“backgroundhangmonitor” ping

Whenever a thread monitored by the Background Hang Monitor hangs, a stack and some non-identifying information about the hang is captured. When 50 of these hangs are collected across all processes, or the browser exits, this ping is transmitted with the collected hang information.

This ping is only collected in nightly builds, to avoid the high volume of pings which would be produced in Beta.

Structure:

{
  "type": "backgroundhangmonitor",
  ... // common ping data
  "environment": { ... },
  "payload": {
    "timeSinceLastPing": <number>, // uptime since last backgroundhangmonitor ping (ms).
    "modules": [
      [
        <string>, // Name of the file holding the debug information.
        <string> // Breakpad ID of this module.
      ],
      ...
    ],
    "hangs": [
      {
        "duration": <number>, // duration of the hang in milliseconds.
        "thread": <string>, // name of the hanging thread.
        "runnableName": <string>, // name of the runnable executing during the hang.
                                  // Runnable names are only collected for the XPCOM main thread.
        "process": <string>, // Type of process that hung, see below for a list of types.
        "remoteType": <string>, // Remote type of process which hung, see below.
        "annotations": [ ... ], // A list of annotations on the hang, see below.
        "pseudoStack": [ ... ], // List of label stack frames and js frames.
        "stack": [ ... ], // interleaved hang stack, see below.
      },
      ...
    ]
  }
}

Note

Hangs are collected whenever the current runnable takes over 128ms.

Process Types

The process field is a string denoting the kind of process that hung. Hangs are currently sent only for the processes below:

Kind

Description

default

Main process, also known as the browser process

tab

Content process

gpu

GPU process

Remote Type

The remoteType field is a string denoting the type of content process that hung. As such it is only non-null if processType contains the tab value.

The supported remoteType values are documented in the crash ping documentation: Remote Process Types.

Stack Traces

Each hang object contains a stack field which has been populated with an interleaved stack trace of the hung thread. An interleaved stack consists of a native backtrace with additional frames interleaved, representing chrome JS and label stack entries.

The structure that manages the label stack and the JS stack was called “PseudoStack” in the past and is now called “ProfilingStack”.

Note that this field only contains native stack frames, label stack and chrome JS script frames. If native stacks can not be collected on the target platform, or stackwalking was not initialized, there will be no native frames present, and the stack will consist only of label stack and chrome JS script frames.

A single frame in the stack is either a raw string, representing a label stack or chrome JS script frame, or a native stack frame.

Label stack frames contain the static string from all insances of the AUTO_PROFILER_LABEL* macros. Additionally, dynamic strings are collected from all usages of the AUTO_PROFILER_LABEL_DYNAMIC*_NONSENSITIVE macros. The dynamic strings are simply appended to the static strings after a space character.

Current dynamic string collections are as follows:

Static string

Dynamic

ChromeUtils::Import

Associated chrome:// or resource:// URI

nsJSContext::GarbageCollectNow

GC reason string

mozJSSubScriptLoader::DoLoadSubScriptWithOptions

Associated chrome:// or resource:// URI

PresShell::DoFlushPendingNotifications

Flush type

nsObserverService::NotifyObservers

Associated observer topic

Native stack frames are as such:

[
  <number>, // Index in the payload.modules list of the module description.
            // -1 if this frame was not in a valid module.
  <string> // Hex string (e.g. "FF0F") of the frame offset in the module.
]

Annotations

The annotations field is an array of key-value pairs, for example if the user was interacting during a hang the annotations field would look something like this:

[
    [
        "UserInteracting",
        "true"
    ]
]

The following annotations are currently present in tree:

Name

Description

UserInteracting

“true” if the user was actively interacting

pluginName

Name of the currently running plugin

pluginVersion

Version of the currently running plugin

HangUIShown

“true” if the hang UI was shown

HangUIContinued

“true” if continue was selected in the hang UI

HangUIDontShow

“true” if the hang UI was not shown

Unrecovered

“true” if the hang persisted until process exit

Additional annotations can be added at run-time via User Interactions.