retryflake

Run flaky commands with retries, exponential backoff, and jitter. One binary, zero dependencies — for the CI step that fails every third run, the deploy script that hits a rate limit, the pip install that drops once a week.

pip install retryflake

retryflake --retries 5 -- make test
retryflake --retries 5 --timeout 120 -- ./deploy.sh
retryflake --retry-on "rate limit" --retry-on "429" -- ./call-api.sh
retryflake --retries 0 -- ./must-pass-once.sh

Output streams live to your terminal exactly as the command produces it; retryflake's own chatter (attempt 2/4 failed (exit 1), retrying in 3.2s) goes to stderr, and --quiet silences it.

Behavior

• Exit code is the command's exit code: 0 on success, the last attempt's code on failure, 124 if an attempt was killed by --timeout, 127/126 if the command can't start (no pointless retries), 130 on Ctrl-C (interrupts the child too, never retried).
• Backoff is delay * 2^attempt, capped at --max-delay (defaults: 1s base, 60s cap), with full jitter on by default (--no-jitter disables it).
• --timeout kills the whole process group per attempt: SIGTERM, then SIGKILL after a 5s grace period. Timeouts always retry.
• --retry-on (repeatable) retries only when the attempt's output contains the substring — retry on rate limit, not on invalid API key. A bounded 128 KiB tail is kept for matching, so memory stays flat on huge logs.

Reference

POSIX only (Linux, macOS). Python 3.9+.

License

MIT.