741315 - [skiplist] Add dvm.*, _JNIEnv.*, JNI_CreateJavaVM.*, missing libc, libmozglue, liblog and libz symbols to the prefix skip list

Status: VERIFIED FIXED

(Socorro :: Infra, task)

Description

[Scoobidiver (away)]

Please add:
* 'dvm.*'
* '_JNIEnv.*'
* 'JNI_GetCreatedJavaVM.*'
* '__memcmp16'
* 'strzcmp16'
* 'strcat'
* 'dlmalloc_trim'
* 'dlmalloc'
* 'huge_dalloc'
* '_files_getaddrinfo'
* 'getanswer'
* 'moz_free'
* 'init_library'
* '__android_log_assert'
remove:
* 'dvmAbort'
* 'dvmStringLen'
* '_JNIEnv::CallObjectMethod'
* 'JNI_CreateJavaVM'

Comment 1

[Robert Kaiser]

Chris, Naoki: Do we need all those on the prefix list?

Scoobidiver: I get a feeling that you are becoming obsessed with adding stuff to the prefix list. ;-)
We should make sure if those are really useful for developers.

Comment 2

[Scoobidiver (away)]

(In reply to Robert Kaiser (:[e-mail]) from comment #1)
> Scoobidiver: I get a feeling that you are becoming obsessed with adding
> stuff to the prefix list. ;-)
Is there any side effect on Socorro (performance...) except gives work to Socorro's team?
> We should make sure if those are really useful for developers.
Those are not Mozilla's libraries. Developers are probably interested by the first Firefox/Core/Graphics component in the stack. Signatures should reflect that.
In addition, how do you know JNI_GetCreatedJavaVMs and AndroidGLController::ProvideEGLSurface are about the same problem? JNI_GetCreatedJavaVMs | AndroidGLController::ProvideEGLSurface is more explicit and allows people who file bugs to group signatures that are related.

Comment 3

[Robert Kaiser]

(In reply to Scoobidiver from comment #2)
> (In reply to Robert Kaiser (:[e-mail]) from comment #1)
> > Scoobidiver: I get a feeling that you are becoming obsessed with adding
> > stuff to the prefix list. ;-)
> Is there any side effect on Socorro (performance...) except gives work to
> Socorro's team?

I'd guess there is an effect on processor performance at least.

> Those are not Mozilla's libraries.

That's no argument for prefixing. If the external code we call is actually the thing that has the bug and is crashing, the signature is accurate with having only the external code on it.

> In addition, how do you know JNI_GetCreatedJavaVMs and
> AndroidGLController::ProvideEGLSurface are about the same problem?

I don't need to. It's up to the developers to find that out and/or tell us that skiplisting would help there.

Comment 4

[Chris Peterson [:cpeterson]]

I'm not familiar with the skiplist syntax, but these function names look like reasonable additions for the skiplist.

Comment 5

[Robert Kaiser]

This list and the one in bug 741222 look so long and scary for me, I wonder if we really need all of those on the skiplist and about the concrete examples we need to add them for.

Comment 6

[Scoobidiver (away)]

I am not scared by a prefix skip list.
Are there metrics to measure processor load performances? What is the impact of adding one signature?

Comment 7

[Scoobidiver (away)]

I added the dependent bugs for the first three signatures in comment 0.

Comment 8

[K Lars Lohn [:lars] [:klohn]]

I am not concerned that signature generation will have any impact on the amount of time that it takes to process a crash.  The processor is IO bound with communicating with Postgres, HBase and minidump_stackwalk (mdsw).  Processing crash on average takes 2.05 seconds. The biggest time expense of processing a crash is running mdsw.  Pre and post processing of a crash (including signature generation) is trivial compared with the IO waits.

The processor is written such that is does as much as it can in parallel with the running of mdsw.  If the crashing thread is early in the output of mdsw, signature generation will happen before mdsw completes.  In that case, skiplist additions will have no impact on the length of time it takes to process a crash.

Comment 9

[Naoki Hirata :nhirata (please use needinfo instead of cc)]

Please add:
* 'dvm.*'
* '_JNIEnv.*'
* 'JNI_GetCreatedJavaVM.*'
* '__memcmp16'
* 'strzcmp16'
* 'strcat'
* 'dlmalloc_trim'
* 'dlmalloc'
* 'huge_dalloc'
* '_files_getaddrinfo'
* 'getanswer'
* 'moz_free'
* 'init_library'
* '__android_log_assert'

remove:
* 'dvmAbort'
* 'dvmStringLen'
* '_JNIEnv::CallObjectMethod'

We still want to keep 'JNI_CreateJavaVM' I believe.

Comment 11

[Laura Thomson :laura]

In https://github.com/lauraxt/socorro/commit/5e952c67b42d141367ad113ed1253e9e43a78744

Comment 12

[[github robot]]

Commit pushed to master at https://github.com/mozilla/socorro

https://github.com/mozilla/socorro/commit/5e952c67b42d141367ad113ed1253e9e43a78744
Fixes bug 741315, Add dvm.*, _JNIEnv.*, JNI_CreateJavaVM.*, missing

Comment 13

[[github robot]]

Commit pushed to stage at https://github.com/mozilla/socorro

https://github.com/mozilla/socorro/commit/e644fe97c73aaf5d8e7bdc329d6c173c83af0010
Fixes bug 741315, Add dvm.*, _JNIEnv.*, JNI_CreateJavaVM.*, missing
libc, libmozglue, liblog and libz symbols to the prefix skip list

Comment 14

[Naoki Hirata :nhirata (please use needinfo instead of cc)]

I did a query... it looks right:
https://crash-stats.mozilla.com/query/query?product=FennecAndroid&version=ALL%3AALL&range_value=1&range_unit=days&date=06%2F06%2F2012+22%3A40%3A23&query_search=signature&query_type=contains&query=&reason=&build_id=&process_type=any&hang_type=any&do_query=1

We probably want to do a preprocess push as well.

Comment 15

[[github robot]]

Commit pushed to https://github.com/mozilla/socorro

https://github.com/mozilla/socorro/commit/e644fe97c73aaf5d8e7bdc329d6c173c83af0010
Fixes bug 741315, Add dvm.*, _JNIEnv.*, JNI_CreateJavaVM.*, missing