“main” ping

This is the “main” Telemetry ping type, whose payload contains most of the measurements that are used to track the performance and health of Firefox in the wild. It includes the histograms and other performance and diagnostic data.

This ping is triggered by different scenarios, which is documented by the reason field:

  • aborted-session - this ping is regularly saved to disk (every 5 minutes), overwriting itself, and deleted at shutdown. If a previous aborted session ping is found at startup, it gets sent to the server. The first aborted-session ping is generated as soon as Telemetry starts
  • environment-change - the Environment changed, so the session measurements got reset and a new subsession starts
  • shutdown - triggered when the browser session ends. For the first browsing session, this ping is saved to disk and sent on the next browser restart. From the second browsing session on, this ping is sent immediately on shutdown using the Ping Sender, unless the OS is shutting down
  • daily - a session split triggered in 24h hour intervals at local midnight. If an environment-change ping is generated by the time it should be sent, the daily ping is rescheduled for the next midnight
  • saved-session - the “classic” Telemetry payload with measurements covering the whole browser session (only submitted for a transition period)

Most reasons lead to a session split, initiating a new subsession. We reset important measurements for those subsessions.

After a new subsession split, the internal-telemetry-after-subsession-split topic is notified to all the observers. This is an internal topic and is only meant for internal Telemetry usage.

Note

saved-session is sent with a different ping type (saved-session, not main), but otherwise has the same format as discussed here.

Structure:

{
  version: 4,

  info: {
    reason: <string>, // what triggered this ping: "saved-session", "environment-change", "shutdown", ...
    revision: <string>, // the Histograms.json revision
    timezoneOffset: <integer>, // time-zone offset from UTC, in minutes, for the current locale
    previousBuildId: <string>, // null if this is the first run, or the previous build ID is unknown

    sessionId: <uuid>,  // random session id, shared by subsessions
    subsessionId: <uuid>,  // random subsession id
    previousSessionId: <uuid>, // session id of the previous session, null on first run.
    previousSubsessionId: <uuid>, // subsession id of the previous subsession (even if it was in a different session),
                                  // null on first run.

    subsessionCounter: <unsigned integer>, // the running no. of this subsession since the start of the browser session
    profileSubsessionCounter: <unsigned integer>, // the running no. of all subsessions for the whole profile life time

    sessionStartDate: <ISO date>, // hourly precision, ISO date in local time
    subsessionStartDate: <ISO date>, // hourly precision, ISO date in local time
    sessionLength: <integer>, // the session length until now in seconds, monotonic
    subsessionLength: <integer>, // the subsession length in seconds, monotonic

    flashVersion: <string>, // obsolete, use ``environment.addons.activePlugins``
    addons: <string>, // obsolete, use ``environment.addons``
  },

  processes: {...},
  childPayloads: [...], // only present with e10s; reduced payloads from content processes, null on failure
  simpleMeasurements: {...},

  // The following properties may all be null if we fail to collect them.
  histograms: {...},
  keyedHistograms: {...},
  chromeHangs: {...},
  threadHangStats: [...], // obsolete in firefox 57, use the 'bhr' ping
  capturedStacks: {...},
  log: [...],
  webrtc: {...},
  gc: {...},
  fileIOReports: {...},
  lateWrites: {...},
  addonDetails: {...},
  UIMeasurements: [...],  // Android only
  slowSQL: {...},
  slowSQLstartup: {...},
}

info

sessionLength

The length of the current session so far in seconds. This uses a monotonic clock, so this may mismatch with other measurements that are not monotonic like calculations based on Date.now().

If the monotonic clock failed, this will be -1.

Note that this currently does not behave consistently over our supported platforms:

  • On Windows this uses GetTickCount64(), which does increase over sleep periods
  • On macOS this uses mach_absolute_time(), which does not increase over sleep periods
  • On POSIX/Linux this uses clock_gettime(CLOCK_MONOTONIC, &ts), which should not increase over sleep time

See bug 1204823 for details.

subsessionLength

The length of this subsession in seconds. This uses a monotonic clock, so this may mismatch with other measurements that are not monotonic (e.g. based on Date.now()).

If sessionLength is -1, the monotonic clock is not working.

Also see the remarks for sessionLength on platform consistency.

processes

This section contains per-process data.

Structure:

"processes" : {
  // ... other processes ...
  "parent": {
    scalars: {...},
    keyedScalars: {...},
    events: {...},
  },
  "content": {
    scalars: {...},
    keyedScalars: {...},
    histograms: {...},
    keyedHistograms: {...},
    events: {...},
  },
  "gpu": {
    // ...
  }
}

histograms and keyedHistograms

This section contains histograms and keyed histograms accumulated on content processes. Histograms recorded on a content child process have different character than parent histograms. For instance, GC_MS will be much different in processes.content as it has to contend with web content, whereas the instance in payload.histograms has only to contend with browser JS. Also, some histograms may be absent if never recorded on a content child process (EVENTLOOP_UI_ACTIVITY is parent-process-only).

This format was adopted in Firefox 51 via bug 1218576.

scalars and keyedScalars

This section contains the Scalars that are valid for the current platform. Scalars are only submitted if data was added to them, and are only reported with subsession pings. The recorded scalars are described in the Scalars.yaml file. The info.revision field indicates the revision of the file that describes the reported scalars.

events

This section contains the Events that are recorded for the current subsession. Events are not always recorded, recording has to be enabled first for the Firefox session.

The recorded events are defined in the Events.yaml. The info.revision field indicates the revision of the file that describes the reported events.

childPayloads

The Telemetry payloads sent by child processes, recorded on child process shutdown (event content-child-shutdown observed). They are reduced session payloads, only available with e10s. Among some other things, they don’t contain histograms, keyed histograms, add-on details, or UI Telemetry.

Note: Child payloads are not collected and cleared with subsession splits, they are currently only meaningful when analysed from saved-session or main pings with reason set to shutdown.

Note: Before Firefox 51 and bug 1218576, content process histograms and keyedHistograms were in the individual child payloads instead of being aggregated into processes.content.

simpleMeasurements

This section contains a list of simple measurements, or counters. In addition to the ones highlighted below, Telemetry timestamps (see here and here) can be reported.

totalTime

A non-monotonic integer representing the number of seconds the session has been alive.

uptime

A non-monotonic integer representing the number of minutes the session has been alive.

addonManager

Only available in the extended set of measures, it contains a set of counters related to Addons. See here for a list of recorded measures.

UITelemetry

Only available in the extended set of measures. For more see UITelemetry data format.

startupInterrupted

A boolean set to true if startup was interrupted by an interactive prompt.

js

This section contains a series of counters from the JavaScript engine.

Structure:

"js" : {
  "setProto": <unsigned integer>, // Number of times __proto__ is set
}

maximalNumberOfConcurrentThreads

An integer representing the highest number of threads encountered so far during the session.

startupSessionRestoreReadBytes

Windows-only integer representing the number of bytes read by the main process up until the session store has finished restoring the windows.

startupSessionRestoreWriteBytes

Windows-only integer representing the number of bytes written by the main process up until the session store has finished restoring the windows.

startupWindowVisibleReadBytes

Windows-only integer representing the number of bytes read by the main process up until after a XUL window is made visible.

startupWindowVisibleWriteBytes

Windows-only integer representing the number of bytes written by the main process up until after a XUL window is made visible.

debuggerAttached

A boolean set to true if a debugger is attached to the main process.

shutdownDuration

The time, in milliseconds, it took to complete the last shutdown.

failedProfileLockCount

The number of times the system failed to lock the user profile.

savedPings

Integer count of the number of pings that need to be sent.

activeTicks

Integer count of the number of five-second intervals (‘ticks’) the user was considered ‘active’ (sending UI events to the window). An extra event is fired immediately when the user becomes active after being inactive. This is for some mouse and gamepad events, and all touch, keyboard, wheel, and pointer events (see EventStateManager.cpp). This measure might be useful to give a trend of how much a user actually interacts with the browser when compared to overall session duration. It does not take into account whether or not the window has focus or is in the foreground. Just if it is receiving these interaction events. Note that in main pings, this measure is reset on subsession splits, while in saved-session pings it covers the whole browser session.

pingsOverdue

Integer count of pending pings that are overdue.

histograms

This section contains the histograms that are valid for the current platform. Flag and count histograms are always created and submitted, with their default value being respectively false and 0. Other histogram types (see Choosing a Histogram Type) are not created nor submitted if no data was added to them. The type and format of the reported histograms is described by the Histograms.json file. Its most recent version is available here. The info.revision field indicates the revision of the file that describes the reported histograms.

keyedHistograms

This section contains the keyed histograms available for the current platform.

As of Firefox 48, this section does not contain empty keyed histograms anymore.

threadHangStats

As of Firefox 57 this section is no longer present, and has been replaced with the ‘bhr’ ping.

Contains the statistics about the hangs in main and background threads. Note that hangs in this section capture the C++ pseudostack and an incomplete JS stack, which is not 100% precise. For particularly egregious hangs, and on nightly, an unsymbolicated native stack is also captured. The amount of time that is considered “egregious” is different from thread to thread, and is set when the BackgroundHangMonitor is constructed for that thread. In general though, hangs from 5 - 10 seconds are generally considered egregious. Shorter hangs (1 - 2s) are considered egregious for other threads (the compositor thread, and the hang monitor that is only enabled during tab switch).

To avoid submitting overly large payloads, some limits are applied:

  • Identical, adjacent “(chrome script)” or “(content script)” stack entries are collapsed together. If a stack is reduced, the “(reduced stack)” frame marker is added as the oldest frame.
  • The depth of the reported pseudostacks is limited to 11 entries. This value represents the 99.9th percentile of the thread hangs stack depths reported by Telemetry.
  • The native stacks are limited to a depth of 25 stack frames.

Structure:

"threadHangStats" : [
  {
    "name" : "Gecko",
    "activity" : {...}, // a time histogram of all task run times
    "nativeStacks": { // captured for all hangs on nightly, or egregious hangs on beta
      "memoryMap": [
        ["wgdi32.pdb", "08A541B5942242BDB4AEABD8C87E4CFF2"],
        ["igd10iumd32.pdb", "D36DEBF2E78149B5BE1856B772F1C3991"],
        // ... other entries in the format ["module name", "breakpad identifier"] ...
      ],
      "stacks": [
        [
          [
            0, // the module index or -1 for invalid module indices
            190649 // the offset of this program counter in its module or an absolute pc
          ],
          [1, 2540075],
          // ... other frames ...
        ],
        // ... other stacks ...
      ]
    },
    "hangs" : [
      {
        "stack" : [
          "Startup::XRE_Main",
          "Timer::Fire",
          "(content script)",
          "IPDL::PPluginScriptableObject::SendGetChildProperty",
          ... up to 11 frames ...
        ],
        "nativeStack": 0, // index into nativeStacks.stacks array
        "histogram" : {...}, // the time histogram of the hang times
        "annotations" : [
          {
            "pluginName" : "Shockwave Flash",
            "pluginVersion" : "18.0.0.209"
          },
          ... other annotations ...
        ]
      },
    ],
  },
  ... other threads ...
]

capturedStacks

Contains information about stacks captured on demand via Telemetry API. For more information see stack capture.

This is similar to chromeHangs, but only Precise C++ stacks on the main thread of the parent process are reported. This data is only available on Windows, either in Firefox Nightly or in builds using --enable-profiling switch.

Limits for captured stacks are the same as for chromeHangs (see below). Furthermore:

  • the key length is limited to 50 characters,
  • keys are restricted to alpha-numeric characters and -.

The module names can contain unicode characters.

Structure:

"capturedStacks" : {
  "memoryMap": [
    ["wgdi32.pdb", "08A541B5942242BDB4AEABD8C87E4CFF2"],
    ["igd10iumd32.pdb", "D36DEBF2E78149B5BE1856B772F1C3991"],
    // ... other entries in the format ["module name", "breakpad identifier"] ...
  ],
  "stacks": [
    [
       [
         0, // the module index or -1 for invalid module indices
         190649 // the offset of this program counter in its module or an absolute pc
       ],
       [1, 2540075],
       // ... other frames ...
    ],
    // ... other stacks ...
  ],
  "captures": [["string-key", stack-index, count], ... ]
}

chromeHangs

Contains the statistics about the hangs happening exclusively on the main thread of the parent process. Precise C++ stacks are reported. This is only available on Nightly Release on Windows, when building using “–enable-profiling” switch.

Some limits are applied:

  • Reported chrome hang stacks are limited in depth to 50 entries.
  • The maximum number of reported stacks is 50.

The module names can contain unicode characters.

Structure:

"chromeHangs" : {
  "memoryMap" : [
    ["wgdi32.pdb", "08A541B5942242BDB4AEABD8C87E4CFF2"],
    ["igd10iumd32.pdb", "D36DEBF2E78149B5BE1856B772F1C3991"],
    ... other entries in the format ["module name", "breakpad identifier"] ...
   ],
  "stacks" : [
    [
      [
        0, // the module index or -1 for invalid module indices
        190649 // the offset of this program counter in its module or an absolute pc
      ],
      [1, 2540075],
      ... other frames, up to 50 ...
     ],
     ... other stacks, up to 50 ...
  ],
  "durations" : [8, ...], // the hang durations (in seconds)
  "systemUptime" : [692, ...], // the system uptime (in minutes) at the time of the hang
  "firefoxUptime" : [672, ...], // the Firefox uptime (in minutes) at the time of the hang
  "annotations" : [
    [
      [0, ...], // the indices of the related hangs
      {
        "pluginName" : "Shockwave Flash",
        "pluginVersion" : "18.0.0.209",
        ... other annotations as key:value pairs ...
      }
    ],
    ...
  ]
},

log

This section contains a log of important or unusual events reported through Telemetry.

Structure:

"log": [
  [
    "Event_ID",
    3785, // the timestamp (in milliseconds) for the log entry
    ... other data ...
  ],
  ...
]

At present there is one known users of this section: Telemetry Experiments.

Telemetry Experiments uses it to note when experiments are activated and terminated.

webrtc

Contains special statistics gathered by WebRTC related components.

So far only a bitmask for the ICE candidate type present in a successful or failed WebRTC connection is getting reported through C++ code as IceCandidatesStats, because the required bitmask is too big to be represented in a regular enum histogram.

Note: in most cases the webrtc dictionary inside of IceCandidatesStats will simply be empty as the user has not used any WebRTC PeerConnection at all during the ping report time.

Structure:

"webrtc": {
  "IceCandidatesStats": {
    "webrtc": {
      "34526345": {
        "successCount": 5
      },
      "2354353": {
        "failureCount": 1
      }
    },
  }
},

gc

Contains statistics about selected garbage collections. To avoid bloating the ping, only a few GCs are included. There are two selection strategies. We always save the two GCs with the worst max_pause time. Additionally, in content processes, two collections are selected at random. If a GC runs for C milliseconds and the total time for all GCs since the session began is T milliseconds, then the GC has a C/T probablility of being selected for one of these “slots”.

Structure:

"gc": {
  "random": [
    {
      // Timestamps are in milliseconds since startup. All the times here
      // are wall-clock times, which may not be monotonically increasing.
      "timestamp": 294872.2,
      // All durations are in milliseconds.
      "max_pause": 73.629,
      "total_time": 364.951, // Sum of all slice times.
      "zones_collected": 9,
      "total_zones": 9,
      "total_compartments": 309,
      "minor_gcs": 44,
      "store_buffer_overflows": 19,
      "mmu_20ms": 0,
      "mmu_50ms": 0,
      // Reasons include "None", "NonIncrementalRequested",
      // "AbortRequested", "KeepAtomsSet", "IncrementalDisabled",
      // "ModeChange", "MallocBytesTrigger", "GCBytesTrigger",
      // "ZoneChange", "CompartmentRevived".
      "nonincremental_reason": "None",
      "allocated": 37, // In megabytes.
      "added_chunks": 54,
      "removed_chunks": 12,
      // Total number of slices (some of which may not appear
      // in the "slices" array).
      "num_slices": 15,
      // We record at most 4 slices.
      "slices": [
        {
          "slice": 0,  // The index of this slice.
          "pause": 23.221,  // How long the slice took.
          "when": 0,  // Milliseconds since the start of the GC.
          "reason": "SET_NEW_DOCUMENT",
          // GC state when the slice started
          "initial_state": "NotActive",
          // GC state when the slice ended
          "final_state": "Mark",
          // Budget is either "Xms", "work(Y)", or
          // "unlimited".
          "budget": "10ms",
          // Number of page faults during the slice.
          "page_faults": 0,
          "start_timestamp": 294875,
          "end_timestamp": 294879,
          // Time taken by each phase. There are at most 65 possible
          // phases, but usually only a few phases run in a given slice.
          "times": {
            "wait_background_thread": 0.012,
            "mark_discard_code": 2.845,
            "purge": 0.723,
            "mark": 9.831,
            "mark_roots": 0.102,
            "buffer_gray_roots": 3.095,
            "mark_cross_compartment_wrappers": 0.039,
            "mark_c_and_js_stacks": 0.005,
            "mark_runtime_wide_data": 2.313,
            "mark_embedding": 0.117,
            "mark_compartments": 2.27,
            "unmark": 1.063,
            "minor_gcs_to_evict_nursery": 8.701,
            ...
          }
        },
        { ... },
      ],
      // Sum of the phase times across all slices, including
      // omitted slices. As before, there are <= 65 possible phases.
      "totals": {
        "wait_background_thread": 0.012,
        "mark_discard_code": 2.845,
        "purge": 0.723,
        "mark": 9.831,
        "mark_roots": 0.102,
        "buffer_gray_roots": 3.095,
        "mark_cross_compartment_wrappers": 0.039,
        "mark_c_and_js_stacks": 0.005,
        "mark_runtime_wide_data": 2.313,
        "mark_embedding": 0.117,
        "mark_compartments": 2.27,
        "unmark": 1.063,
        "minor_gcs_to_evict_nursery": 8.701,
        ...
      }
    },
    ... // Up to four more selected GCs follow.
  ],
  "worst": [
    ... // Same as above, but the 2 worst GCs by max_pause.
  ]
},

fileIOReports

Contains the statistics of main-thread I/O recorded during the execution. Only the I/O stats for the XRE and the profile directories are currently reported, neither of them disclosing the full local path.

Structure:

"fileIOReports": {
  "{xre}": [
    totalTime, // Accumulated duration of all operations
    creates, // Number of create/open operations
    reads, // Number of read operations
    writes, // Number of write operations
    fsyncs, // Number of fsync operations
    stats, // Number of stat operations
  ],
  "{profile}": [ ... ],
  ...
}

lateWrites

This sections reports writes to the file system that happen during shutdown. The reported data contains the stack and the file names of the loaded libraries at the time the writes happened.

The file names of the loaded libraries can contain unicode characters.

Structure:

"lateWrites" : {
  "memoryMap" : [
    ["wgdi32.pdb", "08A541B5942242BDB4AEABD8C87E4CFF2"],
    ... other entries in the format ["module name", "breakpad identifier"] ...
   ],
  "stacks" : [
    [
      [
        0, // the module index or -1 for invalid module indices
        190649 // the offset of this program counter in its module or an absolute pc
      ],
      [1, 2540075],
      ... other frames ...
     ],
     ... other stacks ...
  ],
},

addonDetails

This section contains per add-on telemetry details, as reported by each add-on provider. The XPI provider is the only one reporting at the time of writing (see DXR). Telemetry does not manipulate or enforce a specific format for the supplied provider’s data.

Structure:

"addonDetails": {
  "XPI": {
    "adbhelper@mozilla.org": {
      "scan_items": 24,
      "scan_MS": 3,
      "location": "app-profile",
      "name": "ADB Helper",
      "creator": "Mozilla & Android Open Source Project",
      "startup_MS": 30
    },
    ...
  },
  ...
}

UITelemetry

See the UITelemetry data format documentation.

slowSQL

This section contains the informations about the slow SQL queries for both the main and other threads. The execution of an SQL statement is considered slow if it takes 50ms or more on the main thread or 100ms or more on other threads. Slow SQL statements will be automatically trimmed to 1000 characters. This limit doesn’t include the ellipsis and database name, that are appended at the end of the stored statement.

Structure:

"slowSQL": {
  "mainThread": {
    "Sanitized SQL Statement": [
      1, // the number of times this statement was hit
      200  // the total time (in milliseconds) that was spent on this statement
    ],
    ...
  },
  "otherThreads": {
    "VACUUM /* places.sqlite */": [
      1,
      330
    ],
    ...
  }
},

slowSQLStartup

This section contains the slow SQL statements gathered at startup (until the “sessionstore-windows-restored” event is fired). The structure of this section resembles the one for slowSQL.

UIMeasurements

This section is Android-only and contains UI specific Telemetry measurements and events (see here).

Structure:

"UIMeasurements": [
  {
    "type": "event", // either "session" or "event"
    "action": "action.1",
    "method": "menu",
    "sessions": [],
    "timestamp": 12345,
    "extras": "settings"
  },
  {
    "type": "session",
    "name": "awesomescreen.1",
    "reason": "commit",
    "start": 123,
    "end": 456
  }
  ...
],