New issue
Advanced search Search tips

Issue 904935 link

Starred by 1 user

Issue metadata

Status: Assigned
Owner:
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Chrome
Pri: 1
Type: Feature

Blocking:
issue 904881


Show other hotlists

Hotlists containing this issue:
CrOSParallelCQ


Sign in to add a comment

ScheduleSlaves

Project Member Reported by cindyb@google.com, Nov 13

Issue description

What is says
 
Status: Available (was: Untriaged)
Cc: dgarr...@chromium.org jclinton@chromium.org
Owner: athilenius@chromium.org
Jason and Don can you confirm my assumption:

Not applicable; Chrome CQ will launch 'children' builds in all cases. In the PostSubmit case, an 'orchestration' build will be launched along with the builders that just spin-waits until they are done.

The only reason this assumption would be wrong is if we must do something in PostSubmit before builders are started (which I think the annealing builder takes care of).
Chrome CQ deals only with PreSubmit case. A separate system is used for Post-Submit builds, luci-scheduler.appspot.com.

LUCI Scheduler will schedule only the orchestration build which will schedule actual builds as children and wait. Scheduler will wait for the orchestration build. The reason why LUCI Scheduler won't be used to schedule actual builders, is that we want all of them to checkout the same revision.

Note that annealing builder and orchestration builder are decoupled because the former spins fast, and the latter waits for its children.
Status: Assigned (was: Available)
Nodir, that should be fine. The Annealing builder will be pushing commits to an 'annealing-manifest' (names TBD) which LUCI CQ can use as a trigger. That manifest is a full pin.
If we plan to have recipe based master builders, we'll need this stage. We probably will for a while.

That applies to release builders, full builders, incrementals, postsubmit, etc.
Cc: tandrii@chromium.org
Yea, even if we adopt the Browser CQ to run the CQ post-submit builder as proposed in the email thread, I'm not sure how we would implement release builders (and others) if this stage didn't exist. That seems like expanding the remit of Browser CQ quite a lot. But maybe that's in the cards, Andrii? (Feel free to reply in the email thread instead.)

This should also be a pretty simple stage to implement.

Parent builder generates / publishes a buildspec.

Parent builder schedules children from it's build config with that buildspec.

Since the new sync script supports explicit buildspec definitions, that seems straight forward. A helper script to publish buildspec's would also be needed (this is currently done by sync stages, but that's conceptually wrong).

Sign in to add a comment