My Acer touchscreen occasionally displays “Web process crashed.” Sometimes the dashboard recovers without intervention a few minutes later.
Setup:
- Raspberry Pi with approximately 4 GB RAM
- Home Assistant OS 18.2
- Home Assistant Core 2026.9.2
- HAOS Kiosk Display 1.3.2
- Browser refresh: 600 seconds
- Watchdog enabled
Host logs show out-of-memory kills of WebKitWebProces on September 11, 13 and 16, 2026. At the September 16 event, the renderer used approximately 1130 MiB resident memory and 783 MiB swap, with system swap almost exhausted.
The displayed dashboard includes weather, entity cards, a scrolling clock, and Powerwall visualization/history. It has no camera cards. A later 24-minute measurement showed kiosk memory fluctuating around 25%, without sustained growth, so we have not identified what triggers the exhaustion.
A separate possible recovery bug was identified during AI-assisted source review of haoskiosk/userconf.lua: the load-status callback increments consecutive_load_failures when status is "failed", but appears to fall through to an unconditional reset to zero. That would prevent successive failures from reaching MAX_LOAD_FAILURES=5. This finding has not been tested in the running browser, and we do not know whether renderer crashes trigger that callback.
Is there a supported fix for these memory crashes, and a way to recover automatically when the renderer fails while the kiosk remains running?
I can provide relevant host and kiosk log excerpts.
My Acer touchscreen occasionally displays “Web process crashed.” Sometimes the dashboard recovers without intervention a few minutes later.
Setup:
Host logs show out-of-memory kills of WebKitWebProces on September 11, 13 and 16, 2026. At the September 16 event, the renderer used approximately 1130 MiB resident memory and 783 MiB swap, with system swap almost exhausted.
The displayed dashboard includes weather, entity cards, a scrolling clock, and Powerwall visualization/history. It has no camera cards. A later 24-minute measurement showed kiosk memory fluctuating around 25%, without sustained growth, so we have not identified what triggers the exhaustion.
A separate possible recovery bug was identified during AI-assisted source review of haoskiosk/userconf.lua: the load-status callback increments consecutive_load_failures when status is "failed", but appears to fall through to an unconditional reset to zero. That would prevent successive failures from reaching MAX_LOAD_FAILURES=5. This finding has not been tested in the running browser, and we do not know whether renderer crashes trigger that callback.
Is there a supported fix for these memory crashes, and a way to recover automatically when the renderer fails while the kiosk remains running?
I can provide relevant host and kiosk log excerpts.