Add max total crash limit to stop relaunching persistently failing pr…

[kezhangMS]

…ocesses

The existing crash protection only stops relaunching a process if it
exits more than 10 times within 60 seconds. This misses the case where
a process fails at a slower but persistent rate (e.g. 2 times per
minute), which causes it to be relaunched indefinitely.

Add a total crash counter (max 50) alongside the existing per-minute
rolling window so that persistently failing processes are eventually
stopped regardless of crash rate.

[benhillis]

Triage note (2026-05): Thanks — the rationale (slow but persistent crash loops bypass the per-minute rate limit and run forever) is correct and the fix is the right shape. Two small things for the next round:

1. c_maxTotalCrashes = 20 over the lifetime of a WSLGd process is quite tight — a long-lived session that crashes Weston a few times across, say, a network change, a sleep/wake, and an rdp-disconnect over the course of a workday could plausibly hit it. Consider 50 or making it a function of uptime. Happy to take 20 if you''ve actually seen the pathology you''re guarding against trigger inside that budget.
2. Please add a brief log line when the total cap is reached that distinguishes it from the per-minute cap in /mnt/wslg/stderr.log — your LOG_INFO does that, good.

Will queue for an internal buddy build once you reply or push an update.

[kezhangMS]

Thanks for the comments. Updated the total crash threshold from 20 to 50.

[Copilot]

Pull request overview

Adds a cumulative crash counter to ProcessMonitor::Run so that processes which fail at a slow but persistent rate are eventually given up on, in addition to the existing per-minute rolling-window limit. Also replaces magic numbers in the log message with named constants.

Changes:

• Introduces totalCrashes map and c_maxTotalCrashes constant to track lifetime crash counts per command.
• Refactors the per-minute threshold to a named constant c_maxCrashesPerMinute and updates the log message.
• Adds a new log/stop branch when total crashes exceed the lifetime threshold.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.