No need to set generic Infra components when it has a clear subcomponent already.

I've looked at recent logs of slow /tasks/new, and they all have unexplainable stalls, like this one: https://screenshot.googleplex.com/bzGNOfw9UQR.png (a request comes in, does nothing for 30 sec, then hits the log line at the very start of the handler (before any expensive operations; what did it do for 30 sec?).

This looks very similar to https://bugs.chromium.org/p/chromium/issues/detail?id=704907#c38

I hope the work Marc-Antoine does in https://bugs.chromium.org/p/chromium/issues/detail?id=704236 will reduce load on our GAE instances, and this random stalls will disappear.

Could you please investigate this further? Random 30 second stalls seems like a *bad* thing and shouldn't be happening no matter the load.

If you think it is a problem in AppEngine we should talk to AppEngine SRE and see if they can help us diagnose the problem.

This was before M-A's scheduler optimizations, please reopen if it's still happening.

estaab@ I don't think this has anything to do with the scheduler? 

vadimsh@ said "recent logs of slow /tasks/new and they all have unexplainable stalls". This feels like an issue in request handling or something configured wrong on appengine?

I really think this still needs investigating.

The swarming scheduler (before M-A optimizations) was huge CPU hog. GAE instances were CPU-throttled and it caused all request to be slow.

Spot checking /tasks/new logs now shows they are consistently <2s (usually under 500 ms). Monarch metrics seem to agree: http://shortn/_s6kYemX4Zv (though they tend to lie, since they don't include time spent in GAE routing guts, that get affected by CPU overloads too).

Issue has not been modified or commented on in the last 365 days, please re-open or file a new bug if this is still an issue.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot