On 12-09-28 10:03 PM, Seth Fowler wrote:
On Sep 28, 2012, at 6:28 PM, Boris Zbarsky <bzbar...@mit.edu> wrote:
The point is that the patches that _do_ need to land urgently are blocked on
lack of resources because people are wasting too many cycles with unnecesary
try pushes.
This sounds like a notion of priority might be helpful.
- Seth
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
From what I can see, there are actually two use cases for try (correct
me if I'm wrong):
1. A patch that is expected to succeed, but you want to run it through
try to verify.
2. A patch that is less certain, so you want to run it through try to
shotgun-identify issues to address.
For "optimistic" pushes, we expect that the patch goes from green =>
green. For "pessimistic" pushes we expect a listing of all oranges that
arise from that push.
The first case can be handled differently than the latter. Trivial
pushes can be tried in batch: with multiple patches tried at once (as
long as they apply cleanly on top of each other), with a high
expectation of success. If there's an unexpected failure (which should
be rare), it can be bisected quickly by repeating only the failed tests
in a binary search to identify the patch that introduced them. If there
are a lot of failures in the tests against the batched patches, then we
fall back to running them all through try individually.
The approach could free up significant compute resources for the second
case.
-kannan
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform