Re: WAL archive space planning?

[Ray Stell <stellr@xxxxxx>]

On 3/13/17 11:53 PM, Steven Chang wrote:

Dears,

     Just read a posgres monitoring moudule named check-postres,  which is a ready enhance package on Debian Linux Distribution.

     

   https://bucardo.org/check_postgres/check_postgres.pl.html

   

   here is its readme about wal_files

   Maybe you can also use this module to help plan your  space usage of WAL files.

 wal_files

       ("symlink: check_postgres_wal_files") Checks how many WAL files exist in the

       pg_xlog directory, which is found off of your data_directory, sometimes as a

       symlink to another physical disk for performance reasons. This action must be

       run as a superuser, in order to access the contents of the pg_xlog directory.

       The minimum version to use this action is Postgres 8.1. The --warning and

       --critical options are simply the number of files in the pg_xlog directory. What

       number to set this to will vary, but a general guideline is to put a number

       slightly higher than what is normally there, to catch problems early.

       Normally, WAL files are closed and then re-used, but a long-running open

       transaction, or a faulty archive_command script, may cause Postgres to create

       too many files. Ultimately, this will cause the disk they are on to run out of

       space, at which point Postgres will shut down.

2017-02-22 3:45 GMT+08:00 Keith <keith@xxxxxxxxxxx>:

On Tue, Feb 21, 2017 at 2:33 PM, Ray Stell <stellr@xxxxxx> wrote:

On 2/21/17 12:26 PM, Keith wrote:

On Tue, Feb 21, 2017 at 12:22 PM, Ray Stell <stellr@xxxxxx> wrote:

On 2/21/17 12:09 AM, Steven Chang wrote:

check this, his 2nd part introduces Stream Replication Implementation and tell you px_log retention and wal archive related parameters.

https://www.itenlight.com/blog/2016/05/19/PostgreSQL+HA+with+pgpool-II+-+Part+2

2017-02-18 2:30 GMT+08:00 Ray Stell <stellr@xxxxxx>:

I was "planning" to turn on WAL archiving on a postgresql 9.4.11 server that currently is running with "wal_level=hot_standby" and streaming to a standby.  I thought there would be a relationship between the rate of pg_xlog files and archive generation.   When I turned up the archive_command/mode I found the scale of the archive target was wrong as I had based it on the pg_xlog file creation rate.

When I turned on the archive command for a few minutes, pg_xlog dir contained these files for the time period:

-rw------- 1 postgres postgres 16777216 Feb 17 06:47 00000001000023AC0000007F
-rw------- 1 postgres postgres 16777216 Feb 17 06:45 00000001000023AC0000007E

The archive command wrote 126, 16MB files:

-rw------- 1 postgres postgres 16777216 Feb 17 06:47 000000010000237700000056
-rw------- 1 postgres postgres 16777216 Feb 17 06:47 000000010000237700000055
-rw------- 1 postgres postgres 16777216 Feb 17 06:47 000000010000237700000054
-rw------- 1 postgres postgres 16777216 Feb 17 06:47 000000010000237700000053
...

-rw------- 1 postgres postgres 16777216 Feb 17 06:45 0000000100002376000000DC
-rw------- 1 postgres postgres 16777216 Feb 17 06:45 0000000100002376000000DB
-rw------- 1 postgres postgres 16777216 Feb 17 06:45 0000000100002376000000DA
-rw------- 1 postgres postgres 16777216 Feb 17 06:45 0000000100002376000000D9

On servers that are not nearly as busy, I observe a one-to-one relationship between these files/rates.

Is there a good WAL archive space planning guide?

Is the way to collect planning data for this to turn on wal_debug?

TIA!




--
Sent via pgsql-admin mailing list (pgsql-admin@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-admin

I'm interested in why I might observed a dramatic difference between pg_xlog files and the archive target files. Other dbs I see have them pretty much one-to-one, but not in this case here.  What might cause the large variation?  I got only a few pg_xlogs files each minute and 42 archive files/min.  

Ray,

Did you see my previous response about the archive_timeout setting?

Lost your post somehow, but I see it in the list archive.  Thanks, Keith.

There was a 30 minute timeout set, but I would not think that would increase the archive file generation as observed, does it? 

" When this parameter is greater than zero, the server will switch to a new segment file whenever this many seconds have elapsed since the last segment file switch, and there has been any database activity, including a single checkpoint. (Increasing checkpoint_timeout will reduce unnecessary checkpoints on an idle system.) "

Your post seems to indicate having a non-zero value might increase the number of archive files, is that so?  Maybe it is in milliseconds which might make sense with the observed archive rate, but the doc says seconds and it was set to 1800.  I can set it to 0 and see if it changes things, but I'm a little confused.

If it was set to 30 minutes, then no, it wouldn't explain what you saw. But we didn't know what it was set to, so that's why I asked. Make sure you're checking the value actually set in the database, and not just what's in postgresql.conf, since it could've changed but never been reloaded to put in place (type "show archive_timeout" in psql).

If that's not it, not sure what may be causing what you're seeing.

Keith

Thanks, Steven,

An interesting statement:

"long-running open transaction, or a faulty archive_command script, may cause Postgres to create too many files. Ultimately, this will cause the disk they are on to run out of space, at which point Postgres will shut down."  

The statement, "long-running open transaction...may cause Postgres to create too many files" makes it sound like there could be the case where there is not a 1-to-1 relationship between pg_xlog and the archive.  I've never seen that before and didn't think that was the case.  Maybe I'm reading too much into the statement. 

I have to suspect that the "faulty" nature of the archive_command was that it was just asking too much of the I/O subsystem and was just getting way behind.  Shockingly way behind.  I didn't expect it.   It is a very busy system will many long running transactions.  If I ever get back to that problem, my first pass will be to reduce the load request.  I suppose I can do some bonnie tests to simulate the archive writing.  I was doing some multiplexing of the files in the archive shell.  Let's see what writing them once locally does first.  If it can't do that and keep up, I'm going to need a "bigger boat."   It's a production system so I only get a few cracks at it a year.   I'll probably wait for the next patch release.  Going back to the last backup is within the SLA, but I don't like not having the archive.