Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: 'alter session set sql_trace=true' bug or feature ?
VC wrote:
> Hello Holger,
>
> "Holger Baer" <holger.baer_at_science-computing.de> wrote in message
> news:bs9d7c$pvi$1_at_news.BelWue.DE...
>
>>VC wrote: >> >>Sorry if I missed something in this thread sofar, but what is actually
>>problem? If you turn on tracing, then your goal is not to measure elapsed
>>time for the complete query, but rather create a trace file with explain
>>waits etc.
>>So while an massivly increased response time might bother you, it doesn't
>>sql_trace useless.
See, that was my question. Until now I've only seen examples with wallclock timings. But when you say that tkprof reports them as well, I see your problem.
>
> Now, we also have a complex query that runs fro about 10 minutes. Enabling
> trace in order to obtain tkprof data makes it run close to an hour. Your
> recommendations ?
>
What should I recommend faced with an obvious bug? ;-) But since you asked so nicely, I repeated your steps on Windows with 9.2.0.4 first with sql_trace = true, then with event '10046 trace name context forever, level 12'. Strangely enough, when tracing is on, after each fetch there is a fairly high wait reported for 'SQL*NET message from client' which in turn amounts to about 87% of the total runtime.
This could be reproduced with pl/sql developer, so it seems not a sql*plus bug, but in the oracle kernel.
So with the current state of 9.2.0.4 your only chance seems to step back from 9.2.0.4 to any version that had not this problem (perhaps an extra tuning instance)?
Regards,
Holger Received on Tue Dec 23 2003 - 07:52:05 CST