Re: pg_walinspect - a new extension to get raw WAL data and WAL stats

At Wed, 16 Mar 2022 20:49:12 +0530, Ashutosh Sharma <ashu(dot)coek88(at)gmail(dot)com> wrote in
> I can see that the pg_get_wal_records_info function shows the details
> of the WAL record whose existence is beyond the user specified
> stop/end lsn pointer. See below:
>
> ashu(at)postgres=# select * from pg_get_wal_records_info('0/01000028',
> '0/01000029');
> -[ RECORD 1 ]----+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> start_lsn | 0/1000028
> end_lsn | 0/100009F
> prev_lsn | 0/0
...
> record_length | 114
...
> In this case, the end lsn pointer specified by the user is
> '0/01000029'. There is only one WAL record which starts before this
> specified end lsn pointer whose start pointer is at 01000028, but that
> WAL record ends at 0/100009F which is way beyond the specified end
> lsn. So, how come we are able to display the complete WAL record info?
> AFAIU, end lsn is the lsn pointer where you need to stop reading the
> WAL data. If that is true, then there exists no valid WAL record
> between the start and end lsn in this particular case.

You're right considering how pg_waldump behaves. pg_waldump works
almost the way as you described above. The record above actually ends
at 1000099 and pg_waldump shows that record by specifying -s 0/1000028
-e 0/100009a, but not for -e 0/1000099.

# I personally think the current behavior is fine, though..

It still suggests unspecifiable end-LSN..

> select * from pg_get_wal_records_info('4/4B28EB68', '4/4C000060');
> ERROR: cannot accept future end LSN
> DETAIL: Last known WAL LSN on the database system is 4/4C000060.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center