You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I've been running a two week eval, solo, before pushing this out to others. Every real failure I hit left the admin error log completely empty. I know there's a family of open issues that echo this (#375 , #377 , #602 ). So maybe the useful framing is - the error logs intake is too narrow to catch the failures that actually happen.
These are the ones i'm facing:
An app publish that crashed at boot (bad entrypoint). The deployment kept reporting "running" with a stale endpoint. deploy: publish pre-flight — inherit the entrypoint #442 's preflight covers the entrypoint case now, but a crash at boot after a successful publish still shows up nowhere.
The sandbox exec transport degraded after one full length (300s) foreground command. After that every blocking command over ~10s failed instantly with fetch failed while the container stayed healthy and fast commands worked fine. NOthing recorded. I see sandbox: out-of-band computer status and restart via the execute tool #469 added detection and auto-restart on the sprite backend, but on the local docker backend its still totally silent.
A run that died mid task with stopped:true and an empty reply. The runs view shows it as done, attempts 1.
I get that the log can't record what the process didn't survive. But "failed inside a run that completed" and "failed above the run layer" both route to nothing right now. If deploy boot-crashes, exec-transport faults on all backends, and abnormal run termination each wrote an error event, the error page would be the first place I check instead of the last.
I've been running a two week eval, solo, before pushing this out to others. Every real failure I hit left the admin error log completely empty. I know there's a family of open issues that echo this (#375 , #377 , #602 ). So maybe the useful framing is - the error logs intake is too narrow to catch the failures that actually happen.
These are the ones i'm facing:
entrypoint). The deployment kept reporting "running" with a stale endpoint. deploy: publish pre-flight — inherit the entrypoint #442 's preflight covers the entrypoint case now, but a crash at boot after a successful publish still shows up nowhere.fetch failedwhile the container stayed healthy and fast commands worked fine. NOthing recorded. I see sandbox: out-of-band computer status and restart via the execute tool #469 added detection and auto-restart on the sprite backend, but on the local docker backend its still totally silent.stopped:trueand an empty reply. The runs view shows it as done, attempts 1.I get that the log can't record what the process didn't survive. But "failed inside a run that completed" and "failed above the run layer" both route to nothing right now. If deploy boot-crashes, exec-transport faults on all backends, and abnormal run termination each wrote an error event, the error page would be the first place I check instead of the last.