|
Thank you for a very thoughtful and detailed reply, James. Hopefully
we'll get this one figured out.
Drew
Drew,
Performance problems are notoriously
difficult to diagnose, especially remotely. Remember too that the
benchmarks you are thinking of are for optimally configured systems running in
the lab, not real-world results. But here's a couple of points you
might investigate.
(1) What you
see in trapd.log is not necessarily what is coming in. It's what
trapd.processed and logged. Logging is the last thing trapd does with
the trap, after he's processed it in every other way. It does record
that a particular trap was received and processed at a particular time, but
that's about all. So seeing Cisco traps in trapd.log 5 seconds apart
means that's how fast trap is processing them, not how fast they are arriving.
What might you not see in the log? Any traps configured to "Don't
Log or Display" in xnmtrap. That action puts the trap category in
trapd.conf to "Ignore". So you could go to /usr/OV/conf/C (don't
forget the "C" here) and do "grep Ignore trapd.conf " and see whether you have
any of those. If you do, then you are not seeing those in the log.
For diagnostic purposes you should alter those entries to "Log Only" so
you can get a better idea of the work trapd is actually doing.
(2) To get closer to what is coming in,
you could turn on the trapd.trace. You'll see a message about each
trap being received from address so-and-so every time one is pulled off
the queue for processing. If you want to see the contents of those
incoming traps, then you also need to have trapd running with the -x option to
hex dump an incoming packets. Now I said closer to what is coming in,
because obviously trapd will cannot trace a trap until he has started to read
it. When won't he read immediately? When there is no break between
incoming traps. If traps arrive too quickly, rather than pull them
off one at a time and process them, trapd queues them so that he doesn't lose
any. He won't start processing them again until there's a break in the
incoming flow. In that case you should see a bunch of trap queued
messages but no intervening processing in the trace. I suspect that this
is really what's going on. You get a big burst of traps, so all trap
processing slows while we queue them, and then once the burst subsides,
processing starts up again. But now the bottle neck is going to be in
nvcorrd and nvserverd, who have been idle for awhile, and now have a lot to
do. It's like a snake swallowing an egg; you see a big lump moving
along until it is totally digested. You have to turn on the nvcorrd
trace (nvcdebug -d all) to see what nvcorrd's doing, and one benefit of that
is that you can see how long it takes him to process just one trap, given the
rulesets and event windows you have going at the time. Look for the
eye-catcher "Received a trap" and "Finished with the trap". From the one
to the other is the transit time through nvcorrd. Not much you can do
if you don't like it, other than to reduce the load.
(3) Obviously if you want to assess what the real
incoming trap rate is, you need an outside analysis tool, such as an iptrace
for port 162. Then you can use ipreport of the data and see. Those are
AIX commands by the way -- there are similar tools on Solaris and Linux but I
haven't used them much.
(4)
If you cannot reduce the incoming rates to keep processing from being
overloaded then you might consider installing an MLM and using it as a trap
filter, tossing out duplicates and only passing on to trapd what you really
want to see.
HTH
James Shanks
Level 3 Support
for Tivoli NetView for UNIX and Windows
Tivoli Software / IBM
Software Group
"Van Order, Drew \(US -
Hermitage\)" <dvanorder@deloitte.com>
Sent by: owner-nv-l@lists.us.ibm.com
09/17/2004 10:12 AM
|
|
To
| <nv-l@lists.us.ibm.com>
|
|
cc
|
|
|
Subject
| RE: [nv-l] nvtecia
still hanging or falling behind processing
TEC_ITS.rs |
|
James,
We finally had missed heartbeats to
track. We can see the heartbeat trap in trapd.log, but no corresponding entry
in nvserverd. This appears to confirm the holdup is on the NV side, and again,
we had an increase in Cisco traps (one every 5 seconds for about 2 hours prior
to missing the first heartbeat), but nothing near NV's limit. Trapd.log shows
it is starting to fall behind as well during this period--as an example, the
missed heartbeat TEC event for 6 PM last night did not show in trapd until
6:48 PM. The 7 PM heartbeat shows in trapd at 7:21 PM and is in nvserverd at
7:45, so it had almost caught up by then.
So the TEC
adapter never stopped, but we've got to figure out why trapd and the processes
in between seem to stumble under load, but not a heavy one. We know Cisco
devices can send some traps at rates faster than one per second. Is it
possible devices are machine gunning traps even though NV shows one every 5
seconds or so? That's the only thing I can think of that could set trapd
behind based on what we are seeing.
Thanks everyone--Drew
-----Original Message-----
From:
owner-nv-l@lists.us.ibm.com [mailto:owner-nv-l@lists.us.ibm.com] On Behalf
Of James Shanks
Sent: Thursday, September 16, 2004 3:40
PM
To: nv-l@lists.us.ibm.com
Subject: RE: [nv-l] nvtecia
still hanging or falling behind processing TEC_ITS.rs
So what's different? Is your wpostemsg to
@EventServer like your tecint.conf file?
We are back to this being a TEC issue
and not a NetView one. So unless you want to open a problem to TEC
support, you'll have to do some more detective work yourself.
If both the wpostemsg and
the tecint.conf have @EventServer, then I don't know what to tell you.
If not, then reconfigure your tecint.conf using serversetup to use the
non-TME method (which requires that a different daemon be started than when
you use the TME method). For non-TME forwarding, /usr/OV/bin/nvserverd
is started. For TME forwarding, it is /usr/OV/bin/spmsur, who then
starts /usr/OV/bin/tme_nvserverd. To which from one to the other
requires that you go through serversetup, which will reconfigure this
automatically, or that you manually alter the /usr/OV/conf/ovsuf file to
start the correct daemons. But note that when you go through
serversetup, your special customization to the Nvserverd entries is lost.
The fact that events are going to the cache means that nvserverd
got the event, formatted it, did his tec_put_event( ) and all went fine, but
then TEC library code in trying to send to the TEC server found that it could
not, that it has lost connection to the TEC server, for some reason known only
to those internal routines. And without a diag (as in "diagnosis") file
configured in here so that the internal TEC library code will trace itself, no
one can tell you what it's doing or why. And you have to get that diag
file, called ".ed_diag_config" from TEC Support and they are the ones who have
to look at the traces. No one on the NetView side can assist at this
point.
James
Shanks
Level 3 Support for Tivoli NetView for UNIX and
Windows
Tivoli Software / IBM Software Group
"Edwards, JT - ESM"
<JEdwards3@wm.com>
Sent by:
owner-nv-l@lists.us.ibm.com
09/16/2004 04:00 PM
|
|
To
| "'nv-l@lists.us.ibm.com'"
<nv-l@lists.us.ibm.com>
|
|
cc
|
|
|
Subject
| RE: [nv-l] nvtecia
still hanging or falling behind processing TEC
_ITS.rs |
|
Yes it
does.
-----Original
Message-----
From: owner-nv-l@lists.us.ibm.com
[mailto:owner-nv-l@lists.us.ibm.com]On Behalf Of James
Shanks
Sent: Thursday, September 16, 2004 2:32 PM
To:
nv-l@lists.us.ibm.com
Subject: RE: [nv-l] nvtecia still hanging or
falling behind processing TEC _ITS.rs
Wpostemsg does not go through the internal
adapter. Does that get to the TEC server?
James Shanks
Level 3 Support
for Tivoli NetView for UNIX and Windows
Tivoli Software / IBM
Software Group
"Edwards, JT - ESM"
<JEdwards3@wm.com>
Sent by:
owner-nv-l@lists.us.ibm.com
09/16/2004 03:17 PM
|
|
To
| "'nv-l@lists.us.ibm.com'"
<nv-l@lists.us.ibm.com>
|
|
cc
|
|
|
Subject
| RE: [nv-l] nvtecia
still hanging or falling behind processing TEC
_ITS.rs |
|
Well at this
point. We are now getting events caching. From there what can we
do?
A
wpostemsg does not clear the cache.
-----Original Message-----
From:
owner-nv-l@lists.us.ibm.com [mailto:owner-nv-l@lists.us.ibm.com]On Behalf
Of James Shanks
Sent: Wednesday, September 15, 2004 10:16
PM
To: nv-l@lists.us.ibm.com
Subject: RE: [nv-l] nvtecia
still hanging or falling behind processing TEC _ITS.rs
No. The errno 827 indicates that there is a problem
initializing the JVM -- Java Virtual Machine. In almost every case I have seen
this indicates that the nvserverd daemon does not have the correct library
path for Java or the ZCE_CLASSPATH variable is not set. Since it is only set
in /etc/netnmrc, if you ovstop all the daemons and restart them with just
ovstart, you will lose it. So Mike is right. The usual fix is to ovstop nvsecd
and then restart with /etc/netnmrc (/etc/init.d/netnmrc on Solaris or Linux).
This issue has been fixed in the upcoming FixPack 2 (FP02) by updating the
NVenvironment script so that if you run that before you do ovstart, it will
source the correct environment for you, and then the daemons will inherit it
when you do the ovtstart.
But I still don't know why you are not
getting an nvserverd.log which shows the same tec_create_handle failure that
you see in the formatted nettl. We do get that here.
James
Shanks
Level 3 Support for Tivoli NetView for UNIX and Windows
Tivoli
Software / IBM Software Group
This message (including any attachments) contains confidential
information intended for a specific individual and purpose, and is protected
by law. If you are not the intended recipient, you should delete this message.
Any disclosure, copying, or distribution of this message, or the taking of any
action based on it, is strictly prohibited.
This message (including any attachments) contains confidential information intended for a specific individual and purpose, and is protected by law. If you are not the intended recipient, you should delete this message. Any disclosure, copying, or distribution of this message, or the taking of any action based on it, is strictly prohibited.
|