Provide a way to disable CQ retries |
||
Issue descriptionA bit of background: A bad roll of the Fuchsia SDK autoroller made it through the CQ because the tests ending up passing on retry: https://ci.chromium.org/p/chromium/builders/luci.chromium.try/fuchsia_x64/154584 This ended up being a bad roll, causing lots of flakiness across all Fuchsia test suites. We would like a way to prevent this from happening again. Our idea would be to disable CQ retries on Fuchsia bots for the Fuchsia autoroller CLs. I am not familiar enough with Infra to know how we could best meet that goal, feel free to redirect or reformulate as you wish.
,
Nov 28
I think it's also reasonable to entertain possibility of annotating individual builders in cq.cfg with retry directives if e.g., fuchsia_arm64 builder is always required. Something like this:
..
builder {
name: "fuchsia_arm64"
retry_policy {
failures: 0
infra_failures: 1
}
}
,
Nov 28
I like the idea of #c2. IIUC, we are moving away from dynamically adding footers for CQ.
,
Nov 28
I suspect we'll still want a way to tweak the CQ on a per-CL basis, though (e.g., for autorollers). I'm not sure what our plan is for that beyond gerrit footers. I know we replaced PRESUBMIT's addition of Cq-Include-Trybots; if there are plans to expand that to what the autorollers do, it's news to me.
,
Nov 28
There had been offline discussions about having CQ somehow understand DEPS.py and trigger different trybots depending on which file line gets changed.
However, given path_regexp feature, there is an easy but ugly workaround:
* autorollers modifies not just DEPS file, but also some other whitespace-like file, say autoroll-fuchsia-trigger-bots.whitespace.
* cq.cfg contains:
builder {
name: "fuchsia"
path_regexp: "autoroll-fuchsia-trigger-bots.whitespace"
}
,
Nov 28
That indeed is ugly. |
||
►
Sign in to add a comment |
||
Comment 1 by jbudorick@chromium.org
, Nov 27Components: -Infra Infra>Platform>CQ Infra>Client>Chrome