Git · Automation · Cron

14 entries

Unattended automation mostly fails without telling anyone. That is the theme of this chapter.

An auto-push that swallows failures produces permanent desync

Symptom

No git identity in the container: rebase dies halfway

Symptom

Two schedulers producing the same artifact collide every day

Symptom

Cron + a `pgrep` duplicate-run guard catches itself

Symptom

Forget the `trap` on a lock file and every subsequent run dies

Symptom

The safeguard blocks the thing it was protecting (the general case)

Same shape three times in one day.

Shared hosting's process limit makes your watchdog kill its own child

Symptom

`nohup` alone dies when the session ends

Symptom

Widen the filter and already-closed items come back to life

Symptom

"Which file actually runs" is settled only by following the call chain to the end

Symptom

Tokens

When a resource runs out, your diagnostic tools go with it

The incident

Long polling wants exactly one consumer — never run two processes on one token

Symptom

Flipping one feature switch turned off an unrelated feature too

Symptom