This page presents data captured from base::ThreadPool histograms at a given point in time so it can be used in future design decisions.
All data is 28-day aggregation on Stable channel.
Number of tasks between two waits by a foreground worker thread in a browser/renderer process.
Histogram name: ThreadPool.NumTasksBetweenWaits.(Browser/Renderer).Foreground Date: August 2019 Values in tables below are percentiles.
| Number of tasks | Browser process | Renderer process |
|---|---|---|
| 1 | 87 | 92 |
| 2 | 95 | 98 |
| 5 | 99 | 100 |
| Number of tasks | Browser process | Renderer process |
|---|---|---|
| 1 | 81 | 90 |
| 2 | 92 | 97 |
| 5 | 98 | 100 |
| Number of tasks | Browser process | Renderer process |
|---|---|---|
| 1 | 92 | 96 |
| 2 | 97 | 98 |
| 5 | 99 | 100 |
Number of tasks run by ThreadPool while task was queuing (from time task was posted until time it was run). Recorded for dummy heartbeat tasks in the browser process. The heartbeat recording avoids dependencies between this report and other work in the system.
Histogram name: ThreadPool.NumTasksRunWhileQueuing.Browser.* Date: September 2019 Values in tables below are percentiles.
Note: In renderer processes, on all platforms/priorities, 0 tasks are run while queuing at 99.5th percentile.
| Number of tasks | USER_BLOCKING | USER_VISIBLE | BEST_EFFORT |
|---|---|---|---|
| 0 | 95 | 93 | 90 |
| 1 | 98 | 95 | 92 |
| 2 | 99 | 96 | 93 |
| 5 | 100 | 98 | 95 |
| Number of tasks | USER_BLOCKING | USER_VISIBLE | BEST_EFFORT |
|---|---|---|---|
| 0 | 100 | 100 | 99 |
| 1 | 100 | 100 | 99 |
| 2 | 100 | 100 | 99 |
| 5 | 100 | 100 | 100 |
| Number of tasks | USER_BLOCKING | USER_VISIBLE | BEST_EFFORT |
|---|---|---|---|
| 0 | 99 | 98 | 97 |
| 1 | 100 | 99 | 99 |
| 2 | 100 | 99 | 99 |
| 5 | 100 | 100 | 100 |
For all priorities, 0 tasks are run while queueing at 99.5th percentile.
The number of tasks that run while a BEST_EFFORT task is queued is unexpectedly low. We should explore creating threads less aggressively, at the expense of keeping BEST_EFFORT tasks in the queue for a longer time. See Bug 906079.