Hi,

The pg_stat_progress_data_checksums docs says that databases_total and
databases_done are reported as NULL for workers, but I observed that
they are reported as 0 instead:

    =# select * from pg_stat_progress_data_checksums;
      pid  | datid | datname  |  phase   | databases_total |
databases_done | relations_total | relations_done | blocks_total |
blocks_done
    
-------+-------+----------+----------+-----------------+----------------+-----------------+----------------+--------------+-------------
     34873 |     0 | (null)   | enabling |               3 |
   0 |          (null) |         (null) |       (null) |      (null)
     34874 |     5 | postgres | enabling |               0 |
   0 |             297 |              0 |       163935 |         427
    (2 rows)

The cause is that pg_stat_progress_data_checksums uses -1 as the
sentinel value displayed as NULL, but pgstat_progress_start_command()
initializes all progress counters to zero. Data checksum progress did not
consistently reset those counters to -1, so some counters could
incorrectly appear as 0 instead of NULL.

The attached patch fixes this by initializing all data checksum progress
counters to -1 immediately after progress reporting starts for both
launcher and worker processes.

While looking at this code, I also found that blocks_done is not reset
when a worker starts processing a new relation fork. As a result,
blocks_done can temporarily exceed blocks_total, or a stale blocks_done
value can remain visible for an empty relation fork. The attached patch
resets blocks_done together with blocks_total when starting each
relation fork.

Regards,

-- 
Fujii Masao

Attachment: v1-0001-Fix-data-checksum-progress-counter-initialization.patch
Description: Binary data

Reply via email to