Re: 10.1.0.3 RAC on Solaris 8-9

From: Yong Huang <yong321_at_yahoo.com>
Date: 5 Oct 2004 08:50:04 -0700
Message-ID: <b3cb12d6.0410050750.3c50a29e@posting.google.com>

"Alexey Sergeyev" <saefido7_at_devexperts.com> wrote in message news:<cjr5o6$170d$1_at_news.rtcomm.ru>...

>...

> Good things - CRS is going to evicte the problem node. Now it has to
> synchronize the cluster:
>
> 2004-09-28 13:21:50.729 [8] >TRACE: clssnmDoSyncUpdate: Initiating sync 3
> 2004-09-28 13:21:50.729 [4] >TRACE: clssnmHandleSync: Acknowledging sync:
> src[1] seq[10] sync[3]
> 2004-09-28 13:21:51.208 [1] >USER: NMEVENT_SUSPEND [00][00][00][02]
>
> Here Oracle totally hangs. What is going on in these 5 minutes?????
>
> 2004-09-28 13:26:55.749 [8] >WARNING: clssnmWaitOnEvictions: Unconfirmed
> dead node count 1
> 2004-09-28 13:26:55.750 [4] >USER: clssnmHandleUpdate: SYNC(3) from
> node(1) completed
> 2004-09-28 13:26:55.750 [4] >USER: clssnmHandleUpdate: NODE(1) IS ACTIVE
> MEMBER OF CLUSTER
> 2004-09-28 13:26:56.330 [14] >USER: NMEVENT_RECONFIG [00][00][00][02]

>...

> It looks like new 300 sec. timeout was introduced in 10.1.0.3, but where can
> i change it?

Just a guess. Metalink Note:220970.1 talks about VIP and says "Without using VIPs, clients connected to a node that died will often wait for a TCP/IP timeout period before getting an error. This can be as long as 10 minutes or more." Maybe that 10 minutes is 5 minutes in your case. If this is not configured anywhere under $ORACLE_HOME, it may be controlled by one of your system TCP parameters. I think you're using Solaris. For your convenience, run this

for i in $(ndd /dev/tcp \? | awk '{print $1}'); do   echo $i; ndd /dev/tcp $i
done > nddoutput

Look at nddoutput to see if there's anything close to 5 minutes plus 4 seconds (304000 ms).

As I said, it's a guess. I suggest you open a Tar with Oracle and post back when you find the answer. We'll all benefit. Thanks.

Yong Huang Received on Tue Oct 05 2004 - 10:50:04 CDT