Improve handling of dropped relations for REINDEX DATABASE/SCHEMA/SYSTEM
authorMichael Paquier <[email protected]>
Wed, 2 Sep 2020 00:08:12 +0000 (09:08 +0900)
committerMichael Paquier <[email protected]>
Wed, 2 Sep 2020 00:08:12 +0000 (09:08 +0900)
commit1d65416661bbb0b165865a521ce038ffb61b12ad
tree2b5bb3b6cb60419afef36aeb88fb2462eee52009
parent4c51a2d1e4b750bc11b8de9a85b079a14f798741
Improve handling of dropped relations for REINDEX DATABASE/SCHEMA/SYSTEM

When multiple relations are reindexed, a scan of pg_class is done first
to build the list of relations to work on.  However the REINDEX logic
has never checked if a relation listed still exists when beginning the
work on it, causing for example sudden cache lookup failures.

This commit adds safeguards against dropped relations for REINDEX,
similarly to VACUUM or CLUSTER where we try to open the relation,
ignoring it if it is missing.  A new option is added to the REINDEX
routines to control if a missed relation is OK to ignore or not.

An isolation test, based on REINDEX SCHEMA, is added for the concurrent
and non-concurrent cases.

Author: Michael Paquier
Reviewed-by: Anastasia Lubennikova
Discussion: https://postgr.es/m/20200813043805[email protected]
src/backend/access/table/table.c
src/backend/catalog/index.c
src/backend/commands/indexcmds.c
src/include/access/table.h
src/include/nodes/parsenodes.h
src/test/isolation/expected/reindex-schema.out[new file with mode: 0644]
src/test/isolation/isolation_schedule
src/test/isolation/specs/reindex-schema.spec[new file with mode: 0644]