AynSiddiqui commented on issue #2443: URL: https://github.com/apache/iceberg-go/issues/2443#issuecomment-6879463888
#2492 has merged. Here is the plan for the retry PR. In `catalog/rest/rest.go`, create one key and body per call. Reuse both for resends after eligible 5xx responses, timeouts or connection errors. Return every 4xx immediately. A transaction refreshed after a 409 makes a new call with a new key. Measure the advertised lifetime from the first attempt. The explicit retry loop will not resend with a missing, zero, negative or invalid lifetime. Stop before a delay or send would exceed the window. Return `ErrCommitStateUnknown` for an ambiguous commit when expiry prevents a resend; for other calls, return the last error. Honor cancellation and bound the retry count. Move scan plan and task POSTs to the same retry policy. They will stop resending on 408 and 429, a change from their current behavior. Scan GET polling stays unchanged. Tests will cover matching keys and bodies on retries, fresh keys on separate calls and 409 refreshes, and no explicit resend on 4xx. They will also cover expiry, a slow first attempt, cancellation, missing or invalid lifetimes, and both scan POSTs. @zeroshade, you said `net/http` replay is fine when the server advertises support. `TestMutationTransportReplay` shows it also replays with `PT0S`, and a slow first attempt may outlast the lifetime. Should we leave that behavior as-is, or bound transport replay too, for example by omitting the key or disabling replay when the lifetime is zero? Should caller-supplied scan keys be suppressed without advertisement, and should their age from an earlier call reduce the available window? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
