Crash reports filed as GitHub issues automatically
Pool the spare power of every computer into one cloud. Two apps (sys0
Brought to you by:
a-hackers-guide
Menu
▾
▴
#31 Crash reports filed as GitHub issues automatically
Originally created by: A-Hackers-Guide
When sys0 or sys1 crashes, the project should hear about it without anyone copying logs.
- the watchdog captures the panic and stack trace of every crash, scrubbed of anything personal: no IP addresses, host names, user names, file paths, join codes, keys or app data; only the version, OS, CPU and the code locations
- clients hand reports to their node; the node collects them, groups repeats by their stack signature and shows them on the dashboard
- the node forwards each new crash to a small intake service whose code lives in this repo (
intake/). Only the intake holds a GitHub token (fine-grained, issues only), so no token is ever in an app. It files one issue per distinct crash, labeledcrash, and counts repeats on the existing issue, with rate limits - on by default, can be turned off; the dashboard shows exactly what is sent
- until the intake is deployed, the dashboard offers a prefilled "Report on GitHub" link instead
Originally posted by: A-Hackers-Guide
Built in v0.8.0: crashes are captured, scrubbed, grouped on the node, shown under Machines → Crash reports with a prefilled "Report on GitHub" link, and sent to the crash intake when one is configured.
Left to do (needs a maintainer): deploy
intake/to Cloudflare with a fine-grained token that can only write issues on this repository, then setcrash.DefaultIntake. Steps in docs/CRASHES.md.