Merge lp:~julian-edwards/launchpad/double-build-bug-705342 into lp:launchpad
Status: | Merged | ||||
---|---|---|---|---|---|
Approved by: | Julian Edwards | ||||
Approved revision: | no longer in the source branch. | ||||
Merged at revision: | 12322 | ||||
Proposed branch: | lp:~julian-edwards/launchpad/double-build-bug-705342 | ||||
Merge into: | lp:launchpad | ||||
Prerequisite: | lp:~jml/launchpad/failing-tests-bug-711209 | ||||
Diff against target: |
96 lines (+43/-5) 4 files modified
lib/canonical/widgets/__init__.py (+12/-0) lib/lp/buildmaster/model/builder.py (+10/-0) lib/lp/buildmaster/tests/mock_slaves.py (+2/-2) lib/lp/buildmaster/tests/test_builder.py (+19/-3) |
||||
To merge this branch: | bzr merge lp:~julian-edwards/launchpad/double-build-bug-705342 | ||||
Related bugs: |
|
Reviewer | Review Type | Date Requested | Status |
---|---|---|---|
Leonard Richardson (community) | Approve | ||
Review via email: mp+48219@code.launchpad.net |
Commit message
[r=leonardr]
Description of the change
The attached bug describes how some build logs end up with two apparently separate builds in them. This happens when a buildd-admin kills the builder's processes off but doesn't kill everything quite properly for whatever reason. Normally the buildd-manager will handle this fine for virtual builders, as when it sees an ABORTED state it resets the builder.
However for nonvirtual builders we just try and clean the slave which is clamitous due to a bug in sbuild which won't kill the actual build process off, thus leaving the builder open to receiving a second build.
This change just fails the (nonvirtual) builder if it gets into this state, since there's nothing we can do on the manager side.