gkoszyk opened a new issue, #4135: URL: https://github.com/apache/iggy/issues/4135
### Test to fix `given_a_failed_writeback_when_a_fresh_handle_syncs_then_it_should_not_report_success` — `core/simulator/src/storage/tests.rs:2610`, ignored in #4132. It asserts that after a device drops an inode's dirty pages, a descriptor opened **afterwards** must not report a successful `fsync` over the lost bytes. ### Why it fails `checkpoint_files` (`core/journal/src/partition_journal.rs:356-361`) opens its own descriptor for each dirty path and syncs through that, rather than through the writer that produced the bytes. `synced_files` (`core/partitions/src/persistence.rs:795-800`) covers consumer-offset paths only. Under `durability = replicated` + `consumer_offset_durability = persisted`, `IggyIndexWriter` is constructed with `fsync = runtime.durability.is_persisted()` = `false` (`core/server/src/partition_helpers.rs:391`, `:461`), so that fresh-descriptor `fsync` is the **only** index barrier in that configuration. Linux samples errseq at open. A writeback error recorded before that open is reported to nobody, the `fsync` returns success, and the WAL then reclaims history the index never received. The simulator could not express this before #4132 — it had no per-inode error state and no per-handle errseq sampling. `SimStorage::fail_writeback` adds both. ### Done when `synced_files` covers segment and index paths synced through their live writers, or those writers are fsynced in `commit_messages_inner` before the checkpoint is requested. `#[ignore]` removed and passing without `--ignored`. Found by an adversarial review of #4092. See also #4128 — same class of defect on the *purge* path rather than the checkpoint path. -- 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]
