1261108 - Cannot measure BLOCKED_ON_PLUGIN_MODULE_INIT_MS for e10s experiment participants in Beta46noapz

Status: RESOLVED INCOMPLETE

(Core Graveyard :: Plug-ins, defect, P1)

Description

[Chris H-C :chutten|PTO (back Oct 23)]

There are only about 12 parent-process reports of BLOCKED_ON_PLUGIN_MODULE_INIT_MS from the Beta46noapz experiment (bug 1253669)

I don't know why.

Comment 1

[Chris H-C :chutten|PTO (back Oct 23)]

Here's my analysis showing the dearth of responses: https://gist.github.com/chutten/275109b6da95cda0a5aed577e7a4fb29

I wonder if this is just because MODULE_INIT_MS didn't make it in until beta4 which was targeted at March 21[0], and the experiment ended March 23 around midnight, Eastern[1].

[0]: bug 1256943 comment 7
[1]: "endTime: 1458788400" is Wednesday, March 23, 2016 11:00:00 PM GMT-4:00 DST (from bug 1253669)

Comment 2

[Benjamin Smedberg]

Chris can you try this in a debugger? You should be hitting http://mxr.mozilla.org/mozilla-central/source/dom/plugins/ipc/PluginModuleParent.cpp#2665 the first time you use Flash in a process (and later, if you stop using for a few minutes and then come back). I bet we're skipping one of the checks somewhere.

Comment 3

[(No longer employed by Mozilla) Aaron Klotz]

For the e10s case he actually needs
http://mxr.mozilla.org/mozilla-central/source/dom/plugins/ipc/PluginModuleParent.cpp#131

FWIW I can easily view this histogram by going to a flash site, then checking for its accumulation in about:telemetry.

Comment 4

[Chris H-C :chutten|PTO (back Oct 23)]

So... that a breakpoint set in plugin-container for NPP_NewInternal hits for me with e10s enabled is a bad sign?

Comment 5

[(No longer employed by Mozilla) Aaron Klotz]

Remember that this histogram exists for both chrome and content; both breakpoints will be hit in different contexts.

The NPP_NewInternal will be hit when logging that probe for blocking *content*.

When we want to know how much time is spent blocking *chrome*, you need to break at the location that I specified, which is the fix I added in bug 1256943.

Comment 6

[Chris H-C :chutten|PTO (back Oct 23)]

Spiffy. Works for me.

I see no evidence that the collection is flawed.

I can only surmise that Beta users in the experiment didn't update to builds containing the MODULE_INIT-reporting patch until too late for us to get a sample bigger than 12.