Fix a random failure in 038_save_logical_slots_shutdown.pl.
authorAmit Kapila <[email protected]>
Wed, 13 Mar 2024 03:03:26 +0000 (08:33 +0530)
committerAmit Kapila <[email protected]>
Wed, 13 Mar 2024 03:03:26 +0000 (08:33 +0530)
The test ensures that all the WAL on the publisher is sent to the
subscriber before shutdown by comparing the confirmed_flush_lsn of the
associated slot with the shutdown_checkpoint WAL location. But if the
shutdown_checkpoint location falls into a new page in the WAL then the
check won't work. So, ensure that the shutdown_checkpoint WAL record
doesn't fall into a new page.

Reported-by: Bharath Rupireddy
Author: Bharath Rupireddy
Reviewed-by: Vignesh C, Kuroda Hayato, Amit Kapila
Discussion: https://postgr.es/m/CALj2ACVLzH5CN-h9=S26mdRHPuJ9yDLUw70yh4JOiPw03WL0CQ@mail.gmail.com

src/test/recovery/t/038_save_logical_slots_shutdown.pl

index 5a4f5dc1d4048d5595b22600ddb2290d7fe92a83..2a3bf1f8da9f83788935dbe06241776714255789 100644 (file)
@@ -60,6 +60,12 @@ $node_subscriber->start;
 $node_publisher->safe_psql('postgres', "CREATE TABLE test_tbl (id int)");
 $node_subscriber->safe_psql('postgres', "CREATE TABLE test_tbl (id int)");
 
+# To avoid a shutdown checkpoint WAL record (that gets generated as part of
+# the publisher restart below) falling into a new page, advance the WAL
+# segment. Otherwise, the confirmed_flush_lsn and shutdown_checkpoint
+# location won't match.
+$node_publisher->advance_wal(1);
+
 # Insert some data
 $node_publisher->safe_psql('postgres',
    "INSERT INTO test_tbl VALUES (generate_series(1, 5));");
@@ -81,6 +87,10 @@ is($result, qq(5), "check initial copy was done");
 
 my $offset = -s $node_publisher->logfile;
 
+# Note: Don't insert any data on the publisher that may cause the shutdown
+# checkpoint to fall into a new WAL file. See the comments atop advance_wal()
+# above.
+
 # Restart the publisher to ensure that the slot will be flushed if required
 $node_publisher->restart();