olivierli@chromium.org catan-team@chromium.org

Every 10 seconds, a boolean is recorded to this histogram to indicate whether {ProcessAndThreadType} was hung. The thread is hung if the current work unit started running more than 10 seconds ago and this was not labeled as "working as intended" by InvalidateActiveExpectations() (Prior to 2021-07-09 with IgnoreHangsInScope). Recording is skipped if the hang watcher detects an irregular heartbeat (e.g. when the machine sleeps). Warning: Starting with 104.0.5112.79 hang counts have gone way up on macOS which is due to additional coverage added and not necessarily to regressions. See: crrev.com/c/3035824

olivierli@chromium.org catan-team@chromium.org

HangWatcher is a class that periodically checks on watched threads to see if they are making progress or if they are hung. This metric is recorded when the HangWatcher resumes from sleeping for its next watch time. The value is the time difference between when the HangWatcher woke up and the planned wake-up time. Not waking up in time is a bad state. It could be caused by an overloaded machine or another problem like system sleep. Knowing whether the HangWatcher wakes up in a timely manner in the average case can help analyse trial results and is interesting to track over time. An increasing value for this histogram could hint at a larger problem and would prevent missinterpreting a reduced inflow of hang reports as a improvement when it is in fact caused by hindered capabilities.

olivierli@chromium.org catan-team@chromium.org

Because HangWatcher start is delayed until the sandbox is initialized on Linux and ChromeOS, this measures the delta between the time HangWatcher would have started in an unsandboxed process and the time the HangWatcher actually started. The metric is emitted in the process type's main function (e.g. RendererMain, UtilityMain) right after the sandbox is engaged and right before HangWatcher is started. This will be used to evaluate whether hangs are missed during startup because of this delay.