500007 - try server OSX builds should include symbols for Shark

Status: RESOLVED FIXED

(Release Engineering :: General, defect, P2)

Description

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

When these builds are created to test performance-related patches, it's helpful to be able to profile them.  Can we include the same set of symbols that we do in the shark builds? (Function symbols only)

Comment 1

[Chris Cooper [:coop] (he/him)]

Vlad: is this a dupe of bug 495030? Armen's going to do some work in that bug today to try to generate symbols in a fashion similar to nightlies/releases.

Comment 2

[(not currently active) Ted Mielczarek]

This would be as simple as adding "ac_add_options --disable-install-strip" to the mozconfig:
http://hg.mozilla.org/build/buildbot-configs/file/3bf03608f457/mozilla2/macosx/mozilla-central/shark/mozconfig#l19
(that instructs the packager not to strip out debug symbols).

Comment 3

[Chris Cooper [:coop] (he/him)]

Running a new mozconfig through the try server now.

Comment 4

[Nick Thomas [:nthomas] (UTC+12)]

We'd should check if tests like Ts are affected at all. If so, someone would have to make a call about the utility of talos comparisons vs full symbols.

Comment 5

[Chris Cooper [:coop] (he/him)]

Attachment: Disable install strip for Mac try server builds

Vlad: can you check the build at http://build.mozilla.org/tryserver-builds/[e-mail]-1245951407 for symbols? It was generated with this patched mozconfig.

Comment 6

[Chris Cooper [:coop] (he/him)]

I made another run (http://build.mozilla.org/tryserver-builds/[e-mail]-1249419341) and verified that the symbols are present in shark.

As for ts, looking at some comparative graphs:

try run, tiger: 1052.89
m-c talos tiger (3, avg): ~1010, ~1010, ~1125  

http://graphs-stage.mozilla.org/graph.html#tests=[{%22test%22:%2212%22,%22branch%22:%226%22,%22machine%22:%22126%22},{%22test%22:%2212%22,%22branch%22:%221%22,%22machine%22:%22278%22},{%22test%22:%2212%22,%22branch%22:%221%22,%22machine%22:%22277%22},{%22test%22:%2212%22,%22branch%22:%221%22,%22machine%22:%22276%22}]

try run leopard (2): 1375.32, 1390.79
m-c leopard (avg): ~1535

http://graphs-stage.mozilla.org/graph.html#tests=[{%22test%22:%2212%22,%22branch%22:%226%22,%22machine%22:%22238%22},{%22test%22:%2212%22,%22branch%22:%226%22,%22machine%22:%22237%22},{%22test%22:%2212%22,%22branch%22:%221%22,%22machine%22:%22141%22}]

Comment 7

[Chris Cooper [:coop] (he/him)]

Bugzilla isn't marking up the links well, so I'll try again:

Tiger graph:
http://bit.ly/xl9yA

Leopard graph:
http://bit.ly/eW5cy

Comment 8

[Chris Cooper [:coop] (he/him)]

Comment on attachment 385199 [details] [diff] [review]
Disable install strip for Mac try server builds

Ts doesn't seem unduly affected IMO.

Comment 9

[Nick Thomas [:nthomas] (UTC+12)]

Comment on attachment 385199 [details] [diff] [review]
Disable install strip for Mac try server builds

> Ts doesn't seem unduly affected IMO.

Agreed. There's a Tp3 RSS spike - about 30M or 1% on Leopard, and something similar on Tiger, but that doesn't seem like the end of the world. Do you know why the mac unit test build bombed out ?

Comment 10

[Chris Cooper [:coop] (he/him)]

(In reply to comment #9) 
> Agreed. There's a Tp3 RSS spike - about 30M or 1% on Leopard, and something
> similar on Tiger, but that doesn't seem like the end of the world. Do you know
> why the mac unit test build bombed out ?

Huh, I thought I had killed all the other builds I didn't care about. I'll run it again before pushing and make sure it's not something dangerous.

Comment 11

[Chris Cooper [:coop] (he/him)]

(In reply to comment #9)
> Agreed. There's a Tp3 RSS spike - about 30M or 1% on Leopard, and something
> similar on Tiger, but that doesn't seem like the end of the world. Do you know
> why the mac unit test build bombed out ?

The issue here is that unittests use their own configs. I provided a new mozconfig through the interface that had --disable-tests specified (the non-unittest default), but the new mozconfig gets used by all slaves regardless of platform or test payload.

Supplying a proper unittest config with the stripping disabled works fine, i.e. shows same warnings as current trunk: http://tinderbox.mozilla.org/showlog.cgi?log=MozillaTry/1249653693.1249661307.17858.gz&fulltext=1

Going to carry-forward the review to land the strip change for the unittest config as well, if that's ok.

Comment 12

[Chris Cooper [:coop] (he/him)]

Comment on attachment 385199 [details] [diff] [review]
Disable install strip for Mac try server builds

http://hg.mozilla.org/build/buildbot-configs/rev/b9921f77d54b

Comment 13

[Chris Cooper [:coop] (he/him)]

Updated the buildbot-configs clones on the tryserver masters.