672086 - Workers: Recycle threads to prevent churn and stack recursion problems

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Attachment: Patch, v1

Workers currently create their own thread and then destroy it when they complete. If many workers all finish at the same time then multiple thread shutdown events will hit the main thread at the same time. Each of these events spins the event loop and so we can easily run out of stack space. Besides all that there's no reason not to hang on to finished threads and reassign them to new workers if possible.

Attached patch hangs on to a max of 20 finished (idle) threads for 30 seconds. They're reassigned in LIFO order to hopefully keep memory from paging.

Comment 1

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Comment on attachment 546362 [details] [diff] [review]
Patch, v1

Review of attachment 546362 [details] [diff] [review]:
-----------------------------------------------------------------

Comment 2

[Ben Turner (not reading bugmail, use the needinfo flag!)]

http://hg.mozilla.org/integration/mozilla-inbound/rev/e385a9429c3a

Comment 3

[Marco Bonardo [:mak] (Away Apr 25 - May 5)]

http://hg.mozilla.org/mozilla-central/rev/e385a9429c3a

Comment 4

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Any particular reason to not use nsIThreadPool here?

Comment 5

[Ben Turner (not reading bugmail, use the needinfo flag!)]

(In reply to comment #4)
> Any particular reason to not use nsIThreadPool here?

Well, we want to allow a capped number of threads per domain, so we would have to have a threadpool per origin (threadpools will run anything in the queue, no way to filter currently). If we did that then free threads in one domain's pool couldn't be used by another domain that hadn't yet reached its limit since the pool manages all its own threads. The end result would be that it causes us to have more idle threads than we really want.

Is that argument compelling enough?

Comment 6

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Definitely!