1051051 - b2g hits >70MB in 24 hours

Status: RESOLVED FIXED

(Firefox OS Graveyard :: Stability, defect, P1)

Description

[Doug Turner (:dougt)]

This bug should focus on the js-non-window leak.  The fd leak was fixed in 1048906.


+++ This bug was initially created as a clone of Bug #1049806 +++

There is not test steps  for this. But please note that we are running stability test on device for more than 24 hours. 

Observation:
1) b2g is using 430 FDs and it has opened 202 FD with /data/local/webapps/marketplace.firefox.com/application.zip . This points to FD leak in b2g process
2) about-memory-45 (collected at device timestamp [1]) shows that b2g is allocating 
     23.14 MB (100.0%) -- explicit
     ├──16.58 MB (71.63%) -- js-non-window
     │  ├──15.77 MB (68.13%) -- zones/zone(0xNNN)
     │  │  ├───7.25 MB (31.34%) ── unused-gc-things [3]
     │  │  ├───6.79 MB (29.35%) -- compartment([System Principal], inProcessTabChildGlobal?ownedBy=chrome://b2g/content/shell.html)



[1] b2g memory growth device timestamp 2014-08-06 11:04:14 . You can try to b2g memory usage using this timestamp in all attached log files.
[2] FULL LOG LOCATION: https://drive.google.com/file/d/0B1cSMS8_GuAERVJlVE1OM1dvc28/edit?usp=sharing . It has DMD logs, gc/cc logs, about-memory* reports, lsof logs etc.


I also attached a diff between about-memory-0 and about-memory-45 as attachment to this bug.

Comment 1

[Andrew McCreight [:mccr8]]

I'll take a look at the logs.

Comment 2

[Doug Turner (:dougt)]

Andrew McCreight is going to look at this.  He is awesome.

Comment 3

[Andrew McCreight [:mccr8]]

One thing I noticed in the log is that we have 7000 copies of the SpecialPowers object.  That's probably bug 1048833, which landed on our 2.0 yesterday, after the memory bug was filed.

Comment 4

[Andrew McCreight [:mccr8]]

This is the frequency of various functions in the GC heap:
    7254 Function onInnerWindowDestroyed
    7254 Function SpecialPowers/<.get
     505 Function firstScan
     462 Function toString
     414 Function setEventHandler
     409 Function onReceivedMessage
     330 Function get

The first two are from leaking Special Powers objects.

firstScan and onReceievedMessage look like they could be separate issues, so I'll file new bugs.

Comment 5

[Tapas[:tkundu on #b2g/gaia/memshrink/gfx] (always NI me)]

[Blocking Requested - why for this release]:

Comment 6

[Andrew McCreight [:mccr8]]

I only see 3 ContentParent in the CC log, so I think we aren't leaking those in this test.

Comment 7

[Inder]

Tapas will have an update here by COB today (8/13).

Comment 8

[bhavana bajaj [:bajaj]]

Inder did mention he was still seeing a small leak here, adding the NI to get more info from Inder/tapas.

Comment 9

[Tapas[:tkundu on #b2g/gaia/memshrink/gfx] (always NI me)]

This seems to be fixed now. Thanks for your help :)