Hi,
I did a tiny amount of hacking on hubbot and wanted to share the state.
It now runs permanently on files.cockpit-project.org and works on new and updated pull requests. (But it doesn't use github events yet...)
it has a whitelist of users and wont touch pull requests created by people not in that list.
It will add statuses to the commits it is working on, right next to Travis.
The link in the status goes to a directory with the checked out Cockpit sources. There is a "hubbot.log" file in there that contains the ongoing output of the test run.
Other interesting files are "mock/root.log" and "mock/build.log".
Hubbot rebases the head of a pull request onto master before running VERIFY.
However, it doesn't recognize when master has changed since the last run, and there is currently no way check what master was when hubbot did the rebase. So, some tiny amount of hacking left... :)
The integration tests are of course still brittle. If they fail spuriously, hubbot will not retry automatically. You can manually reset the hubbot status of a commit to "error" and then hubbot will retry the next time it looks at the pull request. (Ask me how to do that if you need it.)
For those with accounts on files.cp.o: check hubbot.service.
On 21.01.2015 12:38, Marius Vollmer wrote:
Hi,
I did a tiny amount of hacking on hubbot and wanted to share the state.
It now runs permanently on files.cockpit-project.org and works on new and updated pull requests. (But it doesn't use github events yet...)
it has a whitelist of users and wont touch pull requests created by people not in that list.
It will add statuses to the commits it is working on, right next to Travis.
This is awesome :)
The integration tests are of course still brittle. If they fail spuriously, hubbot will not retry automatically. You can manually reset the hubbot status of a commit to "error" and then hubbot will retry the next time it looks at the pull request. (Ask me how to do that if you need it.)
Will do.
Stef
Marius Vollmer marius.vollmer@redhat.com writes:
However, it doesn't recognize when master has changed since the last run, and there is currently no way check what master was when hubbot did the rebase. So, some tiny amount of hacking left... :)
I did this now, and a new "hubbot.master" file should appear in the checkout directies with the SHA of the master at the time of rebasing. (Yep, low tech. I should have used MongoDB, but...)
Also, hubbot now runs with TEST_JOBS=4 and a run takes about 10 minutes instead of the previous 30 minutes.
Hubbot is slowly working down the open pull requests, because they don't have the "hubbot.master" file yet.
cockpit-devel@lists.fedorahosted.org