The Impala project uses Gerrit for all our code reviews. Gerrit is a git-based code review tool. The Impala project Gerrit server is here.
To authenticate with Impala's Gerrit server, you'll need a Github account. Once you have one, logging in to Gerrit is as easy as clicking 'Github Sign-in' in the top-right corner.
You will need to authorize Gerrit to access some details of your Github account, and then you should be brought back to the Gerrit homepage. Unfortunately, there's a bug in the Gerrit OAuth plugin that will send you to a page that gives a 404 error. You'll need to manually go back to https://gerrit.cloudera.org/, and at that point you should be signed in as 'Anonymous Coward (X)'.
If your Gerrit username is different than your username on the system you want to send patches from, set your "GERRIT_USER" environment variable to your Gerrit username.
Once you are logged in, click your user name in the top-right corner, and go to 'Settings'. Under 'SSH Public Keys' you'll need to paste in your public key:
cat ~/.ssh/id_\{r or d\}sa.pub
It's reported that RSA keys are denied in new OSes like Ubuntu 22.04 and MacOS 14 (Sonoma). Try other key types like edcsa or ed25519, e.g. try one of these
ssh-keygen -t edcsa -b521 ssh-keygen -t ed25519 -b 4096 |
While you're in the settings page, please make sure to update your profile with your name, and your e-mail address!
Gerrit works by keeping a Git repository for all the projects it maintains, including Impala. The source of truth for what is in Impala is the official Apache git server. Gerrit serves as a staging ground for reviewing patches, and once a patch is approved, a sort of waiting room while patches wait for a committer to officially move them to the Apache git repo.
Sending a patch to Gerrit is exactly the same as 'pushing' a sequence of commits to a remote Git repository:
git push --no-thin asf-gerrit HEAD:refs/for/master
To set up Gerrit as a git remote for your Impala repository, do the following:
cd ${IMPALA_HOME}
git remote add asf-gerrit ssh://<your-github-username>@gerrit.cloudera.org:29418/Impala-ASF |
If using HTTP to connect to Gerrit, use the following command to add the Gerrit remote is also works:
git remote add asf-gerrit http://<your-github-username>@gerrit.cloudera.org:8080/a/Impala-ASF |
Pay attention, when using HTTP to push/pull code, we need to input username/password. Do not use login account directly, we should get username/password from "Settings→HTTP Password" instead.
Gerrit keeps track of different patches by generating a unique Change-Id that is constant over all iterations of a patch (obviously the Git hash won't do, since that changes every time the patch changes). That Change-Id is appended to every commit message:
IMPALA-1726: Move JNI / Thrift utilities to separate header
To avoid pulling in jni.h in the codegen module (where it does not
compile), this patch moves the methods that depend on both Thrift and
JNI to a separate header where it can be included more precisely.
Change-Id: I4d97f1816b24149d9163dce59f31b2afe2642b11
Gerrit provides a "pre-commit hook" that automates Change-Id generation. Installing it is very easy:
cd ${IMPALA_HOME}
curl -o .git/hooks/commit-msg https://gerrit.cloudera.org/tools/hooks/commit-msg
chmod u+x .git/hooks/commit-msg
After the hook is installed, every commit you make will get its own Change-Id. Without it, Gerrit will reject all patches.
Commit all outstanding changes:
cd $IMPALA_HOME git add --update git commit |
Rebase against asf-gerrit/master (good practice to catch conflicts introduced by recent commits)
git fetch asf-gerrit git rebase asf-gerrit/master |
You might have to resolve merge conflicts during 'git rebase'. Follow the instructions, then continue with
git rebase --continue |
Squash everything down to a single commit:
git rebase -i asf-gerrit/master |
Make sure to pick only the first commit and 's' (squash) following lines, then edit the commit message into something coherent. Run the git log command to see that you're exactly one commit ahead of asf-gerrit/master
I have these helpers in my .bashrc to simplify branching from the latest trunk (branch-trunk my_new_branch) and rebase against the latest trunk (rebase-trunk).
|
Do not push directly to asf- |
To send a patch for review, you should push it to a Gerrit-specific 'branch' on the Gerrit remote. That branch has the form refs/for/<branch-name>. That patch will then be visible to code reviewers, who will review your code before it becomes part of Impala. If you want to see what a patch looks like without making it visible to code reviewers, push to the branch refs/drafts/<branch-name>. For the vast majority of patches the target branch will be master, which is where all mainline development happens. During releases, release managers may push commits to a specific release branch, however most new development will happen on master.
In the example below, we're pushing HEAD to be a draft change. Gerrit provides a URL for the review.
$ git push asf-gerrit HEAD:refs/drafts/master Counting objects: 3, done. Delta compression using up to 8 threads. Compressing objects: 100% (3/3), done. Writing objects: 100% (3/3), 332 bytes | 0 bytes/s, done. Total 3 (delta 2), reused 0 (delta 0) remote: Resolving deltas: 100% (2/2) remote: Processing changes: new: 1, refs: 1, done remote: remote: New Changes: remote: http://gerrit.cloudera.org:8080/8335 test [DRAFT] remote: To ssh://XXXX@gerrit.cloudera.org:29418/Impala-ASF |
Visit the URL that Gerrit provided to review the diff before sending it out to reviewers. Check that you've uploaded the right commit and that you've included what you wanted in the diff. When you've reviewed it yourself, hit the "Publish" button (
) to send the review to the community. This will make the review visible, but it won't be an official part of the Impala code base yet.
Note that Gerrit will send all patches between the HEAD of the target branch and your current branch's HEAD for review. Please therefore make certain that you are only pushing the commits you want reviewed: every separate commit translates to an e-mail in everyone's inbox!
The following does not apply if changes are only within the |
The following Jenkins jobs are available to all Impala contributors for testing patches they submit to Gerrit.
bin/run-all-tests.sh in the source tree is the best place to understand the meanings to the options of this job.Neither of these jobs write a comment or Verified bit or attempt to Submit a Gerrit patch set. A submitter ought to monitor his or her job as it progresses. A submitter is encouraged to manually link a successful job in a Gerrit review comment as demonstration of success. When testing a draft, the "Impala Public Jenkins" user needs to be added to the change, so that Jenkins can access it.
Access to these jobs now requires a login with the Contributor role on jenkins.impala.io. Please request access via dev@impala.apache.org, and include your Apache JIRA username. Credentials will be emailed to you. |
Once a change has a +2 from a committer and no -1 from any committer, it is ready to be verified and merged to gerrit. Remember that the source of truth for what is in Impala is the official Apache git server. Gerrit serves as a staging ground for reviewing patches, and once a patch is approved, a sort of waiting room while patches wait for a committer to officially move them to the Apache git repo.
To verify a patch, we use one of two different automated processes. Both are restricted to use by committers. If you are not a committer and your patch has been +2ed, please ask a committer to run the verification for you.
First, log in to https://jenkins.impala.io. If you are a committer and forgot your password, ask dev@impala.apache.org for help. Someone will reset your password.
Then, if your commit changes only files in the docs/ folder:
If your commit doesn't change any files in the docs/ folder:
If your commit touches files both in and out of the docs/ folder, run https://jenkins.impala.io/job/gerrit-verify-dryrun/build?delay=0sec first, following the directions above. Once it succeeds, run https://jenkins.impala.io/job/gerrit-docs-submit/build?delay=0sec.
Verifying and cherry-picking a patch can be simplified with a bookmarklet. Here is how to do that in Chrome:
By changing the javascript URL, you can also make a bookmarklet for the docs Jenkins job. |
When Jenkins has completed successfully, your change is in the master gerrit repo, but not yet in the official Apache repo. Apache does not allow non-sentient robots like Jenkins to submit patches. To get your patch to the official Apache repo, a committer needs to push the commit to the Apache git repository, by running:
bin/push_to_asf.py |
push_to_asf.py checks the latest commits in gerrit and checks if they are in the Apache git repo. If some are not in the Apache git repo yet, it will ask you if you want to update the Apache git repo with the missing commits found in the Gerrit repo. It does not check what your local state is at all. It only compares remote Gerrit with the remote Apache repo. Keep in mind that if you are a committer, it will allow you to commit any change authored by anyone that has passed the Jenkins cherry-picking described above.
According to project bylaws, branch creation is by lazy consensus of the PMC. We recommend that you email the dev mailing list to let people know about your branch so they understand its purpose.
The steps to create a branch are:
git fetch asf-gerrit git checkout -b the-new-branch asf-gerrit/the-new-branch git push apache HEAD:the-new-branch |
Once the branch is created in both gerrit and the ASF repository, all commits should go to gerrit first, then be pushed to the ASF repository with push_to_asf.py.
Before you send a patch, please make sure that you've read the contribution guidelines.
As described above, Gerrit generates one review request per new commit. It's much easier to review everything related to a patch together in one single commit, so please make sure to squash all your related commits together before sending. If you don't, your review requests will probably get forcibly removed
You might want to bookmark this link, which shows all open reviews for Impala.
git-review is a Python module that can simplify publishing gerrit reviews. To install it use:
sudo pip install git-review |
By default, git-review will look for a remote named 'gerrit'. The usage is:
# Send a code review (or update an existing review). This will also do a git rebase -i asf-gerrit/master. git review -r asf-gerrit |
git-review can be used to apply changes from a review locally and compare patch sets. For a full set of options see: https://pypi.python.org/pypi/git-review
The documentation for Gerrit is here:
http://gerrit.cloudera.org/Documentation/index.html