system-config/doc/source/gerrit.rst

title

Gerrit

Gerrit

Gerrit is the code review system used by the OpenStack project. For a full description of how the system fits into the OpenStack workflow, see the development workflow guide.

This section describes how Gerrit is configured for use in the OpenStack project and the tools used to manage that configuration.

At a Glance

Hosts

• http://review.openstack.org
• http://review-dev.openstack.org

Puppet

• https://git.openstack.org/cgit/openstack-infra/puppet-gerrit/tree/
• modules/openstack_project/manifests/review.pp
• modules/openstack_project/manifests/review_dev.pp

Configuration

• modules/openstack_project/templates/review.projects.ini.erb
• gerrit/projects.yaml

Projects

• http://code.google.com/p/gerrit/

Bugs

• https://storyboard.openstack.org/#!/project/715
• http://code.google.com/p/gerrit/issues/list

Resources

• Gerrit Documentation

Installation

Gerrit is installed and configured by Puppet, including specifying the exact Java WAR file that is used. See sysadmin for how Puppet is used to manage OpenStack infrastructure systems.

Cinder Volumes

The Gerrit installation at /home/gerrit2 is located on a Cinder volume. See cinder for details on volume management. Note that SSD volumes are used (and they have a minimum size of 100G).

Gerrit Configuration

Most of Gerrit's configuration is in configuration files or Git repositories (and in our case, managed by Puppet), but a few items must be configured in the database. The following is a record of these changes:

Add information about the CLA:

sudo -u root
mysql
use reviewdb;
insert into contributor_agreements values (
'Y', 'Y', 'Y', 'ICLA',
'OpenStack Individual Contributor License Agreement',
'static/cla.html', 2);

Groups

A number of system-wide groups are configured in Gerrit (rather than via Puppet). When installing a new Gerrit, you should create these by hand (and capture their UUID - you will need them to setup the ACLs later).

The Project Bootstrappers group grants all the permissions needed to set up a new project. Normally, the OpenStack Project Creater account is the only member of this group, but members of the Administrators group may temporarily add themselves in order to correct problems with automatic project creation.

The Third-Party CI group is used to grant +/-1 Verified access to external testing tools on a sandbox project.

The Voting Third-Party CI group is used to grant +/-1 Verified access to external testing tools for all projects.

The Continuous Integration Tools group contains Jenkins and any other CI tools that get +2/-2 access on reviews.

The Release Managers group is used for release managers.

Users

The first user to log in becomes an administrator. Be sure to set an account name and add ssh keys - you'll need those.

Once you've created your groups you should create the openstack-project-creator account by hand (the account name is referenced from modules/openstack_project/templates/review.projects.ini.erb) using:

cat $pubkey | ssh -p 29418 $USER@$HOST gerrit create-account \
  --group "'Project Bootstrappers'" \
  --group Administrators \
  --full-name "'Project Creator'" \
  --email openstack-infra@lists.openstack.org \
  --ssh-key - openstack-project-creator

GitHub Integration

Gerrit replicates to GitHub by pushing to a standard Git remote. The GitHub projects are configured to allow only the Gerrit user to push.

Pull requests can not be disabled for a project in Github, so instead we have a script that runs from cron to close any open pull requests with instructions to use Gerrit.

These are both handled automatically by jeepyb.

Note that the user running Gerrit will need to accept the GitHub host keys. e.g.:

sudo su - gerrit2
ssh github.com

Troubleshooting

When creating a new project, there can be times where the jeepyb automation to create the GitHub project can fail, and leave the project improperly configured. This can cause replication to GitHub to fail. The project in GitHub will be created, but will appear empty. When trying replication from Gerrit, it will show a Permission denied error when trying to push content. To solve that, following steps are needed:

1. Login into github.com, using openstack-project-creator user.

#. Navigate to the failed repository, and enter on Settings > Collaborators & teams option.

1. Add Gerrit as Team member to that project.

After the team has been added, project will start replicating successfully to GitHub.

Auto Review Expiry

Puppet automatically installs a daily cron job called expire-old-reviews onto the Gerrit servers. This script follows two rules:

1. If the review hasn't been touched in 2 weeks, mark as abandoned.
2. If there is a negative review and it hasn't been touched in 1 week, mark as abandoned.

If your review gets touched by either of these rules, it is possible to unabandon a review on the Gerrit web interface.

This process is managed by the jeepyb openstack-infra project.

Gerrit IRC Bot

Gerritbot consumes the Gerrit event stream and announces relevant events on IRC. gerritbot is an openstack-infra project and is also available on Pypi.

Launchpad Bug Integration

In addition to the hyperlinks provided by the regex in gerrit.config, we use a Gerrit hook to update Launchpad bugs when changes referencing them are applied. This is managed by the jeepyb openstack-infra project.

Storyboard Integration

We use the Gerrit its-storyboard plugin to update storyboard stories and tasks when changes referencing them are applied.

New Project Creation

Gerrit project creation is now managed through changes to the openstack-infra/project-config repository. jeepyb handles automatically creating any new projects defined in the configuration files.

Local Git Replica

Gerrit replicates all repos to a local directory so that Apache can serve the anonymous http requests out directly. This is automatically configured by jeepyb.

Access Controls

High level goals:

1. Anonymous users can read all projects.
2. All registered users can perform informational code review (+/-1) on any project.
3. Jenkins can perform verification (blocking or approving: +/-1).
4. All registered users can create changes.
5. The OpenStack Release Manager and Jenkins can tag releases (push annotated tags).
6. Members of $PROJECT-core group can perform full code review (blocking or approving: +/- 2), and submit changes to be merged.
7. Members of Release Managers (Release Manager and delegates), and $PROJECT-milestone (PTL and release minded people) exclusively can perform full code review (blocking or approving: +/- 2), and submit changes to be merged on pre-release stable/* branches.
8. Full code review (+/- 2) of API projects (documentation of the API, not implementation of the API) should be available to the -core group of the corresponding implementation project as well as to the OpenStack Documentation Coordinators.
9. Full code review of stable branches should be available to the -stable-maint group of the project.
10. Drivers (PTL and delegates) of client library projects should be able to add tags (which are automatically used to trigger releases).

To manage API project permissions collectively across projects, API projects are reparented to the "API-Projects" meta-project instead of "All-Projects". This causes them to inherit permissions from the API-Projects project (which, in turn, inherits from All-Projects).

The global Gerrit permissions set out the high level goals (and manage-projects can then override this on a per project basis as needed). To setup the global permissions, first create the groups covered above under Groups.

You need to grant yourself enough access to replace the ACLs over ssh (we use SSH because it's fast, and it gets syntax checked).

1. Visit https://$HOST/#/admin/projects/All-Projects,access and click on Edit.
2. Look for the reference to 'refs/meta/config', click on the drop-box for 'add permission' and choose 'PUSH'.
3. Type in Administrators as the group name
4. Click on Add
5. Click on Save Changes

Then... we need to fetch the All-Projects ACLs, update them, then push the updates back into Gerrit:

export USER=$your_gerrit_user
export HOST=$your_gerrit_host
cd $anywhereyoulike
mkdir All-Projects-ACLs
cd All-Projects-ACLs
git init
git remote add gerrit ssh://$USER@$HOST:29418/All-Projects.git
git fetch gerrit +refs/meta/*:refs/remotes/gerrit-meta/*
git checkout -b config remotes/gerrit-meta/config

There will be two interesting files, groups and project.config. groups contains UUIDs and names of groups that will be referenced in project.config. UUIDs can be found on the group page in Gerrit. Next, edit project.config to look like:

[access "refs/*"]
create = group Project Bootstrappers
create = group Release Managers
forgeAuthor = group Registered Users
forgeCommitter = group Project Bootstrappers
push = +force group Project Bootstrappers
pushMerge = group Project Bootstrappers
pushSignedTag = group Project Bootstrappers
pushTag = group Continuous Integration Tools
pushTag = group Project Bootstrappers
pushTag = group Release Managers
read = group Anonymous Users
editTopicName = group Registered Users

[access "refs/drafts/*"]
push = block group Registered Users

[access "refs/for/refs/*"]
push = group Registered Users

[access "refs/for/refs/zuul/*"]
pushMerge = group Continuous Integration Tools

[access "refs/heads/*"]
label-Code-Review = -2..+2 group Project Bootstrappers
label-Code-Review = -1..+1 group Registered Users
label-Verified = -2..+2 group Continuous Integration Tools
label-Verified = -2..+2 group Project Bootstrappers
label-Verified = -1..+1 group Voting Third-Party CI
label-Workflow = -1..+0 group Change Owner
label-Workflow = -1..+1 group Project Bootstrappers
rebase = group Registered Users
submit = group Continuous Integration Tools
submit = group Project Bootstrappers

[access "refs/meta/config"]
read = group Project Owners

[access "refs/meta/openstack/*"]
create = group Continuous Integration Tools
push = group Continuous Integration Tools
read = group Continuous Integration Tools

[access "refs/zuul/*"]
create = group Continuous Integration Tools
push = +force group Continuous Integration Tools
pushMerge = group Continuous Integration Tools

[capability]
accessDatabase = group Administrators
administrateServer = group Administrators
createProject = group Project Bootstrappers
emailReviewers = deny group Third-Party CI
priority = batch group Non-Interactive Users
runAs = group Project Bootstrappers
streamEvents = group Registered Users

[contributor-agreement "ICLA"]
accepted = group CLA Accepted - ICLA
agreementUrl = static/cla.html
autoVerify = group CLA Accepted - ICLA
description = OpenStack Individual Contributor License Agreement
requireContactInformation = true

[contributor-agreement "System CLA"]
accepted = group System CLA
agreementUrl = static/system-cla.html
description = DON'T SIGN THIS: System CLA (externally managed)

[contributor-agreement "USG CLA"]
accepted = group USG CLA
agreementUrl = static/usg-cla.html
description = DON'T SIGN THIS: U.S. Government CLA (externally managed)

[label "Code-Review"]
abbreviation = R
copyAllScoresOnTrivialRebase = true
copyMinScore = true
function = MaxWithBlock
value = -2 Do not merge
value = -1 This patch needs further work before it can be merged
value = 0 No score
value = +1 Looks good to me, but someone else must approve
value = +2 Looks good to me (core reviewer)

[label "Verified"]
function = MaxWithBlock
value = -2 Fails
value = -1 Doesn't seem to work
value = 0 No score
value = +1 Works for me
value = +2 Verified

[label "Workflow"]
function = MaxWithBlock
value = -1 Work in progress
value = 0 Ready for reviews
value = +1 Approved

[plugin "its-storyboard"]
enabled = true

[project]
description = Rights inherited by all other projects

Now edit the groups file. The format is:

#UUID  Group Name
1234567890123456789012345678901234567890  group-foo

Each of the groups listed above under 'Groups' should have an entry as well as the built in groups such as 'Non-Interactive Users' which may or may not be present in the initial groups file. You can find the UUID values by navigating to Admin -> Groups -> Group Name -> General in the Web UI.

Finally, commit the changes and push the config back up to Gerrit:

git commit -am "Initial All-Projects config"
git push gerrit HEAD:refs/meta/config

Manual Administrative Tasks

The following sections describe tasks that individuals with root access may need to perform on rare occasions.

Renaming a Project

Renaming a project is not automated and is disruptive to developers, so it should be avoided. Allow for an hour of downtime for the project in question, and about 10 minutes of downtime for all of Gerrit. All Gerrit changes, merged and open, will carry over, so in-progress changes do not need to be merged before the move.

Note that some of the steps in the process below are repetitive and so for larger batches a script can be used to generate the command lists for upload to and execution on their respective servers:

#!/bin/sh
#
# Expects a renames.list file in the current directory with:
#
#     stackforge/foo -> openstack/foo
#     openstack/oldbar -> openstack/newbar

echo "\nGerrit database updates\n-----------------------"
for r in `sed 's/ -> /@/' renames.list` ; do
    OLD=`echo $r | cut -d@ -f1`
    NEW=`echo $r | cut -d@ -f2`
    echo "update account_project_watches set project_name = \"$NEW\" where
        project_name = \"$OLD\";"
    echo "update changes set dest_project_name = \"$NEW\",
        created_on = created_on where dest_project_name = \"$OLD\";"
done

echo "\nGerrit filesystem updates\n-------------------------"
for r in `sed 's/ -> /@/' renames.list` ; do
    OLD=`echo $r | cut -d@ -f1` ; NEW=`echo $r | cut -d@ -f2`
    echo "sudo mv ~gerrit2/review_site/git/{$OLD,$NEW}.git"
    echo "sudo mv /opt/lib/git/{$OLD,$NEW}.git"
done

echo "\nGit farm filesystem updates\n---------------------------"
for r in `sed 's/ -> /@/' renames.list` ; do
    OLD=`echo $r | cut -d@ -f1`
    NEW=`echo $r | cut -d@ -f2`
    echo "sudo mv /var/lib/git/{$OLD,$NEW}.git"
done

echo "\nJenkins workspace cleanup\n-------------------------"
for r in `sed 's/ -> /@/' renames.list` ; do
NAME=`echo $r | cut -d@ -f1 | cut -d/ -f2`
echo "sudo ansible-playbook -f 10 \\
    /etc/ansible/playbooks/clean_workspaces.yaml \\
    --extra-vars \"project=$NAME\""
done

To rename a project:

1. Prepare a change to the project-config repo to update things like projects.yaml/ACLs, jenkins-job-builder and gerritbot for the new name. Also add changes to update projects.txt in all branches of the requirements repo, devstack-vm-gate-wrap.sh in the devstack-gate repo, reference/projects.yaml in the openstack/governance repo, and .gitmodules in the openstack/openstack repo if necessary.

2. Stop puppet runs on the puppetmaster to prevent early application of configuration changes:

   sudo crontab -u root -e

   Comment out the crontab entries. Use ps to make sure that a run is not currently in progress. When it finishes, make sure the entry has not been added back to the crontab.

3. Prepare a yaml file called repos.yaml that has a single dictionary called repos with a list of dictionaries each having an old and new entry. Optionally also add a gerrit_groups dict of the same form:

   repos:
   - old: stackforge/awesome-repo
     new: openstack/awesome-repo
   gerrit_groups:
   - old: old-core-group
     new: new-core-group

4. Run the ansible rename repos playbook, passing in the path to your yaml file:

   sudo ansible-playbook -f 10 /opt/system-config/production/playbooks/rename_repos.yaml -e repolist=ABSOLUTE_PATH_TO_VARS_FILE

5. Merge the prepared Puppet configuration change, removing the original Jenkins jobs via the Jenkins WebUI later if needed.

6. Re-enable puppet runs on the puppetmaster:

   sudo crontab -u root -e

7. Submit a change that updates .gitreview with the new location of the project.

Developers will either need to re-clone a new copy of the repository, or manually update their remotes with something like:

git remote set-url origin https://git.openstack.org/$ORG/$PROJECT

Third-Party Testing Access

The command to add an account for an automated system which gets -1/+1 code verify voting rights (as outlined in third-party-testing) looks like:

ssh -p 29418 review.openstack.org "gerrit create-account \
    --group 'Third-Party CI' \
    --full-name 'Some CI Bot' \
    --email ci-bot@third-party.org \
    --ssh-key 'ssh-rsa AAAAB3Nz...zaUCse1P ci-bot@third-party.org' \
    some-ci-bot"

Details on the create-account command can be found in the Gerrit API documentation.

Resetting a Username in Gerrit

Initially if a Gerrit username (which is used to associate SSH connections to an account) has not yet been set, the user can type it into the Gerrit WebUI... but there is no supported way for the user to alter or correct it once entered. Further, if a defunct account has the desired username, a different one will have to be entered.

Because of this, often due to the user ending up with Duplicate Accounts in Gerrit, it may be requested to change the SSH username of an account. Confirm the account_id number for the account in question and remove the existing username external_id for that (it may also be necessary to remove any lingering external_id with the desired username if confirmed there is a defunct account associated with it):

delete from account_external_ids where account_id=NNNN and external_id like 'username:%';

After this, the user should be able to re-add their username through the Gerrit WebUI.

Duplicate Accounts in Gerrit

From time to time, outside events affecting SSO authentication or identity changes can result in multiple Gerrit accounts for the same user. This frequently causes duplication of preferred E-mail addresses, which also renders the accounts unselectable in some parts of the WebUI (notably when trying to add reviewers to a change or members in a group). Gerrit does not provide a supported mechanism for Combining Gerrit Accounts, and doing so manually is both time-consuming and error prone. As a result, the OpenStack infrastructure team does not combine duplicate accounts for users but can clean up these E-mail address issues upon request. To find the offending duplicates:

select account_id from accounts where preferred_email='user@example.com';

Find out from the user which account_id is the one they're currently using, and then null out the others with:

update accounts set preferred_email=NULL, registered_on=registered_on where account_id=OLD;

Then flush Gerrit's caches so any immediate account lookups will hit the current DB contents:

ssh review.openstack.org -p29418 gerrit flush-caches --all

Combining Gerrit Accounts

While not supported by Gerrit, a fairly thorough account merge is documented here (mostly as a demonstration of its unfortunate complexity). Please note that the OpenStack infrastructure team does not combine duplicate accounts for users upon request, but this would be the process to follow if it becomes necessary under some extraordinary circumstance.

Collect as much information as possible about all affected accounts, and then go poking around in the tables listed below for additional ones to determine the account_id number for the current account and any former accounts which should be merged into it. Then for each old account_id, perform these update and delete queries:

delete from account_agreements where account_id=OLD;
delete from account_diff_preferences where id=OLD;
delete from account_external_ids where account_id=OLD;
delete from account_group_members where account_id=OLD;
delete from account_group_members_audit where account_id=OLD;
delete from account_project_watches where account_id=OLD;
delete from account_ssh_keys where account_id=OLD;
delete from accounts where account_id=OLD;
update account_patch_reviews set account_id=NEW where account_id=OLD;
update starred_changes set account_id=NEW where account_id=OLD;
update change_messages set author_id=NEW, written_on=written_on where author_id=OLD;
update changes set owner_account_id=NEW, created_on=created_on where owner_account_id=OLD;
update patch_comments set author_id=NEW, written_on=written_on where author_id=OLD;
update patch_sets set uploader_account_id=NEW, created_on=created_on where uploader_account_id=OLD;
update patch_set_approvals set account_id=NEW, granted=granted where account_id=OLD;

If that last update query results in a collision with an error like:

ERROR 1062 (23000): Duplicate entry 'XXX-YY-NEW' for key 'PRIMARY'

Then you can manually delete the old approval:

delete from patch_set_approvals where account_id=OLD and change_id=XXX and patch_set_id=YY;

And repeat until the update query runs to completion.

After all the described deletes and updates have been applied, flush Gerrit's caches so things like authentication will be rechecked against the current DB contents:

ssh review.openstack.org -p29418 gerrit flush-caches --all

Make the user aware that these steps have also removed any group memberships, preferences, SSH keys, contact information, CLA signatures, and so on associated with the old account so some of these may still need to be added to the new one via the Gerrit WebUI if they haven't been already. With a careful inspection of all accounts involved it is possible to merge some information from the old accounts into new ones by performing update queries similar to the deletes above, but since this varies on a case-by-case basis it's left as an exercise for the reader.

Deleting a User from Gerrit

This isn't normally necessary, but if you find that you need to completely delete an account from Gerrit, perform the same delete queries mentioned in Combining Gerrit Accounts and replace the update queries for account_patch_reviews and starred_changes with:

delete from account_patch_reviews where account_id=OLD;
delete from starred_changes where account_id=OLD;

The other update queries can be ignored, since deleting them in many cases would result in loss of legitimate review history.

Refreshing HTML and CSS configuration

When there is a change in HTML headers, or CSS, this can be applied without the need of restarting Gerrit. To do that, ssh in the Gerrit instance, and touch GerritSiteHeader.html and/or GerritSite.css, under /home/gerrit2/review_site/etc directory.

Deactivating a Gerrit account

To deactivate a Gerrit account (use case can be a failing Third Party CI), you must follow that steps:

1. Identify the account ID of the Third Party CI you need to deactivate. Third-Party CI members can be found on: https://review.openstack.org/#/admin/groups/270,members

   That will give you the name and email of all members. Then you can get the matching numerical account ID with the help of REST API: curl -i -H "Accept: application/json" --digest --user <<gerrit_user>>:<<http_pass>> -X GET https://review.openstack.org/a/accounts/{email}

   This will return a JSON dictionary, that will contain _account_id field.

2. Mark the account as inactive using gerrit ssh api, with: ssh -p 29418 review.openstack.org gerrit set-account --inactive {account-id}

   Alternatively you can use REST API, sending a DELETE for: curl -i -H "Accept: application/json" --digest --user <<gerrit_user>>:<<http_pass>> -X DELETE https://review.openstack.org/a/accounts/{account-id}/active

3. Check if there are active gerrit ssh connections: ssh -p 29418 review.openstack.org gerrit show-connections -n | grep {account-id}

   And kill all of them with subsequent: ssh -p 29418 review.openstack.org gerrit close-connection {connection-id}

4. You can check if the account is properly marked as inactive using REST API, sending a GET for:

   curl -i -H "Accept: application/json" --digest --user <<gerrit_user>>:<<http_pass>> -X GET https://review.openstack.org/a/accounts/{account-id}/active

   A 200 return code means the account is active, and 204 means account inactive.

5. In the case of a failing Third Party CI, if the account caused a loop of comments in a change, you can delete them with following query: delete from change_messages where author_id={account-id} and change_id={change-id};