The tests turn out to cause deadlocks in some circumstances. Fairly
reproducibly so with -DRELCACHE_FORCE_RELEASE
-DCATCACHE_FORCE_RELEASE. Some of the deadlocks may be hard to fix
without disproportionate measures, but others probably should be fixed
- but not in 12.
We discussed removing the new tests until we can fix the issues
underlying the deadlocks, but results from buildfarm animal
markhor (which runs with CLOBBER_CACHE_ALWAYS) indicates that there
might be a more severe, as of yet undiagnosed, issue (including on
stable branches) with reindexing catalogs. The failure is:
ERROR: could not read block 0 in file "base/16384/28025": read only 0 of 8192 bytes
Therefore it seems advisable to keep the tests.
It's not certain that running the tests in isolation removes the risk
of deadlocks. It's possible that additional locks are needed to
protect against a concurrent auto-analyze or such.
Per discussion with Tom Lane.
Discussion: https://postgr.es/m/28926.
1556664156@sss.pgh.pa.us
Backpatch: 9.4-, like
3dbb317d3
\set VERBOSITY default
DROP TABLE reindex_verbose;
--
--- check that system tables can be reindexed
---
--- whole tables
-REINDEX TABLE pg_class; -- mapped, non-shared, critical
-REINDEX TABLE pg_index; -- non-mapped, non-shared, critical
-REINDEX TABLE pg_operator; -- non-mapped, non-shared, critical
-REINDEX TABLE pg_database; -- mapped, shared, critical
-REINDEX TABLE pg_shdescription; -- mapped, shared non-critical
--- Check that individual system indexes can be reindexed. That's a bit
--- different from the entire-table case because reindex_relation
--- treats e.g. pg_class special.
-REINDEX INDEX pg_class_oid_index; -- mapped, non-shared, critical
-REINDEX INDEX pg_class_relname_nsp_index; -- mapped, non-shared, non-critical
-REINDEX INDEX pg_index_indexrelid_index; -- non-mapped, non-shared, critical
-REINDEX INDEX pg_index_indrelid_index; -- non-mapped, non-shared, non-critical
-REINDEX INDEX pg_database_oid_index; -- mapped, shared, critical
-REINDEX INDEX pg_shdescription_o_c_index; -- mapped, shared, non-critical
---
-- REINDEX CONCURRENTLY
--
CREATE TABLE concur_reindex_tab (c1 int);
--- /dev/null
+--
+-- Check that system tables can be reindexed.
+--
+-- Note that this test currently has to run without parallel tests
+-- being scheduled, as currently reindex catalog tables can cause
+-- deadlocks:
+--
+-- * The lock upgrade between the ShareLock acquired for the reindex
+-- and RowExclusiveLock needed for pg_class/pg_index locks can
+-- trigger deadlocks.
+--
+-- * The uniqueness checks performed when reindexing a unique/primary
+-- key index possibly need to wait for the transaction of a
+-- about-to-deleted row in pg_class to commit. That can cause
+-- deadlocks because, in contrast to user tables, locks on catalog
+-- tables are routinely released before commit - therefore the lock
+-- held for reindexing doesn't guarantee that no running transaction
+-- performed modifications in the table underlying the index.
+-- Check reindexing of whole tables
+REINDEX TABLE pg_class; -- mapped, non-shared, critical
+REINDEX TABLE pg_index; -- non-mapped, non-shared, critical
+REINDEX TABLE pg_operator; -- non-mapped, non-shared, critical
+REINDEX TABLE pg_database; -- mapped, shared, critical
+REINDEX TABLE pg_shdescription; -- mapped, shared non-critical
+-- Check that individual system indexes can be reindexed. That's a bit
+-- different from the entire-table case because reindex_relation
+-- treats e.g. pg_class special.
+REINDEX INDEX pg_class_oid_index; -- mapped, non-shared, critical
+REINDEX INDEX pg_class_relname_nsp_index; -- mapped, non-shared, non-critical
+REINDEX INDEX pg_index_indexrelid_index; -- non-mapped, non-shared, critical
+REINDEX INDEX pg_index_indrelid_index; -- non-mapped, non-shared, non-critical
+REINDEX INDEX pg_database_oid_index; -- mapped, shared, critical
+REINDEX INDEX pg_shdescription_o_c_index; -- mapped, shared, non-critical
# These depend on create_misc and create_operator
test: create_index create_index_spgist create_view index_including index_including_gist
+# ----------
+# Has to run in isolation, due to deadlock risk
+# ----------
+test: reindex_catalog
+
# ----------
# Another group of parallel tests
# ----------
test: create_view
test: index_including
test: index_including_gist
+test: reindex_catalog
test: create_aggregate
test: create_function_3
test: create_cast
\set VERBOSITY default
DROP TABLE reindex_verbose;
---
--- check that system tables can be reindexed
---
-
--- whole tables
-REINDEX TABLE pg_class; -- mapped, non-shared, critical
-REINDEX TABLE pg_index; -- non-mapped, non-shared, critical
-REINDEX TABLE pg_operator; -- non-mapped, non-shared, critical
-REINDEX TABLE pg_database; -- mapped, shared, critical
-REINDEX TABLE pg_shdescription; -- mapped, shared non-critical
-
--- Check that individual system indexes can be reindexed. That's a bit
--- different from the entire-table case because reindex_relation
--- treats e.g. pg_class special.
-REINDEX INDEX pg_class_oid_index; -- mapped, non-shared, critical
-REINDEX INDEX pg_class_relname_nsp_index; -- mapped, non-shared, non-critical
-REINDEX INDEX pg_index_indexrelid_index; -- non-mapped, non-shared, critical
-REINDEX INDEX pg_index_indrelid_index; -- non-mapped, non-shared, non-critical
-REINDEX INDEX pg_database_oid_index; -- mapped, shared, critical
-REINDEX INDEX pg_shdescription_o_c_index; -- mapped, shared, non-critical
-
--
-- REINDEX CONCURRENTLY
--
--- /dev/null
+--
+-- Check that system tables can be reindexed.
+--
+-- Note that this test currently has to run without parallel tests
+-- being scheduled, as currently reindex catalog tables can cause
+-- deadlocks:
+--
+-- * The lock upgrade between the ShareLock acquired for the reindex
+-- and RowExclusiveLock needed for pg_class/pg_index locks can
+-- trigger deadlocks.
+--
+-- * The uniqueness checks performed when reindexing a unique/primary
+-- key index possibly need to wait for the transaction of a
+-- about-to-deleted row in pg_class to commit. That can cause
+-- deadlocks because, in contrast to user tables, locks on catalog
+-- tables are routinely released before commit - therefore the lock
+-- held for reindexing doesn't guarantee that no running transaction
+-- performed modifications in the table underlying the index.
+
+
+-- Check reindexing of whole tables
+REINDEX TABLE pg_class; -- mapped, non-shared, critical
+REINDEX TABLE pg_index; -- non-mapped, non-shared, critical
+REINDEX TABLE pg_operator; -- non-mapped, non-shared, critical
+REINDEX TABLE pg_database; -- mapped, shared, critical
+REINDEX TABLE pg_shdescription; -- mapped, shared non-critical
+
+-- Check that individual system indexes can be reindexed. That's a bit
+-- different from the entire-table case because reindex_relation
+-- treats e.g. pg_class special.
+REINDEX INDEX pg_class_oid_index; -- mapped, non-shared, critical
+REINDEX INDEX pg_class_relname_nsp_index; -- mapped, non-shared, non-critical
+REINDEX INDEX pg_index_indexrelid_index; -- non-mapped, non-shared, critical
+REINDEX INDEX pg_index_indrelid_index; -- non-mapped, non-shared, non-critical
+REINDEX INDEX pg_database_oid_index; -- mapped, shared, critical
+REINDEX INDEX pg_shdescription_o_c_index; -- mapped, shared, non-critical