1145007 - Add "resident-peak" memory measurement on Unix

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Nicholas Nethercote [inactive]]

This has been suggested several times recently. It can be read from the "VmHWM" entry in /proc/<pid>/status.

Comment 1

[Nicholas Nethercote [inactive]]

Attachment: (part 1) - Test |residentUnique| in test_memoryReporters.xul

It is the only distinguished amount in nsIMemoryReporterManager that isn't
being tested.

Comment 2

[Nicholas Nethercote [inactive]]

Attachment: (part 2) - Add a "resident-peak" distinguished amount and memory reporter on Linux

Comment 3

[Mike Hommey [:glandium] (PTO. Back on Nov 25)]

Doesn't getrusage give the same value in ru_maxrss?

Comment 4

[Nicholas Nethercote [inactive]]

(In reply to Mike Hommey [:glandium] from comment #3)
> Doesn't getrusage give the same value in ru_maxrss?

Oh, I didn't know about getrusage. It gives the same result and is easier to use, so I'll switch.

Comment 5

[Nicholas Nethercote [inactive]]

Attachment: (part 2) - Add a "resident-peak" distinguished amount and memory reporter on Linux

Now using getrusage().

Comment 6

[Mike Hommey [:glandium] (PTO. Back on Nov 25)]

Note that getrusage is POSIX, you can also use it on OSX as well (or even #ifdef XP_UNIX, likely)

Comment 7

[Eric Rahm [:erahm]]

Comment on attachment 8579805 [details] [diff] [review]
(part 1) - Test |residentUnique| in test_memoryReporters.xul

Review of attachment 8579805 [details] [diff] [review]:
-----------------------------------------------------------------

r=me

Comment 8

[Eric Rahm [:erahm]]

Comment on attachment 8579838 [details] [diff] [review]
(part 2) - Add a "resident-peak" distinguished amount and memory reporter on Linux

Review of attachment 8579838 [details] [diff] [review]:
-----------------------------------------------------------------

lgtm. It would be nice to have this supported on OSX/BSD as well. Can you file a follow up for that?

Comment 9

[Nicholas Nethercote [inactive]]

> It would be nice to have this supported on OSX/BSD as well. Can you
> file a follow up for that?

I'll just do it in this patch, by moving the peak stuff in nsMemoryReporterManager.cpp into a XP_UNIX block.

Comment 10

[Nicholas Nethercote [inactive]]

Attachment: (part 2) - Add a "resident-peak" distinguished amount and memory reporter on Unix

ResidentPeakDistinguishedAmount() is now in the XP_UNIX section, and I'm
handling the different units for different Unices, and I'm handling the
possibility of 0, in case Solaris returns that.

Comment 11

[Eric Rahm [:erahm]]

Comment on attachment 8580322 [details] [diff] [review]
(part 2) - Add a "resident-peak" distinguished amount and memory reporter on Unix

Review of attachment 8580322 [details] [diff] [review]:
-----------------------------------------------------------------

r=me, one super minor suggestion.

::: xpcom/base/nsMemoryReporterManager.cpp
@@ +810,5 @@
> +ResidentPeakDistinguishedAmount(int64_t* aN)
> +{
> +  struct rusage usage;
> +  if (0 == getrusage(RUSAGE_SELF, &usage)) {
> +    // The units for ru_maxrrs:

nit: ru_maxrss

@@ +1923,5 @@
> +
> +/*static*/ int64_t
> +nsMemoryReporterManager::ResidentPeak()
> +{
> +#ifdef HAVE_RESIDENT_PEAK_REPORTER

min--: This could just call GetResidentPeak and drop the #ifdefs.

Comment 12

[Nicholas Nethercote [inactive]]

> > +/*static*/ int64_t
> > +nsMemoryReporterManager::ResidentPeak()
> > +{
> > +#ifdef HAVE_RESIDENT_PEAK_REPORTER
> 
> min--: This could just call GetResidentPeak and drop the #ifdefs.

I thought that too, but this one is static and GetResidentPeak() is not.

Comment 13

[Nicholas Nethercote [inactive]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/228183a22063
https://hg.mozilla.org/integration/mozilla-inbound/rev/3ecc82272861

Comment 14

[Phil Ringnalda (:philor)]

https://hg.mozilla.org/mozilla-central/rev/228183a22063
https://hg.mozilla.org/mozilla-central/rev/3ecc82272861