Timeouts and Retries

By default curl waits a very long time. In a script that is a bug: one unreachable host and your job hangs until someone notices.

curl -s --max-time 5 --connect-timeout 2 URL

--connect-timeout limits how long to spend getting connected. --max-time limits the whole operation. Exceeding either gives exit code 28.

Retrying needs more care than it looks. Repeating a failed GET or PUT is safe, because both are idempotent. Repeating a POST may create a second record, and a timeout is the worst case for it: you never learned whether the first one was processed.

So retry reads freely, and think before retrying writes. When you must, ask the server for a way to make it safe, such as an idempotency key.