+jbudorick for ideas

This seemingly started happening after build revision 743742be5e3ce451a4c5f1970e531a5a97a2f3ee (Let android/builder handle dynamic test target compilation)

it was that. i can address this (and explain it in more detail) later today.

Ok great! I think we'll also have to make sure to go back after this is fixed and purge all those bad builds, so that bisects don't keep grabbing them.

Should the bisect be building everything?

I think the perf builders build everything, Dave can answer that better.

The bisect builders do it on demand when specific revision combos are needed (ie. chromium@X, v8@Y)

Upping priority since this pretty much breaks any bisect descending into rolls.

RE #5: not sure what your asking with "should the bisect be building everything?":

* Do you mean should it rebuild each time? No, it's much faster (and despite this bug generally more reliable) to reuse builds from buildbucket.
* Are you asking when should it use bisect builders vs perf waterfall builders? We build every chromium revision on perf, but we need to build each v8 revision for v8 rolls, and we do that on the bisect builders.

Sorry about the delay here.

#10: It was previously not specifying targets and thus was building all targets. After my change, it started trying to build only chromium_builder_perf, which is empty on Android (https://codesearch.chromium.org/chromium/src/BUILD.gn?rcl=0&l=981). This is in contrast with the Android builders on chromium.perf, which build a specific list of targets. Currently, that's:
 - android_tools
 - chrome_modern_public_apk
 - chrome_public_apk
 - dump_syms
 - microdump_stackwalk
 - monochrome_public_apk
 - push_apps_to_background_apk
 - system_webview_apk
 - system_webview_shell_apk
 - cc_perftests
 - chromium_builder_perf
 - gpu_perftests
 - tracing_perftests

(There are multiple mechanisms by which these are specified; that was the subject of my prior CL.)

Should the bisect bots be building:
 - all targets, as they were before
 - only what they need
 - the same targets as those built by the chromium.perf bots

dtu, simonhatch: can you clarify what they should be building?

re: #c18

Same targets as the Android Builders on chromium.perf.

fix: https://chromium-review.googlesource.com/c/419773/

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/tools/build.git/+/5cb489a5b30cefd55d9d5c980b44bc18272c7a26

commit 5cb489a5b30cefd55d9d5c980b44bc18272c7a26
Author: John Budorick <jbudorick@google.com>
Date: Thu Dec 15 19:12:04 2016

[Android] Specify targets for the android perf bisect builders.

The specified targets match their chromium.perf counterparts.

BUG= 672529 

Change-Id: I27ba34561efa6c88bcc123e52c9ac6da32c3e340
Reviewed-on: https://chromium-review.googlesource.com/419773
Reviewed-by: Simon Hatch <simonhatch@chromium.org>
Reviewed-by: David Tu <dtu@chromium.org>
Commit-Queue: John Budorick <jbudorick@chromium.org>

[modify] https://crrev.com/5cb489a5b30cefd55d9d5c980b44bc18272c7a26/scripts/slave/recipes/android/builder.expected/full_tryserver_chromium_perf_android_arm64_perf_bisect_builder.json
[modify] https://crrev.com/5cb489a5b30cefd55d9d5c980b44bc18272c7a26/scripts/slave/recipes/android/builder.expected/full_tryserver_chromium_perf_android_perf_bisect_builder.json
[modify] https://crrev.com/5cb489a5b30cefd55d9d5c980b44bc18272c7a26/scripts/slave/recipes/android/builder.py

Awesome, thanks John! Next step after we confirm the builds are working is to purge all the broken builds.

https://build.chromium.org/p/tryserver.chromium.perf/builders/android_arm64_perf_bisect_builder/builds/2053

Looks like these are working again! I can go delete the bad builds and close this up.