mirror of
https://github.com/neondatabase/neon.git
synced 2026-01-06 04:52:55 +00:00
fc242afcc2
## Problem When testing tenant splits, I found that PG can get backpressure throttled indefinitely if the split is aborted afterwards. It turns out that each PageServer activates new shard separately even before the split is committed and they may start sending PageserverFeedback to PG directly. As a result, if the split is aborted, no one resets the pageserver feedback in PG, and thus PG will be backpressure throttled forever unless it's restarted manually. ## Summary of changes This PR fixes this problem by having `walprop_pg_process_safekeeper_feedback` simply ignore all pageserver feedback from unknown shards. The source of truth here is defined by the shard map, which is guaranteed to be reloaded only after the split is committed. Co-authored-by: Chen Luo <chen.luo@databricks.com>