A quiet night and a broken pager look the same from the outside. That sentence is the whole problem. Every observability stack I have ever built assumes the difference is obvious, and every incident r
Your probe says "DOWN", but the target responds in 281ms. Here is why "timeout" is a lie, and how to separate a dead target from a blind probe.
A pg_dump killed mid-pipe leaves a file that exits 0, passes gzip -t, and looks completely valid.
A pg_dump killed mid-pipe leaves a file that exits 0, passes gzip -t, and passes pg_restore -l. Measured, with the one check that catches it.
The failures that actually hurt never set a non-zero exit code. How to monitor for the absence of good output instead of the presence of errors.