Does Cloudberry support CREATE INDEX / REINDEX / DROP INDEX CONCURRENTLY?

[yanboer]

Hi,

I checked the documentation for the following statements:

• CREATE INDEX: https://cloudberry.apache.org/docs/sql-stmts/create-index
• DROP INDEX: https://cloudberry.apache.org/docs/sql-stmts/drop-index
• REINDEX: https://cloudberry.apache.org/docs/sql-stmts/reindex

According to the docs, only DROP INDEX mentions support for CONCURRENTLY.

However, in Apache Cloudberry 2.0.0-incubating, all three statements appear to accept the CONCURRENTLY syntax:

gpadmin=# select version();
                                                                                                    version
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 PostgreSQL 14.4 (Apache Cloudberry 2.0.0-incubating build 1) on x86_64-pc-linux-gnu, compiled by gcc (GCC) 11.5.0 20240719 (Red Hat 11.5.0-11), 64-bit compiled on Dec  2 2025 19:35:44 (with assert checking)
(1 row)

gpadmin=# create index CONCURRENTLY on test(name);
CREATE INDEX
gpadmin=# reindex index CONCURRENTLY "test_name_idx";
REINDEX
gpadmin=# drop index CONCURRENTLY "test_name_idx";
DROP INDEX
gpadmin=#

Could you please clarify:

1. Are CREATE INDEX CONCURRENTLY and REINDEX ... CONCURRENTLY officially supported in 2.0.0?
2. If so, are the documentation pages out of date and should they be updated?
3. Are there any known limitations or caveats for these commands, for example with AO/CO tables, partitioned tables, Partial indexes (WHERE predicate), or invalid indexes left behind after a failed concurrent build?

Thanks.

[18w]

Hi @yanboer,

From the code, yes, they are really implemented and not just accepted by the parser. In src/backend/commands/indexcmds.c (same in the 2.0.0-incubating and 2.2.0-incubating tags) CREATE INDEX CONCURRENTLY and REINDEX ... CONCURRENTLY run the usual PG phases, but the QD dispatches each phase to the segments separately (concurrentlyPhase, "dispatch Phase 2/3 of concurrently index build"). So the docs pages look outdated for CREATE INDEX and REINDEX.

Limitations I found in the same file:

• AO/CO tables: cannot create index on append-only table "..." concurrently. The comment says the AO/AOCS table AM doesn't implement index_validate_scan, which phase 3 needs. There's a separate error for unique indexes on AO built concurrently.
• Partitioned tables: cannot create index on partitioned table "..." concurrently (same as PG 14). For REINDEX CONCURRENTLY, partitioned tables/indexes hit cannot reindex this type of relation concurrently, so you have to reindex the leaf partitions one by one.
• Invalid indexes: REINDEX CONCURRENTLY skips them with cannot reindex invalid index "..." concurrently, skipping, and the same for exclusion constraint indexes. A failed CIC leaves an invalid index behind like in PG: drop it (DROP INDEX CONCURRENTLY works) and build it again.
• Partial/expression indexes are supported. edb0f6e ("Fix waits of REINDEX CONCURRENTLY for indexes with predicates or expressions") fixed the waiting logic for them.
• System catalogs can't be reindexed concurrently.

One more thing since your build has assert checking: 867c6a1 on main ("Fix assert failure triggered by CREATE INDEX CONCURRENTLY", #1845) fixes an assertion failure in CIC when resource groups are enabled. Worth checking if your version has it before relying on CIC.

[yanboer]

My understanding is that CIC was inherited from PostgreSQL 14, but making it work in Cloudberry requires additional MPP-specific adaptations, such as proper dispatch logic. These adaptations appear to be in place in the codebase, while the documentation still does not reflect CIC support. Is this primarily a documentation gap, or is there still implementation or validation work required before CIC can be considered fully supported?