Re: ERROR: found multixact from before relminmxid

On 04/06/2018 04:29 PM, Alexandre Arruda wrote:
> 2018-04-06 9:39 GMT-03:00 Tomas Vondra <tomas.vondra@2ndquadrant.com
> <mailto:tomas.vondra@2ndquadrant.com>>:

>
>
>
>     On 04/06/2018 02:09 AM, Alexandre Arruda wrote:
>     > Hi,
>     >
>     > Some time ago, I had this errors frequently showed in logs after some
>     > autovacuum in some tables(pg 9.6). VACUUM FULL or CLUSTER in this tables
>     > show the same and not complete the tasks (showed by some table bloat
>     > select).
>     > Then, I did a full dump/restore into a new version (10.2) and everything
>     > is ok for a couple of months. Now, I have errors like this again:
>     >
>     > db1=# cluster pc44t;
>     >
>     > ERROR:  found multixact 134100944 from before relminmxid 192042633
>     >
>     >  
>     >
>     > Like before, the only way to make the errors to desapear is by
>     > dump/reload the whole table.
>     >
>     >
>     > Thanks for any help.
>     >
>
>     That's going to be hard, unless you still have the cluster around.
>
>     This surely seems like some sort of data corruption issue, but without
>     being able to inspect the data pages it's nearly impossible to determine
>     what went wrong.
>
>     We'd also need more information about what happened to the hardware and
>     cluster before the issues started to appear - crashes, hardware issues.
>     And so on.
>
>     regards
>
>     --
>     Tomas Vondra                  http://www.2ndQuadrant.com
>     PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
>
>
> Hi Tomas,
> The old cluster are gone, unfortunatly.
>
> This server is a 128GB ECC RAM with a dual redundant hba fiber channel
> connect to a sotorage with Raid 6 and I don't have (apparently) any
> errors reported.
> Yesterday I did a test with one table:  some sum aggragates, count(*),
> etc, then dump/reload and repeat the tests the results (of querys) are
> the same, regarding the vacuum problem
> thats disapeared.
>

I'm not sure I understand correctly. So you can reproduce the issue? If
yes, how can you share the scripts/data you use (and how large is it)?
If we could reproduce it locally, it would make the investigation much
easier.

BTW you mentioned you're using PostgreSQL 9.6 - which minor version,
specifically?

regards