This page describes the "master" branches for Launchpad development. Please ask for help immediately if you have any questions. |
Where's trunk?
Launchpad has four master branches (this is unusual; most projects only have one). They are:
master Where most development takes place (except anything that involves changing the database schema; see below).
db-devel Changes that modify the database schema get merged here first.
stable This is fed by regular merges from the master branch, and is deployed to qastaging.launchpad.net frequently.
db-stable This is fed by regular merges from the db-devel branch. It is deployed to staging.launchpad.net frequently, and deployed to launchpad.net on each DB schema deployment. After deployment, db-stable is merged back into master, completing the cycle.
In summary: db-stable is the post-test-run counterpart of db-devel, and stable is the post-test-run counterpart of master, where "post-test-run" means that once code hits foo-devel it is only cleared into foo-stable if the buildbot test runner succeeds.
If you want to know how all of this works behind the scenes, read Trunk/Glue (after reading the rest of this page).
Look at the Pretty Pictures
Diagrams might be the best way to understand this. (If they don't work, there's text afterwards, don't worry.)
Below, we break the process down a bit with a different diagramming approach. Here's what happens when you submit to master:
Where are the expected potential problems in the process? Glad you asked!
It is also possible to submit directly to the db-devel branch.
Let's Try That in Words
Database changes can be destabilizing to other work, so we isolate them out into a separate branch (db-devel). Then there are two arenas for stabilizing changes for deployment: stable (which ends up on qastaging and is fed from the master branch), and db-stable (which ends up on staging and is fed from the db-devel branch).
(Note that the actual images here date from the Bazaar world, and so talk about "devel" rather than "master". You can regard these two names as interchangeable. Feel free to update the images if you have time.)
In summary:
Developers submit to master and db-devel.
Buildbot tests the tip of db-devel, in addition to master. If a db-devel change passes, it is pushed automatically to db-stable.
When changes are approved by buildbot to go from master to stable, a script also generates a PQM request to merge stable with db-devel, sent as if it came from the Launchpad list. The Launchpad list will be informed of merge failures, and Launchpad developers will collectively be responsible for correcting them. (TODO: is this some internal list? Hmmm.)
Staging runs db-stable; qastaging runs stable. We will deploy production DB schema changes from db-stable. (After a deployment, db-stable is manually pushed to devel and stable.)
Problems that can occur
The merge from stable to db-devel fails. That's a real risk but we expect it to occur infrequently. When it does occur, the Launchpad list (TODO: which one?) will be emailed, and it is the collective responsibility of everyone to fix the problem, by manually submitting a branch to db-devel that includes the branch that failed to merge plus any necessary adjustments.
Where to Send Merge Requests
Merge requests that do any of the following should be targeted at the db-devel branch:
adds a cold database patch (see PolicyAndProcess/DatabaseSchemaChangesProcess)
depends on database schema change only in db-devel / db-stable
All other developer merge requests for the launchpad tree should be targeted at the master branch.
The current buildbot-poller.py script on the PQM box monitors buildbot to discover what revisions have passed the tests. When a new revision is "blessed" by buildbot, the script copies it over to stable. The poller script merges the blessed revision into the db-devel branch. It is also responsible for checking a new buildbot build testing db-devel. The poller will copy revisions of db-devel that buildbot blesses to db-stable.
Staging releases should run code from db-stable. qastaging and production should continue to run code from stable.
When a DB schema deployment is made, the db-stable code should be merged into master. This is the responsibility of the person who organised the deployment.
Problems with the automated merge requests are sent to the Launchpad list (TODO: which list?).
When the merge of stable -> db-devel does not work automatically, it should be done manually and submitted. The tests do not need to be run (rationale: the bots don't run the tests when they do it automatically, and the failure messages can clutter up the ML really fast.)
How do I submit to db-devel?
See also WorkingWithDbDevel.
cd ~/canonical/launchpad/launchpad git checkout -b mydbbranch db-devel make build make schema
...hack until ready to submit...
git push -u myusername mydbbranch
Now create and land a merge proposal.
FAQ
Can I land a testfix before buildbot has finished a test run that has failed or will fail?
Yes you can, and please do if appropriate, because this will mean that other developers will not encounter a broken tree at all.
What happens if the automatic merge from stable to db-devel fails?
The (internal! change?) launchpad list gets an email. This is described on the /Trunk page.
A build failed for some reason other than a test failure. What do I do?
Force a build at https://lpbuildbot.canonical.com/force.