On Tuesday, February 10, 2026 5:34 PM shveta malik <[email protected]> wrote: > > Thanks for the patch. > > + * Note that we do not wait and retry if the local slot has been invalidated. > + * In such cases, the corresponding remote slot on the primary is > + likely > + * invalidated as well. Even if only the local slot is invalidated, > + simply > + * retrying synchronization won't suffice, as it requires further user > + actions > + * to verify the server configuration, drop the invalidated slot. > > On thinking more, I realized that if the local slot is invalidated alone > while the > remote-slot is not, we do not wait for the user to drop such an invalidated > slot. Instead slot-sync will drop it internally. See comments atop > drop_local_obsolete_slots(). This makes me wonder whether such a case, > where only the local slot is invalidated, should also set slotsync_pending = > true, since there is a good chance it will get synchronized in subsequent > runs. > OTOH, if we do not wait for such a slot, we could end up in a situation where > the slot (remote one) is valid pre-failover but is invalid (synced one) post- > failover, even after running the API immediately before switchover. Thoughts?
I agree that it makes sense to retry when only the local slot is invalidated. Here is the updated patch. Best Regards, Hou zj
v5-0004-Add-a-taptest.patch
Description: v5-0004-Add-a-taptest.patch
v5-0001-Refactoring-remove-some-unnecessary-func-paramete.patch
Description: v5-0001-Refactoring-remove-some-unnecessary-func-paramete.patch
v5-0002-Refactoring-move-similar-checks-to-a-central-plac.patch
Description: v5-0002-Refactoring-move-similar-checks-to-a-central-plac.patch
v5-0003-Improve-the-retry-logic-in-pg_sync_replication_sl.patch
Description: v5-0003-Improve-the-retry-logic-in-pg_sync_replication_sl.patch
