Bug 86014
| Summary: | Chromium perf bots fail to download the binary ~50% of the time | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Ojan Vafai <ojan> |
| Component: | Tools / Tests | Assignee: | Nobody <webkit-unassigned> |
| Status: | RESOLVED LATER | ||
| Severity: | Normal | CC: | abarth, morrita, rniwa, tony |
| Priority: | P2 | ||
| Version: | 528+ (Nightly build) | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
| Bug Depends on: | |||
| Bug Blocks: | 77037 | ||
Ojan Vafai
This means we're losing a lot of coverage. Not sure who manages these bots...
http://build.webkit.org/waterfall?show=Chromium%20Win%20Release%20(Perf)&show=Chromium%20Mac%20Release%20(Perf)&show=Chromium%20Linux%20Release%20(Perf)
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Ryosuke Niwa
This is because perf bots' cycle time is significantly longer than that of builders especially when changes are trivial. perf. bots get behind, and hits master's threshold to delete uploaded binaries. I've asked wms to increase the threadshold but it didn't help. The only thing it did was increasing the delay and backlogs.
The only solution to this problem is reducing cycle time of perf bots. Otherwise, the backlog will grow indefinitely.
Tony Chang
Do we know which tests are the slowest? Maybe we can get a better cycle time just by fixing a dozen or so tests.
Ryosuke Niwa
Yeah, most of them are Dromaeo tests. We spend roughly 30% of time running the slowest 5 tests. You can process the stdout on bots to find the ranking although I've posted somewhere on Bugzilla as well.
Ryosuke Niwa
We don't have any workable solution for this at the moment.