Skip site navigation (1) Skip section navigation (2)

Re: index corruption on composite primary key indexes

From: Mikael Krantz <mk(at)zigamorph(dot)se>
To: "Ng, Stan" <sng(at)automotive(dot)com>
Cc: Craig Ringer <craig(at)postnewspapers(dot)com(dot)au>, pgsql-bugs(at)postgresql(dot)org
Subject: Re: index corruption on composite primary key indexes
Date: 2010-12-15 00:35:38
Message-ID: (view raw, whole thread or download thread mbox)
Lists: pgsql-bugs
I'm afraid that I'm a bit out of my depth when it comes to index corruption.

I did note a possibly unrelated problem: The unique index includes the
lower(postalcode::text) but your left join is made on
VehicleUsedICCGRadiusCache.PostalCode = t1.PostalCode. It seems likely
that you do this left join to exclude rows already present in
VehicleUsedICCGRadiusCache. If PostalCode can contain mixed case this
will be one possible source of duplicate key violations. Differently
cased PostalCodes will not be excluded by your left join but will be
considered identical by the unique index.

This might not at all be connected to your current problem as it seems
like something that would work or not work in a very consistent manner
regardless on reindexing but it could be useful anyway. Changing the
join condition to use lower() would also make the join able to use
more of the index for the join and could improve performance.

Best Regards

Mikael Krantz

On Tue, Dec 14, 2010 at 7:42 PM, Ng, Stan <sng(at)automotive(dot)com> wrote:
> Ah, I forgot to mention why I suspected corruption. The delta
> application is a single threaded, sequential process. Each delta is done
> within a single transaction. If it fails, there is retry logic that will
> continue to attempt the delta application forever, with about a 5 to 10
> minute wait. This will fail every time until a manual REINDEX command is
> run on the composite primary key index in-between retry attempts. I've
> also backed up the data to another table w/ the same indices and applied
> the delta data manually without any problems while this error was
> happening.
> Oddly enough, even with full logging on, the pgsql logs don't contain
> anything about the duplicate key error or suspect index. Perhaps another
> important piece of information is that we're doing all this in Java
> using the JDBC drivers, so we're seeing this as part of our stacktrace.
> i.e.
> org.postgresql.util.PSQLException: ERROR: duplicate key value violates
> unique constraint "vehicleusediccgradiuscache_i00"
>        at
> org.postgresql.core.v3.QueryExecutorImpl.receiveErrorResponse(QueryExecu
>        at
> org.postgresql.core.v3.QueryExecutorImpl.processResults(QueryExecutorImp
>        at
> org.postgresql.core.v3.QueryExecutorImpl.execute(
> 257)
>        at
> org.postgresql.jdbc2.AbstractJdbc2Statement.execute(AbstractJdbc2Stateme
>        at
> org.postgresql.jdbc2.AbstractJdbc2Statement.executeWithFlags(AbstractJdb
>        at
> org.postgresql.jdbc2.AbstractJdbc2Statement.execute(AbstractJdbc2Stateme
>          ...
> The index looks like this:
> CREATE UNIQUE INDEX vehicleusediccgradiuscache_i00
>  ON vehicleusediccgradiuscache
>  USING btree
>  (datasetid, lower(postalcode::text), productgroupid,
> vehicleclassgroupidic, vehicleusedidic, vehicleuseddisplaypriority,
> vehicleusedsearchradius);
> A single delta is calculated on the master server, which is pulled by
> each client server. Each client puts the delta data into temp tables
> (one each for delete, update, and insert, applied in that order), and
> modifies the data via SQL somewhat like the following:
> insert into
> VehicleUsedICCGRadiusCache(DatasetId,PostalCode,ProductGroupId,VehicleCl
> assGroupIdIC,VehicleUsedIdIC,VehicleUsedDisplayPriority,VehicleUsedSearc
> hRadius,"VehicleUsedPrice.average","VehicleUsedPrice.min","VehicleUsedPr
> ice.max","VehicleUsedIntId.distinct_count") select
> t1.DatasetId,t1.PostalCode,t1.ProductGroupId,t1.VehicleClassGroupIdIC,t1
> .VehicleUsedIdIC,t1.VehicleUsedDisplayPriority,t1.VehicleUsedSearchRadiu
> s,t1."VehicleUsedPrice.average",t1."VehicleUsedPrice.min",t1."VehicleUse
> dPrice.max",t1."VehicleUsedIntId.distinct_count" from t1
> left outer join VehicleUsedICCGRadiusCache on
> VehicleUsedICCGRadiusCache.VehicleUsedSearchRadius=t1.VehicleUsedSearchR
> adius and VehicleUsedICCGRadiusCache.ProductGroupId=t1.ProductGroupId
> and
> VehicleUsedICCGRadiusCache.VehicleUsedDisplayPriority=t1.VehicleUsedDisp
> layPriority and
> VehicleUsedICCGRadiusCache.VehicleUsedIdIC=t1.VehicleUsedIdIC and
> VehicleUsedICCGRadiusCache.DatasetId=t1.DatasetId and
> VehicleUsedICCGRadiusCache.PostalCode=t1.PostalCode and
> VehicleUsedICCGRadiusCache.VehicleClassGroupIdIC=t1.VehicleClassGroupIdI
> C
> where VehicleUsedICCGRadiusCache.VehicleUsedSearchRadius is null
> -----Original Message-----
> From: Craig Ringer [mailto:craig(at)postnewspapers(dot)com(dot)au]
> Sent: Tuesday, December 14, 2010 4:47 AM
> To: Ng, Stan
> Cc: pgsql-bugs(at)postgresql(dot)org
> Subject: Re: [BUGS] index corruption on composite primary key indexes
> On 12/14/2010 10:18 AM, Ng, Stan wrote:
>> Some info on the platform I'm using:
> Please also show your schema and query / queries. If you think it's
> corruption, provide your postgresql logs too, including any mention of
> the names of the suspect indexes.
> Personally I agree with Mikael Krantz; you haven't shown any evidence of
> index corruption, and you're most likely hitting concurrency problems in
> your code.
> --
> Craig Ringer
> --
> Sent via pgsql-bugs mailing list (pgsql-bugs(at)postgresql(dot)org)
> To make changes to your subscription:

In response to


pgsql-bugs by date

Next:From: BERTRAND JoelDate: 2010-12-15 11:19:35
Subject: Re: [8.4.4] Strange bus error on Solaris 10/sparc
Previous:From: Robert HaasDate: 2010-12-14 21:09:11
Subject: Re: BUG #5776: Unable to create view with parameter in PL/pgsql

Privacy Policy | About PostgreSQL
Copyright © 1996-2017 The PostgreSQL Global Development Group