868521 - Don't create a new preallocated process so soon after a process is started

Status: RESOLVED FIXED

(Firefox OS Graveyard :: General, defect)

Description

[Justin Lebar (not reading bugmail)]

I don't know if I regressed this or we were always so bad, but I observe that right after we receive a call, almost the first thing we do is launch a new preallocated process!

That's quite bad; we're just wasting CPU time.

This is easy enough to fix, I think.

Comment 1

[Justin Lebar (not reading bugmail)]

Attachment: Patch, v1

Comment 2

[Dave Hylands [:dhylands]]

Hmm. So when I originally coded this, the first idle didn't occur until several seconds after launching a new app.

Maybe the dialer is different, in that it actually blocks and waits for something right away.

What we discovered with other apps, was that the timeout (which was 5 seconds I think) was occuring before the app had reached idle, so the preallocated process was slowing down the app initialization.

Perhaps we need something like first idle after some delay.

Comment 3

[Justin Lebar (not reading bugmail)]

> Maybe the dialer is different, in that it actually blocks and waits for something right away.

Or perhaps idle happens sooner when we're under CPU stress.

> Perhaps we need something like first idle after some delay.

This is basically delay-after-first-idle instead of first-idle-after-delay.  I hope it's close enough; it's certainly much easier to do.

Comment 4

[Dave Hylands [:dhylands]]

(In reply to Justin Lebar [:jlebar] from comment #3)
> > Maybe the dialer is different, in that it actually blocks and waits for something right away.
> 
> Or perhaps idle happens sooner when we're under CPU stress.

IIRC The idle is that the child is idle, which means that it has no messages to process. This typically would happen when it's waiting for async I/O to complete (or the user to press a mouse or something).

Maybe the idle stuff has regressed from moving stuff off the main thread to other threads, and the idle isn't "good enough" any more.

Comment 5

[Justin Lebar (not reading bugmail)]

> IIRC The idle is that the child is idle, which means that it has no messages to process. 

Right, but the child could become idle if, for example, it's not being pumped as quickly with messages from the main process.  Which could happen under certain types of stress.

Comment 6

[Dave Hylands [:dhylands]]

Comment on attachment 747223 [details] [diff] [review]
Patch, v1

Review of attachment 747223 [details] [diff] [review]:
-----------------------------------------------------------------

This seems like a reasonable compromise for the time being.

Comment 7

[Justin Lebar (not reading bugmail)]

https://hg.mozilla.org/projects/birch/rev/74dbf4035f07

Comment 8

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/74dbf4035f07

Comment 9

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-b2g18/rev/d1b25094d1fd
https://hg.mozilla.org/releases/mozilla-b2g18_v1_0_1/rev/072e642ad2b7

Comment 10

[Cody Roesch [:croesch]]

Can you please provide steps to verify this fix - as we will blackbox test from the UI?

Comment 11

[Deepa Subramanian]

Can you please provide steps to verify this fix - as we will blackbox test from the UI?