[Pgpool-general] one of backends kicked after "VACUUM ANALYZE"

[Tomasz Chmielewski]

On 20.05.2011 14:32, Tatsuo Ishii wrote:
>>>> Are there any other commands which should not be executed through
>>>> pgpool? For example, is REINDEX TABLE safe to execute through pgpool?
>>>
>>> pg_dump/pg_dumpall. REINDEX is safe.
>>
>> Unfortunately, it died for me during REINDEX:
>>
>> ERROR:  kind mismatch among backends. Possible last query was: "REINDEX TABLE core_multistats;" kind details are: 0[E: deadlock detected] 1[C]
>> HINT:  check data consistency among db nodes
>> server closed the connection unexpectedly
>> 	This probably means the server terminated abnormally
>> 	before or while processing the request.
>> connection to server was lost
>>
>>
>> So after all, looks REINDEX is also not so safe?
>
> According to the log there's something wrong going onb the master
> PostgreSQL(deadlock?) I think it's not related to pgpool-II. Can you
> find something related to the problem in the PostgreSQL log?

Not really - although internet search for "postgres deadlock detected 
reindex" shows quite a bit of hits - so I guess that would depend on 
database usage.

Putting the question the other way: is REINDEX TABLE safe, if made 
directly on database backends, without going through pgpool? As I 
understand, it does not change database data visible to the clients, so 
should be safe.


-- 
Tomasz Chmielewski
http://wpkg.org