Nigel Daley wrote:
So far I haven't had any failures. Perhaps we should wait for more experience? Currently, the system will email me after 2 failed build/test attempts on the same patch. I could send these failure emails to hadoop-dev if you'd prefer. But realize that these emails may occasionally be false negatives.
Ideally we'd get a '-1' comment if the patch doesn't apply to trunk, no? This would be annoying if there were lots of false negatives. But it could save folks time, when, e.g., a patch gets stale, if it's noted in the patch.
Should we re-run tests on patches daily, and if something that previously passed now fails or vice versa, note the change in state?
Perhaps there's a way of making this automation more public by running it (and the associated continuous integration server) on apache hosted servers?
The appropriate server for this would be lucene.zones.apache.org, a Solaris zone (virtual solaris box). If you'd like to set these up publicly that'd be great. Alternately, we could do something like add attachments to issues with the log file from the first failing unit test, or somesuch.
Doug
