with the three digit prefix.

Other branches may be created for collaborating on features or for RFC’s that other developers may want to inspect. It is suggested that the JIRA issue ID be used as the prefix, since that triggers certain automation, and it provides a way to account for the branch lifecycle, i.e. "Who’s branch is this, and can I delete it?"

For branches that are NOT associated with JIRA issues, developers should utilize their Apache ID as a branch name prefix. This provides a unique namespace, and also a way to account for the branch lifecycle.

Developers should remove their own branches when they are no longer needed.


Tags are used for milestones, release candidates, and approved partial and complete releases. Tags for a complete release are simply defined by the three-digit version number. Tags for a partial release, should be prefixed by the individual language relevant to that release. For example, if there is an initial partial release for 3.6.0 on gremlin-python then the tag should be

Please refrain from creating arbitrary tags, as they produce permanent clutter.


Each programming language has a runtime that TinkerPop supports. In general, TinkerPop will attempt to support the current LTS version for a particular major version for the lifetime of its minor and patch releases. A deprecation notice will be applied to the last release of TinkerPop that will support an outdated runtime and the runtime will be updated in the following release. Part of each major release cycle, should include some analysis of the current runtime supported.

  • Java - Typically, TinkerPop will find itself bound to the version held by its major dependencies like Apache Spark.

  • Javascript - Consult nodejs/Release for the current active LTS of node and here for npm compatibility.

  • .NET - Consult .NET Release Lifecycle for LTS status.

  • Python - Consult Python.org for the current LTS status.

Issue Tracker Conventions

TinkerPop uses Apache JIRA as its issue tracker. JIRA is a very robust piece of software with many options and configurations. To simplify usage and ensure consistency across issues, the following conventions should be adhered to:

  • An issue’s "status" should generally be in one of two states: open or closed (reopened is equivalent to open for our purposes).

    • An open issue is newly created, under consideration or otherwise in progress.

    • A closed issue is completed for purposes of release (i.e. code, testing, and documentation complete).

    • Issues in a resolved state should immediately be evaluated for movement to closed - issue become resolved by those who don’t have the permissions to close.

  • An issue’s "type" should be one of two options: bug or improvement.

    • A bug has a very specific meaning, referring to an error that prevents usage of TinkerPop AND does not have a reasonable workaround. Given that definition, a bug should generally have very high priority for a fix.

    • Everything else is an improvement in the sense that any other work is an enhancement to the current codebase.

  • The "component" should be representative of the primary area of code that it applies to and all issues should have this property set.

  • Issues are not assigned "labels" with two exceptions:

    • The "breaking" label which marks an issue as one that is representative of a change in the API that might affect users or providers. This label is important when organizing release notes.

    • The "deprecation" label which is assigned to an issue that includes changes to deprecate a portion of the API.

  • The "affects/fix version(s)" fields should be appropriately set, where the "fix version" implies the version on which that particular issue will completed. This is a field usually only set by committers and should only be set when the issue is being closed with a completed disposition (e.g. "Done", "Fixed", etc.).

  • The "priority" field can be arbitrarily applied with one exception. The "trivial" option should be reserved for tasks that are "easy" for a potential new contributor to jump into and do not have significant impact to urgently required improvements.

  • The "resolution" field which is set on the close of the issue should specify the status most closely related to why the issue was closed. In most cases, this will mean "Fixed" for a "Bug" or "Done" for an "Improvement". Only one resolution has special meaning and care should be taken with this particular option: "Later". "Later" means that the item is a good idea but likely will not be implemented in any foreseeable future. By closing uncompleted issues with this resolution, it should be easy to come back to them later when needed.

Code Style

Contributors should examine the current code base to determine what the code style patterns are and should match their style to what is already present. Of specific note however, TinkerPop does not use "import wildcards" - IDEs should be adjusted accordingly to not auto-wildcard the imports.

Build Server

TinkerPop uses GitHub Actions for CI services. The build status can be found here. There is a single "build-test" workflow that runs a number of jobs that break the test execution into a series of smaller test executions. Taken together, they provide a solid cross section of coverage of the code base


When possible, committers should avoid direct "breaking" change (e.g. removing a method from a class) and favor deprecation. Deprecation should come with sufficient documentation and notice especially when the change involves public APIs that might be utilized by users or implemented by providers:

  • Mark the code with the @Deprecated annotation.

  • Use javadoc to further document the change with the following content:

    • @deprecated As of release 3.7.2, replaced by {@link SomeOtherClass#someNewMethod()} - if the method is not replaced then the comment can simply read "not replaced". Additional comments that provide more context are encouraged.

    • @see <a href="https://issues.apache.org/jira/browse/TINKERPOP-XXX">TINKERPOP-XXX</a> - supply a link to the JIRA issue for reference - the issue should include the "deprecation" label.

  • Be sure that deprecated methods are still under test - consider using javadoc/comments in the tests themselves to call out this fact.

  • Create a new JIRA issue to track removal of the deprecation for future evaluation.

  • Update the "upgrade documentation" to reflect the API change and how the reader should resolve it.

The JIRA issues that track removal of deprecated methods should be periodically evaluated to determine if it is prudent to schedule them into a release.

Developing Tests

TinkerPop has a wide variety of test types that help validate its internal code as well as external provider code. There are "unit tests" and "integration tests". Unit tests execute on standard runs of mvn clean install. These tests tend to run quickly and provide a reasonable level of coverage and confidence in the code base. Integration tests are disabled by default and must be explicitly turned on with a special build property by adding -DskipIntegrationTests=false to the mvn execution. Integration tests run slower and may require external components to be running when they are executed. They are "marked" as separate from unit tests by inclusion of the suffix "IntegrateTest".

Here are some other points to consider when developing tests:

  • Avoid use of println in tests and prefer use of a SLF4j Logger instance so that outputs can be controlled in a standard way.

  • If it is necessary to create files on the filesystem, do not hardcode directories - instead, use the TestHelper to create directory structures. TestHelper will properly create file system structure in the appropriate build directory thus allowing proper clean-up between test runs.

  • If writing tests in one of the test suites, like gremlin-test, it is important to remember that if a new Graph instance is constructed within the test manually, that it be closed on exit of that test. Failing to do this cleanup can cause problems for some graph providers.

  • Tests that are designed to use a GraphProvider implementation in conjunction with AbstractGremlinTest and are in the /test directory should not be named with Test as the suffix, as this will cause them to execute in some environments without a GraphProvider being initialized by a suite. These types of tests should be suffixed with Check instead. Please see NativeNeo4jStructureCheck for an example.

Gremlin Language Test Cases

Test cases for the Gremlin Language currently requires that the newly developed test be added in two places:

  1. As a test written in Java in the gremlin-test module within the subpackages of org.apache.tinkerpop.gremlin.process.traversal.step

  2. As a test written in Gherkin in the gremlin-test module in the /features subdirectory

When writing a Java test case for a Gremlin step, be sure to use the following conventions.

  • The name of the traversal generator should start with get, use X for brackets, _ for space, and the Gremlin-Groovy sugar syntax.

    • get_g_V_hasLabelXpersonX_groupXaX_byXageX_byXsumX_name()

  • When creating a test for a step that has both a barrier and sideEffect form (e.g. group(), groupCount(), etc.), test both representations.

    • get_g_V_groupCount_byXnameX()

    • get_g_V_groupCountXaX_byXnameX_capXaX()

  • The name of the actual test case should be the name of the traversal generator minus the get_ prefix.

  • The Gremlin-Groovy version of the test should use the sugar syntax in order to test sugar (as Gremlin-Java tests test standard syntax).

    • g.V.age.sum

  • Avoid using lambdas in the test case unless that is explicitly what is being tested as OLAP systems will typically not be able to execute those tests.

  • AbstractGremlinProcessTest has various static methods to make writing a test case easy.

    • checkResults(Arrays.asList("marko","josh"), traversal)

    • checkMap(new HashMap<String,Long>() {{ put("marko",1l); }}, traversal.next())

Gherkin tests follow some important conventions and have a sub-language that must be adhered to for the tests to function properly. Note that Gherkin tests are designed to support the testing of GLVs and at some point will likely replace the Java tests. If a new Java test is added and an associated Gherkin tests is not, the overall build will fail the FeatureCoverageTest of gremlin-test which validates that all tests written in Java are also implemented in Gherkin.

The basic syntax of a Gherkin test is as follows:

Scenario: g_VX1X_unionXrepeatXoutX_timesX2X__outX_name
  Given the modern graph
  And using the parameter vId1 defined as "v[marko].id"
  And the traversal of
    g.V(vId1).union(__.repeat(__.out()).times(2), __.out()).values("name")
  When iterated to list
  Then the result should be unordered
    | result |
    | ripple |
    | lop |
    | lop   |
    | vadas |
    | josh  |

Scenario Name

The name of the scenario needs to match the name of the Java test. If it does not then the FeatureCoverageTest will fail.


"Given" sets the context of the test. Specifically, it establishes the graph that will be used for the test. It conforms to the pattern of "Given the xxx graph" where the "xxx" may be one of the following:

  • empty

  • modern

  • classic

  • crew

  • sink

  • grateful

Never modify the data of any of the graphs except for the "empty" graph. The "empty" graph is the only graph that is guaranteed to be refreshed between tests. The "empty" graph maybe be modified by the traversal under test or by an additional "Given" option:

Given the empty graph
And the graph initializer of
  g.addV("person").property(T.id, 1).property("name", "marko").property("age", 29).as("marko").
    addV("person").property(T.id, 2).property("name", "vadas").property("age", 27).as("vadas").
    addV("software").property(T.id, 3).property("name", "lop").property("lang", "java").as("lop").
    addV("person").property(T.id, 4).property("name","josh").property("age", 32).as("josh").
    addV("software").property(T.id, 5).property("name", "ripple").property("lang", "java").as("ripple").
    addV("person").property(T.id, 6).property("name", "peter").property("age", 35).as('peter').
    addE("knows").from("marko").to("vadas").property(T.id, 7).property("weight", 0.5).
    addE("knows").from("marko").to("josh").property(T.id, 8).property("weight", 1.0).
    addE("created").from("marko").to("lop").property(T.id, 9).property("weight", 0.4).
    addE("created").from("josh").to("ripple").property(T.id, 10).property("weight", 1.0).
    addE("created").from("josh").to("lop").property(T.id, 11).property("weight", 0.4).
    addE("created").from("peter").to("lop").property(T.id, 12).property("weight", 0.2)

The above configuration will use the "empty" graph and initialize it with the specified traversal. In this case, that traversal loads the "empty" graph with the "modern" graph.

Once the graph for the test is defined, the context can be expanded to include parameters that will be applied to the traversal under test. Any variable value being used in the traversal under test, especially ones that require a specific type, should be defined as parameters. The structure for parameter definition looks like this:

Given the modern graph
And using the parameter vId1 defined as "v[marko].id"

In the above example, "vId1" is the name of the parameter that will be used in the traversal. The end of that line in quotes is the value of that parameter and should use the type system notation that has been developed for the TinkerPop Gherkin tests. The type system notation ensures that different language variants have the ability to construct the appropriate types expected by the tests.

The syntax of the type notation involves a prefix character to help denote the type, a value between two square brackets, optionally suffixed with some additional notation depending on the primary type.

  • Date - dt[xxx] - The "xxx" should be ISO 8601 string.

  • Edge - e[xxx] - The "xxx" should be replaced with a representation of an edge in the form of the vertex_name-edgelabel→vertex_name. This syntax may also include the .id suffix which would indicate getting the edge identifier or the .sid suffix which gets a string representation of the edge identifier.

  • Lambda - c[xxx] - The "xxx" should contain a lambda written in Groovy.

  • List - l[xxx,yyy,zzz,…​] - A comma separated collection of values that make up the list should be added to between the square brackets. These values respect the type system thus allowing for creation of lists of vertices, edges, maps, and any other available type. Spaces are taken literally, therefore it is important to avoid spaces unless they are required for the test.

  • Map - m[xxx] - The "xxx" should be replaced with a JSON string. Note that keys and values will be parsed using the type notation system so that it is possible to have maps containing arbitrary keys and values.

  • Numeric - d[xxx].y - The "xxx" should be replaced with a number. The suffix denoted by "y" should always be included to further qualify the type of numeric. The following options are available:

    • b - 8-bit byte

    • s - 16-bit Short

    • i - 32-bit Integer

    • f - 32-bit Float

    • l - 64-bit Long

    • d - 64-bit Double

    • m - Arbitrary-precision signed decimal numbers (e.g. BigDecimal in Java)

    • n - Arbitrary-precision integers (e.g. BigInteger in Java)

  • Path - p[xxx,yyy,zzz,…​] - A comma separated collection of values that make up the Path should be added to between the square brackets. These values respect the type system thus allowing for creation of Path of vertices, edges, maps, and any other available type.

  • Set - s[xxx,yyy,zzz,…​] - A comma separated collection of values that make up the set should be added to between the square brackets. These values respect the type system thus allowing for creation of sets of vertices, edges, maps, and any other available type.

  • String - Any value not using the system notation will be interpreted as a string by default.

    • str[xxx] (Optional) - xxx should be replaced with a string. Optional notation used for specific string results, such as null and spaces.

  • T - t[xxx] - The "xxx" should be replaced with a value of the T enum, such as id or label.

  • Vertex - v[xxx] - The "xxx" should be replaced with the "name" property of a vertex in the graph. This syntax may include the .id suffix which would indicate getting the vertex identifier or the .sid suffix which gets a string representation of the edge identifier.

In addition, parameter names should adhere to a common form as they hold some meaning to certain language variant implementations:

  • General variables of no particular type should use xx1, xx2 and xx3.

  • A Vertex variable should be prefixed with "v" and be followed by the id, therefore, v1 would signify a Vertex with the id of "1".

  • An Edge variable follows the pattern of vertices but with a "e" prefix.

  • The "id" of a Vertex or Edge is prefixed with "vid"`" or "eid" respectively followed by the id, thus, vid1 would be "1" and refer to the Vertex with that id.

  • Function variables should use l1 and l2.

  • Predicate variables should use pred1.

  • Comparator variables should use c1 and c2.

Finally, specify the traversal under test with the "Given" option "and the traversal":

And the traversal of
  g.V(vId1).union(__.repeat(__.out()).times(2), __.out()).values("name")

The traversal must be written so that it can be parsed by both gremlin-groovy and gremlin-language. Using syntax particular to one but not the other may result in test execution errors.

It will be the results of this traversal that end up being asserted by Gherkin. When writing these test traversals, be sure to always use the method and enum prefixes. For example, use __.out() for an anonymous traversal rather than just out() and prefer Scope.local rather than just local.

If a particular test cannot be written in Gherkin for some reason or cannot be otherwise supported by a GLV, first, consider whether or not this test can be re-written in Java so that it will work for GLVs and then, second, if it cannot, then use the following syntax for unsupported tests:

Scenario: g_V_outXcreatedX_groupCountXxX_capXxX
  Given an unsupported test
  Then nothing should happen because
    The result returned is not supported under GraphSON 2.x and therefore cannot be properly asserted. More
    specifically it has vertex keys which basically get toString()'d under GraphSON 2.x. This test can be supported
    with GraphSON 3.x.


The "When" options get the result from the traversal in preparation for assertion. There are two options to iterate:

  • "When iterated to list" - iterates the entire traversal into a list result that is asserted

  • "When iterated next" - gets the first value from the traversal as the result to be asserted

There should be only one "When" defined in a scenario.


The "Then" options handle the assertion of the result. There are several options to consider:

  • "the result should have a count of xxx" - assumes a list value in the result and counts the number of values in it

  • "the result should be empty" - no results

  • "the traversal will raise an error" - an exception is thrown as a result of traversal iteration

  • "the traversal will raise an error with message comparison text of message" - an exception is thrown as a result of traversal iteration where "comparison" may be one of "containing", "starting", or "ending".

  • "the result should be ordered" - the exact results and should appear in the order presented

  • "the result should be unordered" - the exact results but can appear any order

  • "the result should be of" - results can be any of the specified values and in any order (use when guarantees regarding the exact results cannot be pre-determined easily - see the range()-step tests for examples)

These final three types of assertions mentioned above should be followed by a Gherkin table that has one column, where each row value in that column represents a value to assert in the result. These values are type notation respected as shown in the following example:

Then the result should be unordered
  | result |
  | ripple |
  | lop |
  | lop   |
  | vadas |
  | josh  |

Another method of assertion is to test mutations in the original graph. Again, mutations should only occur on the "empty" graph, but they can be validated as follows:

Scenario: g_V_outE_drop
  Given the empty graph
  And the graph initializer of
  And the traversal of
  When iterated to list
  Then the result should be empty
  And the graph should return 2 for count of "g.V()"
  And the graph should return 0 for count of "g.E()"


Features have tags associated with them to help allow developers to better break up test runs if they desire. There are two types of tags:

  • @StepClass* - Marks the step grouping and is a prefix that precedes and either refers to one of the following:

    • One of the four types of steps: Branch, Filter, Map, and SideEffect (e.g. @StepClassBranch)

    • Semantics which maps to elements of the Gremlin Semantics specification.

    • An Integrated grouping that does not fit those individual classifications well.

  • @Step* - Marks testing for a particular step. While this tag is generally unique to the feature file itself and test filtering could be accomplished at that level by way of the file, the use of the tag is a secondary option and allows filtering to be consistently managed by tags alone. The suffix is the Gremlin step itself (e.g. @StepHas) in all cases except for @StepVertex which covers V(), E(), out(), in(), both(), inE(), outE(), and bothE().

Scenarios have tags associated with them that help identify subsets of tests so that a test runner can selectively include or ignore certain tests. The tags enable the practical and necessary ability for providers to ignore tests that they simply cannot support. It is important to be aware of the following tags when writing tests as not including a tag when one is necessary will cause provider tests to fail:

  • @AllowNullPropertyValues - The scenario requires that the graph be configured with AllowNullPropertyValues as true (meaning that it can store null values).

  • @GraphComputerVerificationInjectionNotSupported - The scenario will not work on with GraphComputer because the inject() step is not supported.

  • @GraphComputerVerificationMidVNotSupported - The scenario will not work on with GraphComputer because the mid-V() step is not supported.

  • @GraphComputerVerificationOneBulk - The scenario will not work because withBulk(false) is configured and that is not compatible with GraphComputer

  • @GraphComputerVerificationReferenceOnly - The scenario itself is not written to support GraphComputer because it tries to reference inaccessible properties that are on elements only available by "reference" (i.e T.id only).

  • @GraphComputerVerificationStrategyNotSupported - The scenario uses a traversal strategy that is not supported by GraphComputer.

  • @GraphComputerVerificationStarGraphExceeded - The scenario itself is not written to support GraphComputer because the traversal does not mind the star graph limitation.

  • @InsertionOrderingRequired - The scenario is reliant on the graph system predictably returning results (vertices, edges, properties) in the same order in which they were inserted into the graph.

  • @MetaProperties - The scenario makes use of meta-properties.

  • @MultiProperties - The scenario makes use of multi-properties.

  • @RemoteOnly - The scenario uses some Gremlin syntax that cannot be supported outside of remote test executions. The best example of this sort of test would be one that uses the remote Lambda syntax.

  • @UserSuppliedVertexIds - The scenario relies on the vertex IDs specified in the dataset used by the scenario.

  • @UserSuppliedEdgeIds - The scenario relies on the edge IDs specified in the dataset used by the scenario.

  • @UserSuppliedVertexPropertyIds - The scenario relies on the vertex property IDs specified in the dataset used by the scenario.

  • @With* - The scenario uses some with() based configuration like strategies:

    • @WithPartitionStrategy

    • @WithProductiveByStrategy

    • @WithReadOnlyStrategy

    • @WithSeedStrategy

    • @WithSubgraphStrategy

Tag filters can be applied to Intellij at execution time by adding a system properties of -Dcucumber.filter.tags=<step-filter>.

Gremlin Socket Server Tests

gremlin-socket-server is an included test server for driver tests which require predefined server behavior. Typically, this is to test scenarios such as the server closing the connection or returning an error code but other response behavior can be added as needed for tests.

Gremlin socket server uses the request id of incoming messages to determine how to respond. Request ids are defined in gremlin-tools/gremlin-socket-server/conf/*.yaml. The server side behavior for each request id is implemented in TestWSGremlinInitializer.

To add new server side behavior, define a new request id in the config yaml, then add a corresponding field in SocketServerSettings. In TestWSGremlinInitializer, add a new block to the if else chain in the decode method which matches the request id. Define all server behavior and responses in this block.

To write the driver test, send a request message to the gremlin socket server port as if it were a normal gremlin server. Override the request id on the request with one defined in the config yaml and gremlin socket server will respond according to the defined behavior.

Ensure that the socket server is running during driver tests. By default, a docker image for gremlin-socket-server is built during mvn install. The simplest way to use the socket server during driver tests is to run a gremlin socket server container during the integration phase of driver tests.

Developing Benchmarks

Benchmarks are a useful tool to track performance between TinkerPop versions and also as tools to aid development decision making. TinkerPop uses OpenJDK JMH for benchmark development. The JMH framework provides tools for writing robust benchmarking code that avoid many of the pitfalls inherent in benchmarking JIT compiled code on the JVM. Example JMH benchmarks can be found here.

TinkerPop benchmarks live in the gremlin-benchmark module and can either be run from within your IDE or as a standalone uber-jar. The uber-jar is the JMH recommended approach and also makes it easy to distribute artifacts to various environments to gather benchmarking numbers. Having said that, in most cases it should be sufficient to run it from within the IDE.

Benchmarks will not run by default because they are time consuming. To enable benchmarks during the test phase do -DskipBenchmarks=false. To change the number of warmup iterations, measurement iterations, and forks you can do mvn clean test -DskipBenchmarks=false -DdefaultForks=5 -DmeasureIterations=20 -DwarmupIterations=20. Benchmark results will be output by default to the benchmarks directory in JSON format.

Benchmarks may also be run from the command line using the JMH runner. Build the uber-jar and simply run java -jar gremlin-benchmark-TP-VERSION.jar. To see a list of JMH runner options, add the -h flag.

The JUnit/JMH integration was inspired by the Netty projects microbenchmarking suite. Please refer to the Netty docs for more details. Presently there are 3 abstract benchmark classes that may be used as building blocks for your benchmarks; AbstractBenchmarkBase, AbstractGraphBenchmark, and AbstractGraphMutateBenchmark.

  • AbstractBenchmarkBase - extend when your benchmark does not require a graph instance

  • AbstractGraphBenchmark - extend when you are benchmarking read operations against a graph

  • AbstractGraphMutateBenchmark - extend when you are benchmarking graph mutation operations eg. g.addV(), graph.addVertex()

Review then Commit

Code modifications must go through a review-then-commit (RTC) process before being merged into a release branch. All committers should follow the pattern below, where "you" refers to the committer wanting to put code into a release branch.

  • Make a JIRA ticket for the software problem you want to solve (i.e. a fix).

  • Fork the release branch that the fix will be put into.

    • The branch name should be the JIRA issue identifier (e.g. TINKERPOP-XXX).

  • Develop your fix in your branch.

  • When your fix is complete and ready to merge, issue a pull request.

    • Be certain that the test suite is passing.

    • If you updated documentation, be sure that the process-docs.sh is building the documentation correctly.

  • Before you can merge your branch into the release branch, you must have at least 3 +1 consensus votes from other committers OR a single +1 from a committer and a seven day review period for objections (i.e. a "cool down period") at which point we will assume a lazy consensus.

    • Please see the Apache Software Foundations regulations regarding Voting on Code Modifications.

    • With the "cool down" process and lazy consensus the single +1 may (should) come from the committer who submitted the pull request (in other words, the change submitter and the reviewer are the same person).

    • Committers are trusted with their changes, but are expected to request reviews for complex changes as necessary and not rely strictly on lazy consensus.

  • Votes are issued by TinkerPop committers as comments to the pull request.

  • Once either consensus position is reached, you are responsible for merging to the release branch and handling any merge conflicts.

    • If there is a higher version release branch that requires your fix (e.g. 3.y-1.z fix going to a 3.y.z release), multiple pull requests may be necessary (i.e. one for each branch).

  • Be conscious of deleting your branch if it is no longer going to be used so stale branches don’t pollute the repository.

These steps also generally apply to external pull requests from those who are not official Apache committers. In this case, the person responsible for the merge after voting is typically the first person available who is knowledgeable in the area that the pull request affects. Any additional coordination on merging can be handled via the pull request comment system.

For those performing reviews as part of this process it is worth noting that the notion of "review" is fairly wide for our purposes. TinkerPop has grown into a large and complex code base and very few people (if anyone) is knowledgeable on all of its modules. Detailed code reviews might often be difficult or impossible as a result.

To be clear, a "review" need not be specifically about the exact nature of the code. It is perfectly reasonable to review (and VOTE) in the following fashion:

  • VOTE +1 - ran docker integration tests and everything passes

  • VOTE +1 - reviewed the code in detail - solid pull request

  • VOTE +1 - agree with the principle of this pull request but don’t fully understand the code

  • VOTE +1 - read through the updated documentation and understand why this is important, nice

Non-committers are welcome to review and VOTE as well and while their VOTEs are not binding, they will be taken as seriously as non-binding VOTEs on releases. Reviewing and VOTEing on pull requests as a non-committer is a great way to contribute to the TinkerPop community and get a good pulse on the changes that are upcoming to the framework.

The following exceptions to the RTC (review-then-commit) model presented above are itemized below. It is up to the committer to self-regulate as the itemization below is not complete and only hints at the types of commits that do not require a review.

  • You are responsible for a release and need to manipulate files accordingly for the release.

    • Gremlin.version(), CHANGELOG dates, pom.xml version bumps, etc.

  • You are doing an minor change and it is obvious that an RTC is not required (would be a pointless burden to the community).

    • The fix is under the commit-then-review (CTR) policy and lazy consensus is sufficient, where a single -1 vote requires you to revert your changes.

    • Adding a test case, fixing spelling/grammar mistakes in the documentation, fixing LICENSE/NOTICE/etc. files, fixing a minor issue in an already merged branch.

When the committer chooses CTR, it is considered good form to include something in the commit message that explains that CTR was invoked and the reason for doing so. For example, "Invoking CTR as this change encompasses minor adjustments to text formatting." CTR based commits will still require manual merging through all release branches. Merges should occur in reverse order, starting with the latest release version first (e.g. if the fix is going to 3.3.x then the change should be merged in the following order master, 3.4-dev, 3.3-dev).

Pull Requests

When submitting a pull request to one of the release branches, be sure it uses the following style:

  • The title of the pull request is the JIRA ticket number + "colon" + the title of the JIRA ticket.

  • The first line of the pull request message should contain a link to the JIRA ticket.

  • Discuss what you did to solve the problem articulated in the JIRA ticket.

  • Discuss any "extra" work done that go beyond the assumed requirements of the JIRA ticket.

  • Be sure to explain what you did to prove that the issue is resolved.

    • Test cases written.

    • Integration tests run (if required for the work accomplished).

    • Documentation building (if required for the work accomplished).

    • Any manual testing (though this should be embodied in a test case).

  • Notes about what you will do when you merge to the respective release branch (e.g. update CHANGELOG).

    • These types of "on merge tweaks" are typically done to extremely dynamic files to combat and merge conflicts.

  • If you are a TinkerPop committer, you can VOTE on your own pull request, so please do so.

A pull request will typically be made to a target branch. Assuming that branch is upstream of other release branches (e.g. a pull request made to for the branch containing 3.3.x must merge to the branch that releases 3.4.x), it is important to be sure that those changes are merged to the downstream release branches. If the merge from one release branch to another is not terribly conflicted, it is likely safe to offer a single pull request and then merge through the release branches after review. If there is conflict or the likelihood of test failures in downstream branches then this process is best handled by multiple pull requests: one to each release branch. Release branches with merged changes should be pushed in reverse order, starting with the latest release version first (e.g. if the fix is going to 3.3.x then the change should be merged in the following order: master, 3.4-dev`, 3.3-dev).

As an example, consider a situation where there is a feature branch named "TINKERPOP-1234" that contains a fix for the 3.4-dev branch:

`git checkout -b TINKERPOP-1234 3.4-dev`
// do a bunch of stuff to implement TINKERPOP-1234 and commit/push
git checkout -b <TINKERPOP-1234-master> master
git merge TINKERPOP-1234

At this point, there are two branches, with the same set of commits going to 3.4-dev and master. Voting will occur on both pull requests. After a successful vote, it is time to merge. If there are no conflicts, then simply git merge both pull requests to their respective branches. If there are conflicts, then there is some added work to do - time to rebase:

git checkout TINKERPOP-1234
git rebase origin/3.4-dev

Depending on the conflict, it might be a good idea to re-test before going any further, otherwise:

git push origin TINKERPOP-1234 --force

Now, git rebase has re-written the commit history, which makes a mess of the other pull request to master. This problem is rectified by essentially re-issuing the PR:

git checkout TINKERPOP-1234-master
git reset --hard origin/master
git merge TINKERPOP-1234

Again, depending on the changes, it may make sense to re-test at this point, otherwise:

git push origin TINKERPOP-1234-master --force

It should now be safe to merge both pull requests to their release branches.

Always take a moment to review the commits in a particular pull request. Be sure that they are all related to the work that was done and that no extraneous commits are present that cannot be explained. Ensuring a pull request only contains the expected commits is the responsibility of the committer as well as the reviewer.


There are many dependencies on other open source libraries in TinkerPop modules. When adding dependencies or altering the version of a dependency, developers must consider the implications that may apply to the TinkerPop LICENSE and NOTICE files. There are two implications to consider:

  1. Does the dependency fit an Apache approved license?

  2. Given the addition or modification to a dependency, does it mean any change for TinkerPop LICENSE and NOTICE files?

Understanding these implications is important for insuring that TinkerPop stays compliant with the Apache 2 license that it releases under.

Regarding the first item, refer to the Apache Legal for a list of approved licenses that are compatible with the Apache 2 license.

The second item requires a bit more effort to follow. The Apache website offers a how-to guide on the approach to maintaining appropriate LICENSE and NOTICE files, but this guide is designed to offer some more specific guidance as it pertains to TinkerPop and its distribution.

To get started, TinkerPop has both "source" and "binary" LICENSE/NOTICE files:

  • Source LICENSE/NOTICE relate to files packaged with the released source code distribution: LICENSE / NOTICE

  • Binary LICENSE/NOTICE relate to files packaged with the released binary distributions:


As dependencies are not typically added to the source distribution (i.e. the source zip distribution), there is typically no need to edit source LICENSE/NOTICE when editing a TinkerPop pom.xml. These files only need to be edited if the distribution has a file added to it. Such a situation may arise from several scenarios, but it would most likely come from the addition of a source file from another library.

  • If the file being bundled is Apache licensed, then add an entry to NOTICE.

  • If the file being bundled is under a different approved license, then add an entry to LICENSE and include a copy of that LICENSE in the root /licenses directory of the code repository.


The binary LICENSE/NOTICE is perhaps most impacted by changes to the various pom.xml files. After altering the pom.xml file of any module, build gremlin-driver, Gremlin Console and Gremlin Server and examine the contents of the binary distributions:

  • target/gremlin-driver-3.7.2-uber.jar

  • target/gremlin-console-3.7.2-uber.jar

  • target/apache-tinkerpop-gremlin-console-3.7.2-distribution.zip

  • target/apache-tinkerpop-gremlin-server-3.7.2-distribution.zip

Apache licensed software does not need to be included in LICENSE, but if the new dependency is an Apache-approved license (e.g. BSD, MIT) then it should be added in the pattern already defined. A copy of the LICENSE should be added to the <project>/src/main/static/licenses directory of the code repository and the maven-shade-plugin section of the gremlin-console and gremlin-driver pom.xml files should be updated to reference this new license file so that it is included in the uber jar.

To determine if changes are required to the NOTICE, first check if the bundled jar has a NOTICE file in it (typically found in /META-INF directory of the jar).

  • If the bundled file does not have a NOTICE, then no changes to TinkerPop’s NOTICE are required.

  • If the NOTICE of the file being bundled is NOT Apache licensed, then there is no change to TinkerPop’s NOTICE.

  • If the NOTICE of the file being bundled is Apache licensed, then include the copyright notification in TinkerPop’s NOTICE.

  • If the NOTICE of the file being bundled is Apache licensed AND is an Apache Software Foundation project, then ONLY include the portion of that NOTICE in TinkerPop’s NOTICE that is unrelated to the Apache boilerplate NOTICE. If there is no such portion that is different than the boilerplate then this NOTICE can be excluded (i.e. don’t alter TinkerPop’s NOTICE at all).

Please refer to the Modifications to Notice section of the Apache "Licensing How-to" for more information.


The documentation for TinkerPop is stored in the git repository in docs/src/ and are then split into several subdirectories, each representing a "book" (or its own publishable body of work). If a new AsciiDoc file is added to a book, then it should also be included in the index.asciidoc file for that book, otherwise the preprocessor will ignore it. Likewise, if a whole new book (subdirectory) is added, it must include an index.asciidoc file to be recognized by the AsciiDoc preprocessor.

Adding a book also requires a change to the root pom.xml file. Find the "asciidoc" Maven profile and add a new <execution> to the asciidoctor-maven-plugin configuration. For each book in docs/src/, there should be a related <execution> that generates the HTML from the AsciiDoc. Follows the patterns already established by the existing <execution> entries, paying special attention to the pathing of the '<sourceDirectory>', <outputDirectory> and <imagesdir>. Note that the <outputDirectory> represents where the book will exist when uploaded to the server and should preserve the directory structure in git as referenced in <sourceDirectory>.

Adding Gremlin code examples to any of the docs/src/recipes or to docs/src/reference/the-traversal.asciidoc also has the effect of improving testing of the Gremlin language. All Gremlin found in code sections that are marked as [gremlin-groovy] are tested in two ways:

  1. When mvn clean install is executed all such Gremlin are passed through the grammar parser to ensure validity. As the grammar parser is not a Groovy parser, the test framework attempts to filter away or ignore things it can’t possibly parse. Ideally, examples should be written in such a way as to be parsed by the grammar, but in cases where it cannot be as such, the test suite simply needs to be modified to suitably ignore the example.

  2. When the documentation is built, the code snippets are actually executed and errors will result in a failure to build the documentation.

Please see the Building and Testing section for more information on how to generate the documentation.

Asciidoc Formatting Tips

Use Asciidoctor

Asciidoc may render differently with different tools. What may look proper and correct with an IDE may be different than what is ultimately generated during the official build of the documentation which uses Asciidoctor. As a result it’s best to not rely on any other view of changes besides one generated by Asciidoctor.

Anonymous Traversal Formatting

The double underscore (i.e. __) does not typically render right and requires such code to be wrapped with `pass:[__]` or `+__+`.

Cause: #1717, #1066

Non-whitespace After Backtick

Use double backtick if there is non-whitespace immediately following the trailing backtick. So rather than: `ScriptInputFormat`'s, prefer ``ScriptInputFormat``'s.

Original: […​] globally available for ScriptInputFormat’s `parse() method

Fixed: […​] globally available for ScriptInputFormat's parse() method

Cause: #1514


The content for the TinkerPop home page and related pages that make up the web site at tinkerpop.apache.org is stored in the git repository under /docs/site. In this way, it becomes easier for the community to provide content presented there, because the content can be accepted via the standard workflow of a pull request. To generate the site for local viewing, run bin/generate-home.sh, which will build the site in target/site/. Note that Node.js and npm have to be installed in order for the script to work. See the JavaScript Environment section for more info about what parts of TinkerPop depend on Node.js and npm. While most of the generated website files can be viewed locally by opening them in a browser, some of them rely on imported resources that will be blocked by the browser’s same-origin policy if not served from a single origin using a web server. The generated website can be served locally by running npx serve target/site/home. PMC members can officially publish the site with bin/publish-home.sh <username>.

"Publishing" does not publish documentation (e.g. reference docs, javadocs, etc) and only publishes what is generated from the content in /docs/site. Publishing the site can be performed out of band with the release cycle and is no way tied to a version. The master branch should always be considered the "current" web site and publishing should only happen from that branch.


TinkerPop uses SLF4j for logging and relies on logback as the implementation. Configuring log outputs for debugging purposes within tests can be altered by editing the logback-test.xml file in each module’s test resources. That file gets copied to the target/test-classes on build and surefire and failsafe plugins in maven are then configured to point at that area of the file system for those configuration files. The XML files can be edited to fine tune control of the log output, but generally speaking the current configuration is likely best for everyone’s general purposes, so if changes are made please revert them prior to commit.

For Providers

Graph Providers are those who develop third-party systems and libraries on top of the various TinkerPop APIs and protocols. They manage their projects independently of TinkerPop in separate repositories. Additional details for providers can be found in the Provider Documentation.

Release Process

This document describes the steps required to release a version of TinkerPop. The release is handled by a "release manager" (a committer fulfills this role), who ensures that the steps in this document are executed. The process is multi-phased and can therefore take several weeks to complete given the time needed for Apache voting and community feedback. Once a release point has been identified, the following phases represent the flow of "release":

  • Release manager key setup.

  • Pre-flight check.

  • Optionally, produce a release candidate for community feedback.

  • Submit the official release for PMC vote.

  • Release and promote.

During release, it is best to use the current SNAPSHOT version of this documentation to ensure that you have the latest updates as almost every release includes improved documentation that may involve new or revised steps to execute.
The following instructions assume that the release manager’s environment is setup properly for release and includes a .glv files in the various GLV modules, so that they all build in full.
The release process needs to be conducted in a Linux environment.

Development Versions

A "development version" or "snapshot" (in Java parlance) is not an "official" release. Artifacts produced for a snapshot are solely for the convenience of providers and other developers who want to use the latest releases of TinkerPop. These releases do not require a VOTE and do not require a "release manager". Any PMC member can deploy them. It is important to note that these releases cannot be promoted outside of the developer mailing list and should not be recommended for use beyond the purpose of evaluation and testing.

A development release must contain the suffix "-SNAPSHOT" in the pom.xml.

For JVM-based artifacts, simply use the following command:

mvn clean deploy

and artifacts will be pushed to the Apache Snapshot Repository. GLV development artifacts must be generated and deployed separately with additional commands:

mvn clean install -Pglv-python
mvn deploy -pl gremlin-python -Dpypi
mvn clean install -pl :gremlin-dotnet,:gremlin-dotnet-source -Dnuget
mvn deploy -pl :gremlin-dotnet-source -Dnuget
mvn deploy -pl gremlin-javascript -Dnpm`

Python, .NET and Javascript do not use the snapshot model the JVM does, however, the build is smart in that it will dynamically generate a development version number for the GLV artifacts when "-SNAPSHOT" is in the pom.xml.

If you wish to verify that mvn deploy works before doing it officially then:

  • For Python, use the testpypi test environment by updating the gremlin-python/pom.xml.

  • For .NET, use the staging.nuget.org environment by updating the gremlin-dot-source/pom.xml

The clean in the above commands is more important to the pypi deployment because the process will deploy anything found in the target/python-packaged/dist directory. Since the names of the artifacts are based on timestamps, they will not overwrite one another and multiple artifacts will get uploaded.
These commands will dynamically edit certain files in the various GLVs given automated version number changes. Take care to commit or not commit changes related to that as necessary.

Release Manager Requirements

If this is your first time as release manager, you will need to set up keys for signing purposes per the Apache release process. Generally speaking, this will mean that you will need to generate a key-pair and then publish your public key.

For a general overview of key basics, refer to this. For detailed step-by-step instructions, please follow the instructions here.

After completing the key-pair setup instructions, be sure to upload your public key to keys.openpgp.org and add it into your Apache LDAP entry (under OpenPGP Public Key Primary Fingerprint). This ensures that the key will be added automatically to this list of committer keys which will be used to validate the distribution.

Your public key also needs to be added to KEYS files in both the development and release distribution directories and committed using Apache Subversion (SVN).

If you are a committer, you will require the assistance of a PMC member to upload your key into the release distribution directory. This can also be done during Release & Promote when release files are uploaded.

Pre-flight Check

The "pre-flight check" is a list of things performed by the release manager during the weeks leading up to a scheduled day to release. These checks will help to ensure that that release day goes smoothly by identifying problems up early and communicating with other members of the community.

  1. If this release cycle includes a major release, ensure that all upgrades for language runtimes have been considered.

  2. Fourteen days before release, issue an email to the dev mailing list to remind the community of the pending release.

    1. Note any important issues open in JIRA in that post.

    2. Request review and update of the "upgrade documentation" and CHANGELOG.

  3. Seven days before release, announce the code freeze on the dev mailing list to remind the community that the branch under release is protected. Tweaks to documentation and other odds and ends related to release are still allowed during this period.

  4. At some point during the week:

    1. Run a local build mvn clean install

    2. Run the full integration test suite: docker/build.sh -t -i -n

    3. Build and test the Docker images: mvn clean install -pl gremlin-server,gremlin-console -DdockerImages

    4. Ensure that the Gremlin.Net.Template gets packaged successfully: mvn clean install -pl :gremlin-dotnet-source -Dnuget

    5. Deploy a final SNAPSHOT to the Apache snapshot repository for Java.

    6. Review LICENSE and NOTICE files to make sure that no changes are needed.

    7. Review javadoc filters on the "Core API" docs to be sure nothing needs to change.

    8. Review JIRA tickets in the release and ensure that:

      1. All tickets categorized by having a "Component" assigned.

      2. All tickets are either of type "Bug" or "Improvement".

      3. All tickets where work was completed are "Closed"

        1. Search for "closed the pull request" in comments for hints on possible tickets that were left open by mistake.

        2. Look for tickets marked as "Resolved" as some users might not have rights to mark as "Closed" - convert these to "Closed".

      4. All tickets not marked "Fixed", "Done", or "Implemented" for their Resolution should not have a Fix Version assigned (use common sense when reviewing these tickets before removing the Fix Version as it is possible the incorrect Resolution may have been assigned).

  5. When all documentation changes are in place, use bin/publish-docs.sh to deploy a final SNAPSHOT representation of the docs and thus validate that there are no issues with the documentation generation process. Request review of the published documentation on the dev mailing list.

Release Candidate

A release candidate is an unofficial release that is represented by a tagged version in the Git repository. It is offered in cases where there is significant change in a particular version and the potential for upgrades and problems might be high. Release candidates do not require a vote thread. Lazy consensus is acceptable for confirming their deployment.

  1. mvn clean install

  2. docker/build.sh -t -i -n

  3. bin/publish-docs.sh <username> - note that under a release candidate the documentation is published as SNAPSHOT

  4. mvn versions:set -DnewVersion=xx.yy.zz -DgenerateBackupPoms=false to update the project files to reference a non-SNAPSHOT version

  5. mvn clean install -DskipTests to update GLV versions

    1. Review Gremlin.Net.csproj, Gremlin.Net.Template.csproj and Gremlin.Net.Template.nuspec in gremlin-dotnet

    2. Review package.json and package-lock.json in gremlin-javascript and in gremlint

  6. pushd gremlin-console/bin; ln -fs ../target/apache-tinkerpop-gremlin-console-xx.yy.zz-standalone/bin/gremlin.sh gremlin.sh; popd

  7. git diff and review the updated files

  8. git commit -a -m "TinkerPop xx.yy.zz release" and git push

  9. git tag -a -m "TinkerPop xx.yy.zz release" xx.yy.zz and git push --tags

  10. mvn clean install

  11. mvn versions:set -DnewVersion=xx.yy.zz-SNAPSHOT -DgenerateBackupPoms=false to go back to SNAPSHOT

  12. mvn clean install -DskipTests to update GLV versions

    1. Review Gremlin.Net.csproj, Gremlin.Net.Template.csproj and Gremlin.Net.Template.nuspec in gremlin-dotnet

    2. Review package.json and package-lock.json in gremlin-javascript and in gremlint

  13. pushd gremlin-console/bin; ln -fs ../target/apache-tinkerpop-gremlin-console-xx.yy.zz-SNAPSHOT-standalone/bin/gremlin.sh gremlin.sh; popd

  14. git commit -a -m "Returned to xx.yy.zz-SNAPSHOT" and git push

  15. Announce the release candidate to dev mailing list and await feedback

  16. Repeat as required or proceed to the next phase

Note that release candidates need not require the release of all artifacts. For example, if the risk is with one particular GLV, then a release candidate can be prepared of just that particular artifact. In those cases, a tag on the commit that represents the release candidate is sufficient and does not necessarily require that the versions be bumped to reflect the actual "-rc" version. In other words, if preparing a release candidate for .NET, then there is no need to go through the processing of bumping versions in Java artifacts and all GLVs. Nor is it necessary to alter the version of .NET to include the release candidate versioning. It can simply be altered locally by the release manager and deployed.

PMC Vote

This section describes the process that process that prepares a release for VOTE by the community. If there are multiple releases (as there usually are) being prepared, it may be best for downstream releases to wait for upstream releases to complete their process to assure that no last minute commits are required to get the upstream release completed. It is up to the discretion of the release managers to decide how they wish to proceed with respect to preparing releases in parallel or in a more serial fashion.

  1. By this point, the testing performed during the code freeze should have validated the release. If however there are additional tests to perform that the release manager feels are relevant, they should be performed now. In other words, there is no need to rebuild the SNAPSHOT yet another time unless there are circumstances that would call its validity into question.

  2. Update CHANGELOG.asciidoc:

    1. Update the release date - the release date is the date of the vote. Double check the header to be sure that it is formatted properly - specifically, ensure that there is an asciidoc anchor of the format above the version/date header. The release date is the intended day of the vote and not necessarily when it happens (usually falls on a Monday given the code freeze pattern that starts a week prior on a Friday). The release date for all versions should be the same as the first release of the set even if the release vote does not begin on the same day. In other words if 3.2.x has a vote date of "October, 1 2018" then 3.3.x should also have that date even if the vote thread for 3.3.x doesn’t start until the day after.

    2. Generate the JIRA release notes report for the current version and append them to the CHANGELOG.asciidoc.

      1. Use an "advanced" search to filter out JIRA issues already released on other versions. For example: project = TINKERPOP and status = Closed AND fixVersion = 3.2.0 ORDER BY type, Id ASC.

      2. Consider use of an "Excel" export to organize and prepare the JIRA tickets to be pasted to CHANGELOG.asciidoc. This formula can help construct each line item for the CHANGELOG if column A is the issue number, B is the issue title and D is the label field: ="* "&A2&" "&B2&(IF(D2="breaking"," *(breaking)*",""))

      3. Be sure to include a link to other versions in the CHANGELOG.asciidoc that were previously released while the current release was under development as this new release will have those changes included within it. Please see 3.2.1 for an example.

    3. Format "breaking" changes to be clearly marked (use JIRA and the "breaking" label to identify those - already accounted for if using the suggested formula above)

  3. Update "upgrade documentation":

    1. Update the release date.

    2. Update the link to CHANGELOG.asciidoc - this link may already be correct but will not exist until the repository is tagged.

  4. mvn versions:set -DnewVersion=xx.yy.zz -DgenerateBackupPoms=false to update project files to reference the non-SNAPSHOT version

  5. mvn clean install -DskipTests to update GLV versions

    1. Review Gremlin.Net.csproj, Gremlin.Net.Template.csproj and Gremlin.Net.Template.nuspec in gremlin-dotnet

    2. Review package.json and package-lock.json in gremlin-javascript and in gremlint

  6. pushd gremlin-console/bin; ln -fs ../target/apache-tinkerpop-gremlin-console-xx.yy.zz-standalone/bin/gremlin.sh gremlin.sh; popd

  7. git diff and review the updated files

  8. mvn clean install - need to build first so that the right version of the console is used with bin/publish-docs.sh

    1. This step should update the Gremlin.Net project file and Gremlin Javascript package file with the newly bumped version.

  9. bin/process-docs.sh and validate the generated documentation locally. Don’t rely on "BUILD SUCCESS" - scroll up through logs to ensure there were no errors and view the HTML directly. Code blocks that did not execute properly have a gray background and do not show the results of the commands. In the event of failure, especially with GraphComputer related code examples, they best recourse is to take the time to test the example manually. While process-docs.sh tries to output the errors it can often point out a red herring where running the example manually will immediately showcase the problem. Unless you immediately know what is wrong, it is almost never a good idea to try to debug the problem based on the error shown by process-docs.sh.

  10. bin/publish-docs.sh <username> - Note that this step requires no additional processing as the previous step handled document generation and this step now merely needs to upload what was generated. Note that this step will be responsible for generating javadoc.

  11. mvn deploy -Papache-release -DcreateChecksum=true -DskipTests - deploy signed artifacts with checksums to Apache Nexus.

  12. Review generated artifacts (request another committer to review as well) then "close" the repo - if the repo is left open it will be automatically dropped after five days and closing the repo will allow it to stay available for a full ninety days which is more than enough time to complete a vote. Do NOT "release" the repository at this time.

  13. Upload artifacts to https://dist.apache.org/repos/dist/dev/tinkerpop for [VOTE] review.

    1. Use svn rm to delete past versions that were up for review in the same line of code. In other words, if uploading 3.2.3 then remove instances of 3.2.2 or any other past 3.2.x releases.

    2. svn co --depth empty https://dist.apache.org/repos/dist/dev/tinkerpop/ dev and mkdir dev/xx.yy.zz

    3. cp ~/.m2/repository/org/apache/tinkerpop/gremlin-console/xx.yy.zz/gremlin-console-xx.yy.zz-distribution.zip* dev/xx.yy.zz

    4. cp ~/.m2/repository/org/apache/tinkerpop/gremlin-server/xx.yy.zz/gremlin-server-xx.yy.zz-distribution.zip* dev/xx.yy.zz

    5. cp ~/.m2/repository/org/apache/tinkerpop/tinkerpop/xx.yy.zz/tinkerpop-xx.yy.zz-source-release.zip* dev/xx.yy.zz

    6. cd dev/xx.yy.zz

    7. rm -f *.md5

    8. for file in *.sha1 ; do artifact=$(basename ${file} .sha1); sha512sum $artifact | awk '{print $1}' > ${artifact}.sha512; rm ${file}; done

    9. ls * | xargs -n1 -I {} echo "mv apache-tinkerpop-{} {}" | sed -e 's/distribution/bin/' -e 's/source-release/src/' -e 's/tinkerpop-tinkerpop/tinkerpop/' -e s'/^\(.*\) \(.*\) \(.*\)$/\1 \3 \2/' | /bin/bash

    10. cd ..; svn add xx.yy.zz/; svn ci -m "TinkerPop xx.yy.zz release"

  14. Execute bin/validate-distribution.sh and any other relevant testing.

  15. git commit -a -m "TinkerPop xx.yy.zz release" and push

  16. git tag -a -m "TinkerPop xx.yy.zz release" xx.yy.zz and git push origin xx.yy.zz

  17. Submit for [VOTE] at dev@tinkerpop.apache.org (see email template below)

  18. Wait for vote acceptance (72 hours)

Release & Promote

Certain steps can only be performed by a PMC member. If you are a committer, please enlist the help of a PMC member for those steps.
  1. Login to Apache Nexus and release the previously closed repository.

  2. Deploy the GLVs

    1. This build will likely occur from the tag for the release, so be sure to checkout the tag first before executing this step.

    2. mvn clean install -DskipTests -Dnuget

    3. mvn deploy -pl gremlin-python -DskipTests -Dpypi

    4. mvn deploy -pl :gremlin-dotnet-source -DskipTests -Dnuget

    5. mvn deploy -pl gremlin-javascript -DskipTests -Dnpm

      1. Consider direct deploy with npm publish --otp <otp> from gremlin-javscript/src/main/javascript/gremlin-javascript/ - if using this method, ensure the npm version aligns with the one expected by the branch.

    6. mvn deploy -pl gremlint -DskipTests -Dnpm

      1. Consider direct deploy with npm publish --otp <otp> from gremlint/ - if using this method, ensure the npm version aligns with the one expected by the branch.

    7. git tag gremlin-go/v3.6.3

    8. git push origin gremlin-go/v3.6.3

    9. GOPROXY=proxy.golang.org go list -m github.com/apache/tinkerpop/gremlin-go/v3@v3.6.3

  3. Review the GLV releases

  4. Deploy the Docker images (PMC member)

    1. mvn deploy -pl gremlin-console -DskipTests -DdockerImages

    2. mvn deploy -pl gremlin-server -DskipTests -DdockerImages

  5. Review the deployed Docker images at Console and Server

  6. svn co --depth empty https://dist.apache.org/repos/dist/dev/tinkerpop dev; svn up dev/xx.yy.zz (PMC member)

  7. svn co --depth empty https://dist.apache.org/repos/dist/release/tinkerpop release; mkdir release/xx.yy.zz (PMC member)

  8. Copy release files from dev/xx.yy.zz to release/xx.yy.zz. (PMC member)

  9. cd release; svn add xx.yy.zz/; svn ci -m "TinkerPop xx.yy.zz release" (PMC member)

  10. Wait for Apache Sonatype to sync the artifacts to Maven Central.

  11. Report the release through reporter.apache.org (an email reminder should arrive shortly following the svn command above to do the release)

  12. Update homepage with references in /site to latest distribution and to other internal links elsewhere on the page.

    1. Typically, this need only be done on the master branch as we only publish from there.

    2. Update index.html.

    3. Update community.html, specifically the version numbers for our official releases.

    4. Update Downloads page, when moving "Current Releases" to "Archived Releases" recall that the hyperlink must change to point to version in the Apache Archives.

    5. Update docs/gremlint/package.json with the latest version of gremlint.

    6. Preview changes locally with bin/generate-home.sh then commit changes to git.

  13. Wait for zip distributions to sync to the Apache mirrors (i.e ensure the download links work from a mirror).

  14. bin/publish-home.sh <username> to publish the updated web site with new releases.

  15. Execute bin/update-current-docs.sh <username> to migrate to the latest documentation set for /current.

  16. This step should only occur after the website is updated and all links are working. If there are releases present in SVN that represents lines of code that are no longer under development, then remove those releases. In other words, if 3.2.0 is present and 3.2.1 is released then remove 3.2.0. However, if 3.1.3 is present and that line of code is still under potential development, it may stay.

  17. Announce release - see templates for suggested form

    1. Email dev@/gremlin-users@ mailing lists

    2. Tweet from @apachetinkerpop (PMC member)

    3. Message the "Announcement" Discord channel (PMC member)

Post-release Tasks

A number of administration tasks should be taken care of after the release is public. Some of these items can be performed during the VOTE period at the release manager’s discretion, though it may be wise to wait until a successful VOTE is eminent before reopening development. When there are multiple release managers, it’s best to coordinate these tasks as one individual may simply just handle them all.

Certain steps can only be performed by a PMC member. If you are a committer, please enlist the help of a PMC member for those steps.
  1. Perform JIRA administration tasks (PMC member):

    1. "Release" the current version and set the "release date"

    2. If there is to be a follow on release in the current line of code, create that new version specifying the "start date"

  2. Prepare Git administration tasks. Apply the following steps as needed per release branch:

    1. Make the appropriate branching changes as required by the release and bump the version to SNAPSHOT with mvn versions:set -DnewVersion=xx.yy.zz-SNAPSHOT -DgenerateBackupPoms=false.

  3. mvn clean install -DskipTests to update GLV versions

    1. Review Gremlin.Net.csproj, Gremlin.Net.Template.csproj and Gremlin.Net.Template.nuspec in gremlin-dotnet

    2. Review package.json and package-lock.json in gremlin-javascript and in gremlint

    3. pushd gremlin-console/bin; ln -fs ../target/apache-tinkerpop-gremlin-console-xx.yy.zz-SNAPSHOT-standalone/bin/gremlin.sh gremlin.sh; popd

    4. Update CHANGELOG and upgrade docs to have the appropriate headers for the next version.

    5. mvn clean install -DskipTests - need to build first so that the right version of the console is used with bin/publish-docs.sh

    6. mvn deploy -DskipTests - deploy the new SNAPSHOT

    7. bin/process-docs.sh and validate the generated SNAPSHOT documentation locally and then bin/publish-docs.sh <username>

    8. Commit and push the SNAPSHOT changes to git

  4. Examine the future.asciidoc and update the "Roadmap" as needed.

  5. Send email to advise that code freeze is lifted.

  6. Consider the changes made to Gremlin and determine if the community needs to organize a PR to [DefinitelyTyped](https://github.com/DefinitelyTyped/DefinitelyTyped/tree/master/types/gremlin) to keep TypeScript up to date.

  7. Ensure that the GLV examples compile and run with the latest image and dependencies: bin/run-examples.sh.

    1. Make changes as necessary to update the examples.

  8. Generate a list of dead branches that will be automatically deleted and post them as a DISCUSS thread for review, then once consensus is reached removed those branches.

  9. Set up the IO tests for the current SNAPSHOT as discussed in the IO Documentation and Testing Section

Apache provides access to download statistics for release here.

Email Templates

Release VOTE

Subject: [VOTE] TinkerPop xx.yy.zz Release


We are happy to announce that TinkerPop xx.yy.zz is ready for release.

The release artifacts can be found at this location:

The source distribution is provided by:

Two binary distributions are provided for user convenience:

The GPG key used to sign the release artifacts is available at:

The online docs can be found here:
        https://tinkerpop.apache.org/docs/xx.yy.zz/ (user docs)
        https://tinkerpop.apache.org/docs/xx.yy.zz/upgrade/ (upgrade docs)
        https://tinkerpop.apache.org/javadocs/xx.yy.zz/core/ (core javadoc)
        https://tinkerpop.apache.org/javadocs/xx.yy.zz/full/ (full javadoc)
        https://tinkerpop.apache.org/dotnetdocs/xx.yy.zz/ (.NET API docs)
        https://tinkerpop.apache.org/jsdocs/xx.yy.zz/ (Javascript API docs)

The Git tag can be found here:

The release notes are available here:

The [VOTE] will be open for the next 72 hours --- closing <DayOfTheWeek> (<Month> <Day> <Year>) at <Time> <TimeZone>.

My vote is +1.

Thank you very much,
<TinkerPop Committer Name>


Subject: [RESULT][VOTE] TinkerPop xx.yy.zz Release

This vote is now closed with a total of X +1s, no +0s and no -1s. The results are:


+1  (X -- list of voters)
0   (0)
-1  (0)


+1 (X -- list of voters)
0  (0)
-1 (0)

Thank you very much,
<TinkerPop Committer Name>

General Release Announcement

The template below refers to the "name of release line" and the "release line logo". Every release line has a name and logo. For example, 3.1.x had the name, "A 187 On The Undercover Gremlinz" and the logo shown here in the upgrade documentation.

Subject: Apache TinkerPop xx.yy.zz Released: [name of release line]


Apache TinkerPop xx.yy.zz has just been released. [some text to introduce the release - e.g. whether or not
there is breaking change, an important game-changing feature or two, etc.]

The release artifacts can be found at this location:


The online docs can be found here:

https://tinkerpop.apache.org/docs/xx.yy.zz/reference/ (user docs)
https://tinkerpop.apache.org/docs/xx.yy.zz/upgrade/#_tinkerpop_xx_yy_zz (upgrade docs)
https://tinkerpop.apache.org/javadocs/xx.yy.zz/core/ (core javadoc)
https://tinkerpop.apache.org/javadocs/xx.yy.zz/full/ (full javadoc)
https://tinkerpop.apache.org/dotnetdocs/xx.yy.zz/ (.NET API docs)
https://tinkerpop.apache.org/jsdocs/xx.yy.zz/ (Javascript API docs)

The release notes are available here:


The Central Maven repo has sync'd as well:


Go artifacts are available via Go Package Manager:


Python artifacts are available in pypi:


.NET artifacts are available in NuGet:


Javascript artifacts are available in npm:


Gremlint is available in npm:


Docker images for Gremlin Console and Gremlin Server can be found on Docker


[include the release line logo image]

Standard Tweet Text

The tweet from @apachetinkerpop doesn’t need to really adhere to any specific format necessarily, but for general patch releases, the following simple text is common:

Apache TinkerPop xx.yy.zz Released. [name of release] [link to gremlin-users announcement] #graphdb #nosql

Remember to include the picture for the release with the tweet.


New Committers

When a candidate is identified by a PMC member as someone who might be a good official committer to TinkerPop, the PMC member should open a DISCUSS thread on the private TinkerPop mailing list. The thread should provide some background and reasoning for why that member believes the candidate would be a good committer. Given enough time for feedback on the candidate and presuming there is still positive interest in doing so, a VOTE thread on the private TinkerPop mailing list is started to get the official stance. As per usual, the VOTE will be made open for no less than 72 hours.

If the VOTE closes with a successful positive vote to make the candidate a committer, then send the following email to the candidate and copy the private TinkerPop mailing list:

SUBJECT: Invitation to become TinkerPop committer: [candidate name]


The TinkerPop Project Management Committee (PMC) hereby offers you committer
privileges to the project. These privileges are offered on the understanding
that you'll use them reasonably and with common sense. We like to work on
trust rather than unnecessary constraints.

Being a committer enables you to more easily make changes without needing to
go through the patch submission process.

Being a committer does not require you to participate any more than you already
do. It does tend to make one even more committed.  You will probably find that
you spend more time here.

Of course, you can decline and instead remain as a contributor, participating
as you do now.

A. This personal invitation is a chance for you to accept or decline in
private.  Either way, please let us know in reply to the
private@tinkerpop.apache.org address only.

B. If you are accepting, the next step is to register an iCLA with the Apache
Software Foundation:
  1. Details of the iCLA and the forms are found through this link:

  2. The form (text or PDF version) provides instructions for its completion
     and return to the Secretary of the ASF at secretary@apache.org.

  3. When you complete iCLA ensure that you include "Apache TinkerPop" in the
     "notify project" project field and choose a preferred unique Apache id.
     Look to see if your preferred id is already taken at
     http://people.apache.org/committer-index.html. This will allow the
     Secretary to notify the PMC when your iCLA has been recorded.

When recording of your iCLA is noticed, you will receive a follow-up message
with the next steps for establishing you as a committer.

Assuming the iCLA included items from step 3, the secretary will handle account creation, otherwise it will be up to the PMC chair to handle such things. Once the account is established the PMC chair will then need to:

  1. Validate that the iCLA is on file either through svn or through people.apache.org.

  2. Request that the account be created. The PMC Chair may do this through the Account Request Form.

  3. Once verified, provide the new committer access to the repository, which is most easily done through Whimsy.

  4. Send an announcement email to the developer and user mailing lists with the template below.

  5. Update JIRA to include this person in the "committers" group.

SUBJECT: New Committer: [committer name]

The Project Management Committee (PMC) for Apache TinkerPop has asked
[committer name] to become a committer and we are pleased to announce their

[describe the nature of the committer's work in the community]

Being a committer enables easier contribution to the project since there is no
need to work via the patch submission process. This should enable better

Finally, new committers should be sent an email that covers some of the administrative elements of their new role. Generally speaking the email can largely just point them to the Initial Setup section of this documentation.

New PMC Members

The process for bringing on new PMC members is not so different from the one for new committers. The process begins with a DISCUSS thread to the private mailing list for building consensus followed by a VOTE thread to confirm. Presuming the new PMC member is a committer already (which is mostly likely for TinkerPop), there should be no need for any additional paperwork. On successful vote however, a NOTICE should be sent to the board@apache.org (copying private@tinkerpop.apache.org):

SUBJECT: [NOTICE] [name of PMC new member] for Apache TinkerPop PMC

Apache TinkerPop proposes to invite [name of PMC new member] ([ApacheId]) to join the PMC.

The vote result is available here:

[include the link to the vote result thread from private]

If the candidate does not (yet) have an Apache account, then please note that fact in the notification email. It is important to verify that this NOTICE was actually included in the board archive to ensure that is has been delivered to the board. After verifying this, the invite may be sent to the new PMC member. The following template may be used for the invitation (be sure to copy private@ when sending the email):

SUBJECT: Invitation to join the Apache TinkerPop PMC: [name of PMC new member]

In recognition of your contributions to Apache TinkerPop, the TinkerPop PMC has voted to make you a PMC member. If you are unfamiliar with what the PMC does, please see this link to be sure you are comfortable with the responsibilities:


As you have already signed an ICLA and have an Apache account no additional paperwork is required by you other than replying to this email (ensuring to include private@ which is CC'd) to say if you accept or reject the invitation. Obviously there is no requirement to accept - you may simply choose to continue working in the capacity of a committer as you are now. We're happy to answer any questions you might have.

Thank you for your efforts on the project.

If they accept, then the PMC chair should refer to the How to Add a New PMC Member section in the Apache docs for how to complete the process. Use Whimsy to update PMC membership.

Board Reports

The PMC Chair is responsible for submitting a report to the board on a quarterly basis. TinkerPop reports on the following months: January, April, July, October.

A draft of the report should be sent to the TinkerPop developer mailing list for review at least three days prior to submitting to the board. The final report can be submitted via Whimsy or via the Apache Reporter Service which already provides a template for the report and explains the different sections.

Contributor Listing

Contributions can come in many forms and extend well beyond code contributions. TinkerPop strives to ensure that contributions are honestly and accurately recognized for the benefit of the individuals who help make TinkerPop possible.

One of the ways in which this recognition is supported is by way of the Contributor Listing on the TinkerPop home page, which list those who have contributed to the project in some way. It lists both committers and PMC members and provides a short "bio" (i.e. a description of their contributions) as well as an indication of whether they are currently participating in the project or not, i.e. active or inactive. The bios have a slightly different form depending on that status:

  • An active bio should be reflective of current contributions, roughly representing the past six months and the following six months of the update.

  • An inactive bio for those not currently contributing to the project, should reflect the full scope of all contributions made by that individual during their active periods.

Irrespective of being active or inactive, the contributor’s name and tenure accomplishments remain present on the front page of the web site. Being "inactive" does not affect the contributors status as an Apache committer or PMC member - that remains unchanged. Should a contributor become "inactive" at some point, there is no special process to become "active" again - they simply update their bio to do so.

Committers and PMC members can keep this information up to date themselves as they have direct access to the Git repository. Reminders to update this information will be sent to the dev list on each release as part of code freeze week. Those who have not shown an update to their bio in git in the past year or have otherwise not replied to the dev list email to indicate their status in that time will be assumed "inactive" for the purpose of the listing.

The following is a template for the "Bio Update" email:

Subject: Apache TinkerPop Bio Update

To Committers/PMC Members,

As an Apache TinkerPop committer and/or PMC member, your name is listed on the TinkerPop home page in the Contributor List[1] with your "bio". If you are active on the project, your "bio" reflects what you have been working on and what you expect to be working on with respect to TinkerPop for recent times (i.e. for the previous six months and the following six months). If you are currently inactive on the project, your "bio" reflects the full scope of all your contributions throughout your active periods. You can refer to the contributor listing policy[2] for full details.

Please take a moment to update your bio directly in Git[3] or, if you would prefer, please reply to this email with your bio update and it will be added for you. If no changes are required, please reply to this email to confirm that this is the case.

[1] https://tinkerpop.apache.org/#contributors
[2] https://tinkerpop.apache.org/docs/current/dev/developer/#contributor-listing
[3] https://github.com/apache/tinkerpop/blob/master/docs/site/home/index.html


November 23, 2015

The meeting was scheduled for 12:00pm EST, started at 12:05pm when sufficient attendance to constitute a quorum was reached and was held via Google Hangout hosted by Marko Rodriguez. Meeting adjourned at 2:35pm EST.

Committers/PPMC Members

  • Daniel Kuppitz

  • Stephen Mallette

  • Dylan Millikin

  • Jason Plurad

  • Marko Rodriguez

  • Ted Wilmes




  • Performed a full review of the list of unresolved tickets in JIRA. The summary of the changes can be found below. Note that releases and assignments made today are not meant to be set in stone. The assignments could change given community discussion or unforeseen events.

  • Decided to add version 3.2.0-incubating to JIRA. This version was added to help better categorize the available issues. It was explicitly decided that no development was expected to be performed on these issue at this time and that we would wait until 3.1.1-incubating was released before thinking too hard on that body of work. This decision means that the master branch will continue to be the current branch of development for the time being and be bound to 3.1.x line of code.

  • Discussed use of JIRA dependency links to show "blockers" and dependencies among different JIRA tickets.

  • Discussed a target date for upcoming releases:

    • 3.1.1-incubating - February 1, 2016

    • 3.2.0-incubating - May, 1, 2016

The following tickets were assigned to 3.1.1-incubating:

ID Description Type


MessageScope.Local.setStaticMessage(M msg)



GraphConfiguration Class



Automatic Traversal rewriting



IsStep broken when profiling is enabled.



Consider a P.instanceOf()



Do we need runtime BigDecimal in more places?



MapReduceHelper should sort respective of the number of reduce tasks



Provide a more general way to set log levels in plugins



Bindings applied to the PluginAcceptor should appear to Gremlin Server



Develop a less error prone way for rewriting strategies



Allow any GraphReader/Writer to be persistence engine for TinkerGraph



Re-examine Sandboxing Abstractions



Validate dependency grabs that have TinkerPop dependencies



Improve the ability to embed Gremlin Server with Channelizer injection



Test case needed for ensuring same cardinality for key.



bin/publish-docs.sh should only upload diffs.



Tie Alias to Transaction Manager in Gremlin Server



Add ability to cancel script execution associated with a Gremlin Server Session



Move the implementations sections of the primary documentation to "provider tutorials."



Add a "clear SNAPSHOT jars" section to the process-docs.sh.



Neo4jGraph should support HighAvailability (Neo4jHA).



Improve error message for wrong order().by() arguments



Warn if Gremlin Server is running prior to generating docs



Barrier steps provide unexpected results in Gremlin OLAP



Consistent test directory usage



Connection errors tend to force a complete close of the channel



Improve usability of .profile() step.



Test XXXGraphComputer on a Hadoop2 cluster (non-pseudocluster).



Support nested-repeat() structures



BLVP shouldn’t clear configuration properties



Fail earlier if invalid version is supplied in validate-distribution.sh



Native TinkerGraph Serializers for GraphSON



ComputerVerificationStrategy not picking up Order local traversal



Add a service script or daemon mode in the distribution



Deprecate support for credentialsDbLocation in Gremlin Server Config



Provide a way to track open Graph instances in tests



Use GraphProvider for id conversion in Groovy Environment test suite



Manage binary LICENSE/NOTICE files with bin/checkLicenseNotice.sh



SparkGraphComputer.submit shouldn’t use ForkJoinPool.commonPool


The following items were identified for 3.2.0-incubating:

ID Description Type


TraverserConverterStep (proposal)



Commutative Step Marker interface



Consider Required TraversalStrategies



TP3 is too prescriptive in exception



Serializer Handshake



Support "barrier syntax" in step labels.



Some basic mathematical functions / steps



Patterns for DSL Development



Choose then Enforce Semantics for Graph.close()



Implement AutoCloseable on TraversalSource



[Proposal] Domain/Range checking during traversal construction.



[Proposal] Make the Gremlin Graph Traversal Machine and Instruction Set Explicit



Mapping Cardinality Interface



How should OLAP treat Collection<Element> objects? No contract is specified.



PropertyMapStep should reuse PropertiesStep



Operator.mean would be nice.



RuntimeStrategy as the general model for all such execution time rewrites/re-orders



Remove GroupCountStep in favor of new Reduce-based GroupStep



FoldStep should default to using BulkSet for efficiency.



Refactor Gremlin Server integration tests to be Client parameterized



Support for partitioned vertices in GraphComputer



Remove the concept of branch/ package.



Features needs to specify whether 2 vertex properties with same key/value is allowed.



Convert LocalTraversals to MatchSteps in OLAP and Solve the StarGraph Problem



Use EventStrategy to solve OLAP bulk mutation of OLTP.



Consider deprecating or better enforcing Graph.Exceptions.elementNotFound



Traversal respecting Thread.interrupt()



Enforce semantics of threaded transactions as manual



Add a Bulk class which is used by Traverser



Provide "vertex query" selectivity when importing data in OLAP.



SubgraphTraversalAnalyzer to determine what is really required from a traversal.



Support reversible traversals in MatchStep (and respective MatchAlgorithms)






TraversalSource should be fluent like GraphComputer



Saving headless traversals for reuse (clone Iterator Fun)



valuesDecr, valuesIncr, keysDecr, and valuesDecr is lame.


The following issues were simply closed during review - the reasons for closing can be found in the comments of the issues themselves:

ID Description Type


Better Methods for Managing ClassPath for Plugins



User Supplied Ids and IO



Subgraph support for VertexProgram


DetachedEdge.attach(Vertex) is too slow.



Provide Traverser.setPath()



Add a TraversalSourceStrategy that provides "locked" values.



Shorthand for install of TinkerPop dependencies



Check feature requirements before opening graph during tests


October 29, 2015

The meeting was scheduled for 1:00pm EST, started on time and was held via Google Hangout hosted by Stephen Mallette. Meeting adjourned at 1:45pm EST.

Committers/PPMC Members

  • Daniel Kuppitz

  • Stephen Mallette

  • Marko Rodriguez




  • Reviewed the scope of 3.1.0-incubating in JIRA in the context of the upcoming release date on November 16, 2015.

  • It was noted that with the new one week code freeze policy that the real cut-off date for work is November 9, 2015.

  • There was general consensus on the call that work should be accomplished in such a way that the code review process not drag into the code freeze time period. In other words, pull requests to the release branch should be completed well ahead of the 9th.

Upon review of the tickets in JIRA assigned to 3.1.0-incubating, the following were removed from the 3.1.0-incubating roadmap:

ID Description Removal Reasoning


Re-examine Sandboxing Abstractions

Low priority and time limitations.


Remove the concept of branch/ package.

Awaiting step library definition in 3.2.0-incubating.


FoldStep should default to using BulkSet for efficiency.

Awaiting step library definition in 3.2.0-incubating.


Operator.mean would be nice.

Awaiting step library definition in 3.2.0-incubating.


Add a TraversalSourceStrategy that provides "locked" values.

Low priority and time limitations.


Remove PropertyMapStep in favor of multi-instruction construction.

Non-trivial given time limitations.


Choose then Enforce Semantics for Graph.close()

Non-trivial given time limitations.


MatchStep in OLAP should be smart about current vertex.

Non-trivial given time limitations.


Make use of a MemoryRDD instead of Accumulators in SparkGraphComputer?

Non-trivial given time limitations.


GraphConfiguration Class

Non-trivial given time limitations.


More output for OLAP jobs

Not clearly defined given time limitations.


Commutative Step Marker interface

Awaiting step library definition in 3.2.0-incubating


Gremlin IO needs to support both OLTP and OLAP naturally.

Not clearly defined given time limitations.


Consider Providing "getOrCreate" Functionality

Not clearly defined given time limitations.