From 2791cf7a0d6e2ac8f0232c15b197dc9284b6bc01 Mon Sep 17 00:00:00 2001 From: Heikki Linnakangas Date: Sat, 10 Feb 2024 00:35:46 +0200 Subject: [PATCH] tests: Make test_vm_bit_clear_on_heap_lock more robust again. When checking that the contents of the VM page in cache and in pageserver match, ignore the LSN on the page. It could be different, if the page was flushed from cache by a checkpoint, for example. Here's one such failure from the CI that this hopefully fixes: https://neon-github-public-dev.s3.amazonaws.com/reports/pr-6687/7847132649/index.html#suites/8545ca7650e609b2963d4035816a356b/5f9018db15ef4408/ In the passing, also remove some log.infos from the loop. I added them while developing the tests, but now they're just noise. --- test_runner/regress/test_vm_bits.py | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-) diff --git a/test_runner/regress/test_vm_bits.py b/test_runner/regress/test_vm_bits.py index 06c30b8d8169..904952d15eb8 100644 --- a/test_runner/regress/test_vm_bits.py +++ b/test_runner/regress/test_vm_bits.py @@ -169,10 +169,14 @@ def test_vm_bit_clear_on_heap_lock(neon_env_builder: NeonEnvBuilder): # The VM page in shared buffer cache, and the same page as reconstructed # by the pageserver, should be equal. + # + # Ignore the LSN on the page though (first 8 bytes). If the dirty + # VM page is flushed from the cache for some reason, it gets WAL-logged, + # which changes the LSN on the page. cur.execute("select get_raw_page( 'vmtest_lock', 'vm', 0 )") - vm_page_in_cache = (cur.fetchall()[0][0])[:100].hex() + vm_page_in_cache = (cur.fetchall()[0][0])[8:100].hex() cur.execute("select get_raw_page_at_lsn( 'vmtest_lock', 'vm', 0, pg_current_wal_insert_lsn() )") - vm_page_at_pageserver = (cur.fetchall()[0][0])[:100].hex() + vm_page_at_pageserver = (cur.fetchall()[0][0])[8:100].hex() assert vm_page_at_pageserver == vm_page_in_cache @@ -203,16 +207,6 @@ def test_vm_bit_clear_on_heap_lock(neon_env_builder: NeonEnvBuilder): for _ in range(1000): cur.execute("select test_consume_xids(10000);") for _ in range(1000): - cur.execute( - "select get_raw_page_at_lsn( 'vmtest_lock', 'vm', 0, pg_current_wal_insert_lsn() )" - ) - page = (cur.fetchall()[0][0])[:100].hex() - log.info(f"VM page contents: {page}") - - cur.execute("select get_raw_page( 'vmtest_lock', 'vm', 0 )") - page = (cur.fetchall()[0][0])[:100].hex() - log.info(f"VM page contents in cache: {page}") - cur.execute("select min(datfrozenxid::text::int) from pg_database") datfrozenxid = int(cur.fetchall()[0][0]) log.info(f"datfrozenxid {datfrozenxid} locking_xid: {locking_xid}")