When running:


which could take several hours, or even days, depending on the size of the database, is there anyway to get a rough estimate of its progress?

I've seen some forum posts claiming you can query index creation status with a query like:

FROM pg_stat_progress_create_index p 
JOIN pg_stat_activity a ON p.pid = a.pid;

The _done/_total columns in combination with phase does provide a rough progress percent. However, this only lists the progress of the currently updating index. It doesn't tell you how many other indexes are pending update, much less how much work there is to do for each.

Edit: I've tried combining the views pg_index, which lists the *_ccnew temporary indexes used by the concurrent process, with pg_stat_progress_create_index like:

SELECT relname,
CASE WHEN blocks_total > 0 THEN (ci.blocks_done/ci.blocks_total::numeric*100)::int ELSE NULL END as blocks_percent,
FROM pg_class as pgc
inner join pg_index as i on i.indexrelid = pgc.oid
left outer join pg_stat_progress_create_index as ci on ci.index_relid = i.indexrelid
WHERE i.indisvalid = false;

but this shows strange results. For my database, it lists ~300 indexes in pg_index that are temporary, and waiting to be updated. However, the one index cross referenced by pg_stat_progress_create_index that updates never is marked valid. It gets to 100% of blocks processed, and then disappears from pg_stat_progress_create_index but its indisvalid stays false. Why is this?

3 Answers 3


After reviewing the columns in pg_index, it looks like Postgres uses indisvalid = false to denote all the indexes being rebuilt and the subset of those with indisready = true denotes the indexes that pg_stat_progress_create_index has processed.

Using that, I think I can calculate a total progress percent with a query like:

    (m.complete_steps/m.total_steps::numeric*100)::int AS total_percent
    MAX(CASE WHEN blocks_total > 0 THEN (ci.blocks_done/ci.blocks_total::numeric*100)::int ELSE NULL END) AS step_percent,
    COUNT(*) AS total_steps,
    COUNT(CASE WHEN i.indisready THEN 1 ELSE NULL END) AS complete_steps
    FROM pg_class AS pgc
    INNER JOIN pg_index AS i ON i.indexrelid = pgc.oid
    LEFT OUTER JOIN pg_stat_progress_create_index AS ci ON ci.index_relid = i.indexrelid
    WHERE i.indisvalid = false
) AS m

REINDEX DATABASE command has no visible progress reporting in terms of overall work. Here is relevant code in postgresql source tree. REINDEX DATABASE will iterate over tables list and call (literally) reindex table for each individual relation in separate transaction. There is no progress report on how many relations have already been processed and how big the remaining list is.

The statuses in pg_stat_progress_create_index are reported by the cyclically called reindex_relation. So, it is useful from the point of view of processing a single index, but does not have information about the total amount of work across the entire database.


Postgres 12 or later has the system view pg_stat_progress_create_index.
It reports ...

One row for each backend running CREATE INDEX or REINDEX, showing current progress.

Find details in the chapter CREATE INDEX Progress Reporting of the manual.

This is potentially very expensive on a busy server!


The manual has a chapter CREATE INDEX Phases describing states of the column phase. In particular, consider:

waiting for old snapshots

CREATE INDEX CONCURRENTLY or REINDEX CONCURRENTLY is waiting for transactions that can potentially see the table to release their snapshots. This phase is skipped when not in concurrent mode. Columns lockers_total, lockers_done and current_locker_pid contain the progress information for this phase.

Long-running transactions can stall the progress. Consider running REINDEX CONCURRENTLY on selected indexes. And REINDEX DATABASE at off-hours when you can afford to lock tables exclusively.

If your server is not actually busy, check for long-running transactions with:

SELECT * FROM pg_stat_activity;

The ones with state = 'idle in transaction' are the prime troublemakers (typically hint at a programming error, where transactions are not committed or rolled back). Those would likely show up in pg_stat_progress_create_index.current_locker_pid of stalled indexes.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.