IDS Forum

Temp Space Not Getting Freed

Posted By: Schmitz, Rob B [EQ]
Date: Monday, 6 October 2008, at 10:57 a.m.

IDS Gurus,

I am seeing some unusual behavior in several instances on a single server. We are running IDS 7.31.UD7 on AIX 5.2. This is a test server which our application testing group uses. The application in question runs a process that reads from the database and builds some sort of cache on the application server. Some of the reads are against views with several outer joins. These reads cause the process to use space in the temp dbspace. As the process proceeds, we can see the amount of free space fall in the temp dbspace - which should be normal. However, when the process completes, the temp space is not freed - even when there are no user sessions in the instance. I always assumed that temp space used by a session was freed at the end of a transaction - at the very least when the session disconnects. All databases involved have unbuffered logging.

The application was experiencing incomplete caches on the application server because the temp space would be completely consumed. We worked around the incomplete cache issue by adding space to the temp dbspace. This provided enough space for the cache to complete, but the problem of not freeing the temp space remains.

I tried to force a checkpoint, perform a fake Level 0 archive, move to the next logical log, and several combinations of these things. Nothing freed up the space unless I bounced the instance. Obviously, this is not a solution for the application.

There are no messages in the online.log file, other than the normal checkpoints.

I should also mention that this instance is involved in Enterprise Replication; however, the read operations only go against one instance and it is the "secondary", or receiving side, of all the replicates. ER should not come into play with this, but I thought I'd best mention it.

One more thing to mention: This server has eight instances. There are four pairs of instances, with each pair comprising a testing environment. Each pair has a "master" or "primary" instance, and a "secondary" or "regional" instance with four regional databases. The masters are named ma1, ma2, ma3, and ma4. The regionals are named reg1, reg2, reg3, and reg4. These read operations are performed only on the regional databases. The application folks first noticed incomplete caches when reading from reg2. The databases in reg2 had recently been refreshed from production so they initially thought it was a bad refresh. However, the data in the regional databases proved to be complete and accurate. We then noticed that other testing environments, which had not been refreshed for some time, were also exhibiting this same behavior. This exonerates the very recent growth in data volume we have seen. Whatever is affecting testing environment 2 now affects all four environments. This leads me to believe it might be a server level issue.

Has anyone seen this sort of behavior?

Thanks in advance for any help and advice.

Rob Schmitz
Embarq Data Management
rob.b.schmitz@embarq.com

Messages In This Thread

• Temp Space Not Getting Freed
  Schmitz, Rob B [EQ] -- Monday, 6 October 2008, at 10:57 a.m.
  • Re: Temp Space Not Getting Freed
    DAVE GRIFFEN -- Tuesday, 7 October 2008, at 3:05 p.m.
    • RE: Temp Space Not Getting Freed
      Schmitz, Rob B [EQ] -- Monday, 13 October 2008, at 5:50 p.m.

IDS Forum is maintained by Administrator with WebBBS 5.12.