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

On Fri, Mar 11, 2022 at 8:22 AM Kyotaro Horiguchi
<horikyota(dot)ntt(at)gmail(dot)com> wrote:
> >
> - The difference between pg_get_wal_record_info and _records_ other than
> - the number of argument is the former accepts incorrect LSNs.
>
> The discussion is somewhat confused after some twists and turns.. It
> should be something like the following.
>
> pg_get_wal_record_info and pg_get_wal_records_info are almost same
> since the latter can show a single record. However it is a bit
> annoying to do that. Since, other than it doens't accept same LSNs for
> start and end, it doesn't show a record when there' no record in the
> specfied LSN range. But I don't think there's no usefulness of the
> behavior.

I would like to reassert the usability of pg_get_wal_record_info and
pg_get_wal_records_info:

pg_get_wal_record_info(lsn):
if lsn is invalid i.e. '0/0' - throws an error
if lsn is future lsn - throws an error
if lsn looks okay, it figures out the next available valid WAL record
and returns info about that

pg_get_wal_records_info(start_lsn, end_lsn default null) -> if start
and end lsns are provided no end_lsn would give the WAL records info
till the end of WAL,
if start_lsn is invalid i.e. '0/0' - throws an error
if start_lsn is future lsn - throws an error
if end_lsn isn't provided by the user - calculates the end_lsn as
server's current flush lsn
if end_lsn is provided by the user - throws an error if it's future LSN
if start_lsn and end_lsn look okay, it returns info about all WAL
records from the next available valid WAL record of start_lsn until
end_lsn

So, both pg_get_wal_record_info and pg_get_wal_records_info are necessary IMHO.

Coming to the behaviour when input lsn is '0/1000000', it's an issue
with XLogSegmentOffset(lsn, wal_segment_size) != 0 check, which I will
fix in the next version.

if (*first_record != lsn && XLogSegmentOffset(lsn, wal_segment_size) != 0)
ereport(WARNING,
(errmsg_plural("first record is after %X/%X, at %X/%X,
skipping over %u byte",
"first record is after %X/%X, at %X/%X,
skipping over %u bytes",
(*first_record - lsn),
LSN_FORMAT_ARGS(lsn),
LSN_FORMAT_ARGS(*first_record),
(uint32) (*first_record - lsn))));

Regards,
Bharath Rupireddy.