Re: "Null Event" Session Wait - hung Sessions

Martin-

The "null" event thing in v$session_wait is a known bug and I think will be fixed in a 92 patchset. My guess in 9.2.0.3 but not before. No one-offs will be created for this one, you will have to put in the patchset.

As for the apparent hang, try using hanganalyze, level 3 when it happens. There is a trace file generated that may seem cryptic at first, but will identify the probable blocker. From there you can proceed with state dumps to identify the specific resource in contention.

Basically, it works through session states, sees who is waiting and then chases up to the session on which it is waiting. The guy you want is not the blocked session, of course, but the mystery session who is doing the blocking. Hanganalyze helps find this bad guy. What makes the output confusing at first is the fact that many sessions may be blocked by one single session, or it may just be one session that is blocked. You get a sort of list of the possibilities.

Get hanganalyze this way, while the hang is occurring:

sqlplus /nolog
connect / as sysdba
oradebug setmypid
oradebug unlimit
oradebug hanganalyze 3

Then disconnect and go find the trace file in udump. Several may be created, depending on the level you run this, but one will be for your os pid and will contain the dependency analysis.

Practice the syntax on a test or dev machine first, so you can execute quickly when you need to. Plus, I am pulling the syntax off the top of my head, so you definitely want to test it. :-)

Once you have done this, if nothing obvious jumps out - like what was going on for the obvious blocker-candidate, then you can run the next time with level 5. Much more volume of output, but you get processstate dumps and errorstacks. The blocker gets is guts dumped to a trace file and you or support ought to be able to figure the cause out.

You probably want to work with support on this one, at least to get you through the hanganalyze trace analysis.

Happy hunting!

• ricky

Martin wrote:
>
> Hi,
>
> I am running a Production 9.2.0.1 Instance on AIX 64bit.
>
> I am encountering sessions on a frequent basis that appear be hung, not
> taking any apparent CPU nor I/O. All I can seem to see in V$SESSION_WAITS
> is wait time increasing against event "NULL Event".
>
> As anyone encountered this ? What does it mean ? I cannot find details
> about it's
> meaning or how to stop it. I have tryed dumping the trace for the session
> but nothing gets written to the dump directory - I have trying
> DBMS_SYSTEM.set_Ev
> and also ORADEBUG but get nought !
>
> These sessions are connecting via JDBC Thin driver - not sure if this is
> significant or not.
>
> Any help would be much appreciated.
>
> Regards
> Martin
Received on Sun Feb 09 2003 - 15:35:18 CST