Re: Wrong statistics for size of XLOG_SWITCH during pg_waldump. - Mailing list pgsql-hackers

From movead.li@highgo.ca
Subject Re: Wrong statistics for size of XLOG_SWITCH during pg_waldump.
Date
Msg-id 2020101512560020200319@highgo.ca
Whole thread Raw
In response to Wrong statistics for size of XLOG_SWITCH during pg_waldump.  ("movead.li@highgo.ca" <movead.li@highgo.ca>)
List pgsql-hackers
Thanks for all the suggestions.

>Yeah.  In its current shape, it means that only pg_waldump would be
>able to know this information.  If you make this information part of
>xlogdesc.c, any consumer of the WAL record descriptions would be able
>to show this information, so it would provide a consistent output for
>any kind of tools.
I have change the implement, move some code into xlog_desc().

>On top of that, it seems to me that the calculation used in the patch
>is wrong in two aspects at quick glance:
>1) startSegNo and endSegNo point always to two different segments with
>a XLOG_SWITCH record, so you should check that ReadRecPtr is not at a
>segment border instead before extracting SizeOfXLogLongPHD, no?
Yes you are right, and I use 'record->EndRecPtr - 1' instead.

>2) This stuff should also check after the case of a WAL *page* border
>where you'd need to adjust based on SizeOfXLogShortPHD instead.
The 'junk_len -= SizeOfXLogLongPHD' code is considered for when the
remain size of a wal segment can not afford a XLogRecord struct for
XLOG_SWITCH, it needn't care *page* border.

>I'm not sure the exact motive of this proposal, but if we show the
>wasted length in the stats result, I think it should be other than
>existing record types.
Yes agree, and now it looks like below as new patch:

movead@movead-PC:/h2/pg/bin$ ./pg_waldump -p ../walbk/ -s 0/3000000 -e 0/6000000 -z
Type                                           N      (%)          Record size      (%)             FPI size      (%)        Combined size      (%)
----                                           -      ---          -----------      ---             --------      ---        -------------      ---
XLOG                                           5 ( 31.25)                  300 (  0.00)                    0 (  0.00)                  300 (  0.00)
XLOGSwitchJunk                                 3 ( 18.75)             50330512 (100.00)                    0 (  0.00)             50330512 (100.00)


movead@movead-PC:/h2/pg/bin$ ./pg_waldump -p ../walbk/ -s 0/3000000 -e 0/6000000 --stat=record
XLOG/SWITCH                                    3 ( 18.75)                   72 (  0.00)                    0 (  0.00)                   72 (  0.00)
XLOG/SWITCH_JUNK                               3 ( 18.75)             50330512 (100.00)                    0 (  0.00)             50330512 (100.00)

The shortcoming now is I do not know how to handle the 'count' of SWITCH_JUNK
in pg_waldump results. Currently I regard SWITCH_JUNK as one count.

>By the way, I noticed that --stats=record shows two lines for
>Transaction/COMMIT. The cause is that XLogDumpCountRecord assumes the
>all four bits of xl_info is used to identify record id.
Thanks,I didn't notice it before, and your patch added into v3 patch attached.



Regards,
Highgo Software (Canada/China/Pakistan)
URL : www.highgo.ca
EMAIL: mailto:movead(dot)li(at)highgo(dot)ca
Attachment

pgsql-hackers by date:

Previous
From: Greg Nancarrow
Date:
Subject: Re: Parallel INSERT (INTO ... SELECT ...)
Next
From: Masahiko Sawada
Date:
Subject: Re: Add Information during standby recovery conflicts