Export Performance Monitoring data to BigQuery

You can export your Performance Monitoring data into BigQuery for further analysis. BigQuery allows you to analyze the data using BigQuery SQL, export it to another cloud provider, and even use the data for your custom ML models.

Enable BigQuery export

1. To get started, use one of the following options:

   • From your Performance Monitoring dashboard, click Link BigQuery just under your Issues feed.

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

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

When you link your project to BiqQuery:

• Firebase exports a copy of your existing data to BigQuery.

  • At initial linking, Firebase automatically schedules your BigQuery tables to backfill data from the past 7 days so that you can start experimenting right away. Allow a few hours for the initial data to be available in BigQuery.

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

• Firebase sets up regular syncs of your data from your Firebase project to BigQuery.

• 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.

What data is exported to BigQuery?

For each app in the project, the export creates a table that includes all the captured performance events. Each row in the table is a single performance event that can be one of the following:

• Duration trace — includes app start, foreground, background, and all developer instrumented traces

  • event_type is DURATION_TRACE
  • event_name is the same as the trace name

• Trace metric — developer instrumented metrics that are associated with traces, previously known as counters

  • event_type is TRACE_METRIC
  • event_name is the name of the metric
  • parent_trace_name is the trace name that contains this metric

• Screen trace — traces spanning the lifetime of a screen

  • event_type is SCREEN_TRACE
  • event_name is prefix _st_ plus the actual screen name

• Network request

  • event_type is NETWORK_REQUEST
  • event_name is the categorized pattern of the network request URL

Each performance event contains attributes of the event (such as country and carrier of the client device), as well as event-specific information:

• Duration traces, trace metrics, and screen traces contain trace_info
• Trace metrics contain trace_info.metric_info
• Screen traces contain trace_info.screen_info
• Network requests contain network_info

Detailed data schema

What can you do with the exported data?

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

View average app start latency break-down by country

SELECT AVG(trace_info.duration_us), country
FROM `table-name`
WHERE _PARTITIONTIME > TIMESTAMP("YYYY-MM-DD")
AND event_type = "DURATION_TRACE"
AND event_name = "_app_start"
GROUP BY 2;

Check the ratio of frozen frames against various conditions

For example, you can check the ratio of frozen frames alongside the amount of time users spend on each screen of your app when on different radio types (WiFi, 4G, etc.).

SELECT
  AVG(trace_info.duration_us / 1000000) AS seconds_on_screen,
  AVG(trace_info.screen_info.frozen_frame_ratio) AS frozen_frame_ratio,
  event_name,
  radio_type
FROM `table-name`
WHERE _PARTITIONTIME > TIMESTAMP("YYYY-MM-DD")
AND event_type = "SCREEN_TRACE"
GROUP BY event_name, radio_type
ORDER BY event_name, radio_type;

Compute cache hit rate for loading certain types of files from disk

This analysis assumes that you have configured a custom trace for loading from disk with a custom attribute named file-extension and a trace metric named cache-hit that is set to 1 if cache hit and 0 if cache miss.

For example, you can compute the cache hit rate for loading PNG files from disk:

SELECT AVG(trace_info.metric_info.metric_value) AS cache_hit_rate
FROM `table-name`
WHERE _PARTITIONTIME > TIMESTAMP("YYYY-MM-DD")
AND event_type = "TRACE_METRIC"
AND event_name = "cache-hit"
AND parent_trace_name = "loadFromDisk"
AND STRUCT("file-extension", "png") IN UNNEST(custom_attributes);

Check for the time of day that users are issuing network requests

For example, you can check at what hour of the day users from the United States are issuing network requests from your app:

SELECT
  count(1) AS hourly_count,
  EXTRACT(HOUR FROM event_timestamp) AS hour_of_day
FROM `table-name`
WHERE _PARTITIONTIME > TIMESTAMP("YYYY-MM-DD")
AND event_type = "NETWORK_REQUEST"
AND country = "US"
GROUP BY 2 ORDER BY 2;

Take your Performance Monitoring data anywhere

Sometimes you want to access your Performance Monitoring data server-side or push it to another third-party solution. There is currently no charge for exporting data.

You can export your data by:

• Using the BigQuery web UI
• Running the CLI command bq extract
• Submitting an extract job via the API or client libraries.

Pricing

There is no charge for exporting data from Performance Monitoring, and BigQuery provides generous free usage limits. For detailed information, refer to BigQuery pricing or the BigQuery sandbox.