[
https://issues.apache.org/jira/browse/FLINK-40278?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Bharath Reddy Gunapati updated FLINK-40278:
-------------------------------------------
Summary: Document that HTTP sink retries resend the whole batch request
(was: Define and implement HTTP sink batch retry semantics)
> Document that HTTP sink retries resend the whole batch request
> --------------------------------------------------------------
>
> Key: FLINK-40278
> URL: https://issues.apache.org/jira/browse/FLINK-40278
> Project: Flink
> Issue Type: Sub-task
> Components: Connectors / HTTP
> Reporter: Bharath Reddy Gunapati
> Priority: Major
> Labels: pull-request-available
>
> The HTTP sink supports per-request and batch request submission. Per-request
> retry can map failures back to individual sink entries. Batch mode combines
> multiple sink entries into
> one HTTP request, so retry behavior must be explicit.
> Define and implement batch retry semantics for HTTP sink.
> Proposed behavior:
> - If a batch HTTP request fails and the response is retryable, retry all
> original entries in that batch.
> - If a batch HTTP request succeeds or is ignored, complete all entries in
> that batch.
> - Partial per-record retry is out of scope unless a future response
> processor can parse target-specific per-record results.
> Acceptance criteria:
> - Original HttpSinkRequestEntry objects are preserved for retry.
> - Batch failures retry the original entries, not reconstructed serialized
> bytes.
> - Batch behavior is documented.
> - Tests cover per-request mode and batch mode.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
