Hi all, Here is a tiny patch removing PreventCommandDuringRecovery() for UNLISTEN. See previous discussion in https://www.postgresql.org/message-id/CADT4RqBweu7QKRYAYzeRW77b%2BMhJdUikNe45m%2BfL4GJSq_u2Fg%40mail.gmail.com .
In a nutshell, this prevents an error being raised when UNLISTEN is issued during recovery. The operation is a no-op (since LISTEN is still disallowed). This logic here is that some clients (namely Npgsql) issue UNLISTEN * to clear connection state (in the connection pool), but this needlessly breaks when the backend is in recovery. On a related note, there currently doesn't seem to be a good way for clients to know whether the backend is in recovery. As a backend can come out of recovery at any point, perhaps an asynchronous ParameterStatus announcing this state change could be useful. Hopefully this also qualifies for backporting to earlier version branches. Shay
diff --git a/src/backend/tcop/utility.c b/src/backend/tcop/utility.c index 970c94ee80..3efd262cb8 100644 --- a/src/backend/tcop/utility.c +++ b/src/backend/tcop/utility.c @@ -629,7 +629,7 @@ standard_ProcessUtility(PlannedStmt *pstmt, { UnlistenStmt *stmt = (UnlistenStmt *) parsetree; - PreventCommandDuringRecovery("UNLISTEN"); + /* allow UNLISTEN during recovery, which is a noop */ CheckRestrictedOperation("UNLISTEN"); if (stmt->conditionname) Async_Unlisten(stmt->conditionname);