Understanding message delivery

FCM provides three sets of tools to help you get insight into message delivery:

• Firebase console message delivery reports
• Aggregated Android SDK delivery metrics from the Firebase Cloud Messaging Data API
• Comprehensive data export to Google BigQuery

The reporting tools described in this page all require Google Analytics in order to function. If Google Analytics is not enabled for your project, you can set it up in the integrations tab of your Firebase project settings.

Keep in mind that the reporting of many of the statistics on this page, are subject to delays up to 24 hours due to batching of analytics data.

Message delivery reports

In the Reports tab in the Firebase console, you can view the following data for messages sent to Android or Apple platform FCM SDKs, including those sent via the Notifications composer and the FCM APIs:

• Sends — The data message or notification message has been enqueued for delivery or has been successfully passed to a third-party service like APNs for delivery. See lifetime of a message for more information.
• Received (available only on Android devices) — The data message or notification message has been received by the app. This data is available when the receiving Android device has FCM SDK 18.0.1 or higher installed.
• Impressions (available only for notification messages on Android devices) — The display notification has been displayed on the device while the app is in the background.
• Opens — The user opened the notification message. Reported only for notifications received when the app is in the background.

This data is available for all messages with a notification payload and all labeled data messages. To learn more about labels, see Adding analytics labels to messages.

When viewing message reports, you can set a date range for the data displayed, with the option to export to CSV. You can also filter by these criteria:

• Platform (iOS or Android)
• App
• Custom analytics labels

Adding analytics labels to messages

Labeling messages is very useful for custom analysis, allowing you to filter delivery statistics by labels or sets of labels. You can add a label to any message sent via the HTTP v1 API by setting the fcmOptions.analyticsLabel field in the message object, or in the platform-specific AndroidFcmOptions or ApnsFcmOptions fields.

Analytics labels are text strings in the format ^[a-zA-Z0-9-_.~%]{1,50}$. Labels can include lower and upper case letters, numbers, and the following symbols:

• -
• ~
• %

Max length is 50 characters. You can specify up to 100 unique labels per day; messages with labels added beyond that limit are not reported.

In the Firebase console messaging Reports tab, you can search a list of all existing labels and apply them singly or in combination to filter the statistics displayed.

Aggregated delivery data via the FCM Data API

The Firebase Cloud Messaging Data API lets you retrieve information that can help you understand the outcomes of message requests targeted to Android applications. The API provides aggregated data across all data collection-enabled Android devices in a project. This includes details on the percentage of messages delivered without delay as well as how many messages were delayed or dropped within the Android Transport Layer. Evaluating this data can reveal broad trends in message delivery and help you find effective ways to improve the performance of your send requests. See Aggregate data timelines for information on date range availability in the reports.

The API provides all data available for a given application. See the API reference documentation.

How is the data broken down?

Delivery data is broken down by application, date, and analytics label. A call to the API will return data for every combination of date, application, and analytics label. For example, a single androidDeliveryData JSON object would look like this:

 {
  "appId": "1:23456789:android:a93a5mb1234efe56",
  "date": {
    "year": 2021,
    "month": 1,
    "day": 1
  },
  "analyticsLabel": "foo",
  "data": {
    "countMessagesAccepted": "314159",
    "messageOutcomePercents": {
      "delivered": 71,
      "pending": 15
    },
   "deliveryPerformancePercents": {
      "deliveredNoDelay": 45,
      "delayedDeviceOffline": 11
    }
  }

How to Interpret the Metrics

Delivery data outlines the percentage of messages that fit each of the following metrics. It is possible that a single message fits multiple metrics. Due to limitations in how we collect the data and the level of granularity at which we aggregated the metrics, some message outcomes are not represented in the metrics at all, so the percentages below will not sum to 100%.

Count Messages Accepted

The only count included in the dataset is the count of messages that were accepted by FCM for delivery to Android devices. All percentages use this value as the denominator. Keep in mind that this count won't include messages targeted to users who have disabled the collection of usage and diagnostic information on their devices.

Message Outcome Percentages

The fields included in the MessageOutcomePercents object provide information on the outcomes of message requests. The categories are all mutually exclusive. It can answer questions such as "Are my messages being delivered?" and "What is causing messages to be dropped?"

For example, a high value for the droppedTooManyPendingMessages field could signal that app instances are receiving volumes of non-collapsible messages exceeding FCM's limit of 100 pending messages. To mitigate this, make sure your app handles calls to onDeletedMessages, and consider sending collapsible messages. Similarly, high percentages for droppedDeviceInactive could be a signal to update registration tokens on your server, removing stale tokens and unsubscribing them from topics. See Manage FCM registration tokens for best practices in this area.

Delivery Performance Percents

The fields in the DeliveryPerformancePercents object provide information about messages that were successfully delivered. It can answer questions such as "Were my messages delayed?" and "Why are messages delayed?" For example, a high value for delayedMessageThrottled would clearly indicate that you are exceeding per-device maximum limits, and should adjust the rate at which you are sending messages.

Message Insight Percentagess

This object provides additional information about all message sends. The priorityLowered field expresses the percentage of accepted messages that had priority lowered from HIGH to NORMAL. If this value is high, try sending fewer high priority messages or ensure that you always display a notification when a high priority message is sent. See our documentation on message priority for more info

How does this data differ from data exported to BigQuery?

The BigQuery export provides individual message logs about message acceptance by the FCM backend and message delivery in the SDK on the device (Steps 2 and 4 of the FCM Architecture). This data is useful for ensuring individual messages were accepted and delivered. Read more about BigQuery data export in the next section.

By contrast, the Firebase Cloud Messaging Data API provides aggregated details about what happens specifically in the Android Transport Layer (or Step 3 of the FCM Architecture). This data specifically provides insight into the delivery of messages from FCM backends to the Android SDK. It’s particularly useful for showing trends as to why messages were delayed or dropped during this transport.

In some cases, it is possible that the two data sets might not match precisely due to the following:

• The aggregated metrics only sample a portion of all messages
• The aggregated metrics are rounded
• We don’t present metrics below a privacy threshold
• A portion of message outcomes are missing due to optimizations in how we manage the large volume of traffic.

Limitations of the API

Aggregate Data Timelines

The API will return 7 days of historical data; however, data returned by this API will be delayed by up to 5 days. For example, on January 20th, the data for January 9th - January 15th would be available, but not for January 16th or later. Additionally, the data is provided at best effort. In the event of a data outage, FCM will work to fix forward and will not backfill the data after the issue is fixed. In larger outages, the data could be unavailable for a week or more.

Data Coverage

The metrics provided by the Firebase Cloud Messaging Data API are meant to provide insight into broad trends of message delivery. However, they do not provide 100% coverage of all message scenarios. The following scenarios are known outcomes not reflected in the metrics.

Expired messages

If the Time To Live (TTL) expires after the end of the given log date, the message won't be counted as droppedTtlExpired on this date.

Messages to inactive devices

Messages sent to inactive devices may or may not show up in the dataset depending on which data path they take. This can lead to some miscounting in the droppedDeviceInactive and pending fields.

Messages to devices with certain user preferences

Users who have disabled the collection of usage and diagnostic information on their devices will not have their messages included in our counting, in keeping with their preferences.

Rounding and Minimums

FCM deliberately rounds and excludes counts where the volumes are not large enough.

BigQuery data export

You can export your message data into BigQuery for further analysis. BigQuery allows you to analyze the data using BigQuery SQL, export it to another cloud provider, or use the data for your custom ML models. An export to BigQuery includes all available data for messages, regardless of message type or whether the message is sent via the API or the Notifications composer.

For messages sent to devices with the following FCM SDK minimum versions, you have the additional option to enable the export of message delivery data for your app:

• Android 20.1.0 or higher.
• iOS 8.6.0 or higher
• Firebase Web SDK 9.0.0 or higher

See below for details on enabling data export for Android and iOS.

To get started, link your project to BigQuery:

1. Choose one of the following options:

   • Open the Notifications composer, then click Access BigQuery at the bottom of the page.

   • From the Integrations page in the Firebase console, click Link in the BigQuery card.

     This page displays FCM export options for all FCM-enabled apps in the project.

2. Follow the on-screen instructions to enable BigQuery.

Refer to Link Firebase to BigQuery for more information.

When you enable BigQuery export for Cloud Messaging:

• Firebase exports your data to BigQuery. Note that the initial propagation of data for export may take up to 48 hours to complete.

  • You can manually schedule data backfills for up to the past 30 days.

• After the dataset is created, the location can't be changed, but you can copy the dataset to a different location or manually move (recreate) the dataset in a different location. To learn more, see Change dataset location.

• Firebase sets up regular syncs of your data from your Firebase project to BigQuery. These daily export operations begin at 4:00 AM Pacific Time and usually finish in 24 hours.

• By default, all apps in your project are linked to BigQuery and any apps that you later add to the project are automatically linked to BigQuery. You can manage which apps send data.

To deactivate BigQuery export, unlink your project in the Firebase console.

Enable message delivery data export

The FCM SDK for Web 9.0.0 provides an Alpha release of delivery data export. Data export is enabled and disabled at the app instance level. When an end user gives consent or rejects data collection, the app should set the experimental enable/disable flag for each app instance as shown:

  // userConsent holds the decision of the user to give big query export consent.
  const userConsent = ...;

  const messaging = getMessagingInSw(app);

  experimentalSetDeliveryMetricsExportedToBigQuery(messaging, userConsent);

What data is exported to BigQuery?

Note that targeting stale tokens or inactive registrations may inflate some of these statistics.

The schema of the exported table is:

What can you do with the exported data?

The following sections offer examples of queries that you can run in BigQuery against your exported FCM data.

Count sent messages by app

SELECT app_name, COUNT(1)
FROM `project ID.firebase_messaging.data`
WHERE
  _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
  AND event = 'MESSAGE_ACCEPTED'
  AND message_id != ''
GROUP BY 1;

Count unique app instances targeted by messages

SELECT COUNT(DISTINCT instance_id)
FROM `project ID.firebase_messaging.data`
WHERE
  _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
  AND event = 'MESSAGE_ACCEPTED';

Count notification messages sent

SELECT COUNT(1)
FROM `project ID.firebase_messaging.data`
WHERE
  _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
  AND event = 'MESSAGE_ACCEPTED'
  AND message_type = 'DISPLAY_NOTIFICATION';

Count data messages sent

SELECT COUNT(1)
FROM `project ID.firebase_messaging.data`
WHERE
  _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
  AND event = 'MESSAGE_ACCEPTED'
  AND message_type = 'DATA_MESSAGE';

Count messages sent to a topic or campaign

SELECT COUNT(1)
FROM `project ID.firebase_messaging.data`
WHERE
  _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
  AND event = 'MESSAGE_ACCEPTED'
  AND bulk_id = your bulk id AND message_id != '';

To track events for a message sent to particular topic, modify this query to replace AND message_id != '' with AND message_id = <your message id>;.

Compute fanout duration for a given topic or campaign

The fanout start time is when the original request is received, and the end time is the time the last individual message targeting a single instance is created.

SELECT
  TIMESTAMP_DIFF(
    end_timestamp, start_timestamp, MILLISECOND
  ) AS fanout_duration_ms,
  end_timestamp,
  start_timestamp
FROM (
    SELECT MAX(event_timestamp) AS end_timestamp
    FROM `project ID.firebase_messaging.data`
    WHERE
      _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
      AND event = 'MESSAGE_ACCEPTED'
      AND bulk_id = your bulk id
  ) sent
  CROSS JOIN (
    SELECT event_timestamp AS start_timestamp
    FROM `project ID.firebase_messaging.data`
    WHERE
      _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
      AND event = 'MESSAGE_ACCEPTED'
      AND bulk_id = your bulk id
      AND message_type = 'TOPIC'
  ) initial_message;

Count percentage of delivered messages

SELECT
  messages_sent,
  messages_delivered,
  messages_delivered / messages_sent * 100 AS percent_delivered
FROM (
    SELECT COUNT(DISTINCT CONCAT(message_id, instance_id)) AS messages_sent
    FROM `project ID.firebase_messaging.data`
    WHERE
      _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
      AND event = 'MESSAGE_ACCEPTED'
  ) sent
  CROSS JOIN (
    SELECT COUNT(DISTINCT CONCAT(message_id, instance_id)) AS messages_delivered
    FROM `project ID.firebase_messaging.data`
    WHERE
      _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
      AND (event = 'MESSAGE_DELIVERED'
      AND message_id
      IN (
        SELECT message_id FROM `project ID.firebase_messaging.data`
        WHERE
          _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
          AND event = 'MESSAGE_ACCEPTED'
        GROUP BY 1
      )
  ) delivered;

Track all events for a given message id and instance id

SELECT *
FROM `project ID.firebase_messaging.data`
WHERE
    _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
    AND message_id = 'your message id'
    AND instance_id = 'your instance id'
ORDER BY event_timestamp;

Compute latency for a given message id and instance id

SELECT
  TIMESTAMP_DIFF(
    MAX(delivered_time), MIN(accepted_time), MILLISECOND
  ) AS latency_ms
FROM (
    SELECT event_timestamp AS accepted_time
    FROM `project ID.firebase_messaging.data`
    WHERE
      _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD')
      AND message_id = 'your message id'
      AND instance_id = 'your instance id'
      AND event = 'MESSAGE_ACCEPTED'
  ) sent
  CROSS JOIN (
    SELECT event_timestamp AS delivered_time
    FROM `project ID.firebase_messaging.data`
    WHERE
      _PARTITIONTIME = TIMESTAMP('date as YYYY-MM-DD') AND
      message_id = 'your message id' AND instance_id = 'your instance id'
      AND (event = 'MESSAGE_DELIVERED'
  ) delivered;