Ridiculously low performance threshold in 1.8

[coop]

Just to second UnterGeek on this one, we're seeing the exact same problem.

I'm running on AIX 5.3 with Zabbix 1.8 and Oracle 10.2 backend (full enterprise license). We compiled against the lib32 OCI libraries just fine.

Running Zabbix 1.6.8 with Oracle 10.2 and the libsqlora8 performed just fine. Running 1.8 on the same server with the same database has issues.

Right now I have 900 items queued up across 5 different item types and our stats are very low...

Number of hosts (monitored/not monitored/templates) 	106 	42 / 14 / 50
Number of items (monitored/disabled/not supported)  	1579 	877 / 617 / 85
Number of triggers (enabled/disabled)[true/unknown/false] 	355 	355 / 0  [1 / 116 / 238]
Number of users (online) 	8 	2
Required server performance, new values per second 	17.5097222222 	-

I can provide any info needed. I'm sure it's a config issue somewhere. Troubleshooting 101 says you look for what's changed. Server and Database are exactly the same as they used to be. Zabbix and Oracle connection is different.

[untergeek]

We set up an 11g database. I was able to connect with Zabbix 1.8 compiled against 10g libraries (but for some reason not with 11g libraries, go figure). There were no problems. We did some further digging to find out why.

Our Oracle DBA did some probing yesterday and found that at least one of the reasons was that some of the queries going to Oracle 10g were not referencing the INDEXES but were doing full history scans (over 4 million in there) instead. This is why the DB was slow to respond. Our DBA made some extra INDEXes and strangely enough the DB became more responsive and the Queued items all went green. There is something else at work here.

To recap: Oracle 11g has no problems, even when accessed by 10g library compiled Zabbix. 10g seems to have query issues with the indexes.

[jansonz]

Hello,

As I was writing before (http://www.zabbix.com/forum/showthre...t=15126&page=2) - I have the same problem. I use 2 servers, One for Zabbix server and other for Oracle.

The performance in which the items are updating is terreble. On the same hardware 1.6 performed without problem, but now I have a lot of problems:

QUEUE OF ITEMS TO BE UPDATED
Items 5 seconds 10 seconds 30 seconds 1 minute 5 minutes More than 10 minutes
Zabbix agent 13 24 38 465 658 8573
Zabbix agent (active) 0 0 0 0 0 0
SNMPv1 agent 5 0 22 11 26 7053
SNMPv2 agent 0 0 0 0 0 1661
SNMPv3 agent 0 0 0 0 0 0
IPMI agent 0 0 0 0 0 0
SSH agent 0 0 0 0 0 0
TELNET agent 0 0 0 0 0 0
Simple check 0 0 0 0 0 1139
Zabbix internal 0 0 0 0 0 1
Zabbix aggregate 0 0 0 0 0 0
External check 0 0 0 0 0 263

What coud be the problem?

[untergeek]

I can tell you this: We had 2 options, Oracle 11g or MySQL. Since we're an Oracle shop and we already have the Veritas clustering set-up with a HUGE high-end SAN, we are going with Oracle 11g. It will be our first ever 11g database in production use, and the DBAs are not thrilled about this, as they would rather have kept us more uniformly with 10g (which is the vast majority of our install base).

I don't know what the Zabbix guys did, but I would love to hear of any Zabbix 1.8 install running on Oracle 10g that doesn't have this problem.

Anyone?

[jansonz]

I have Oracle Database 11g Enterprise Edition Release 11.1.0.6.0

When Zabbix developers plan to make new release, where this problem will be solved?

[sersad]

I have Oracle Database 11g Enterprise Edition Release 11.1.0.6.0

When Zabbix developers plan to make new release, where this problem will be solved?

tray it http://www.zabbix.com/forum/showpost...2&postcount=13

[coop]

tray it http://www.zabbix.com/forum/showpost...2&postcount=13

Just to clarify, these changes were made right in the DB?

[untergeek]

I don't think that will cure anything. The schema is EXACTLY the same for an Oracle 10 as an Oracle 11 install, but somehow it has results so different? Weren't those instructions also Postgres-specific?

There has to be something else going on. I just don't have time to dig to find it so we're using Oracle 11.

[sj7trunks]

Came across a similar problem: Here's how I fixed without changing the database:

# semaphores: semmsl, semmns, semopm, semmni
kernel.sem = 250 32000 100 128

# Controls the maximum shared segment size, in bytes
kernel.shmmax = 68719476736
kernel.shmall = 4294967296
kernel.shmmni = 4096

Then edit zabbix_server.conf and add the following:
CacheSize=64M

Then run sysctl -p /etc/sysctl.conf

-- Might want to raise your CacheSize higher than mine. I'm about to file a bug since zabbix_server still polls for web servers but becomes ultimately unusable with a crazy huge queue with no hints other than the log file mentioning that we need to raise CacheSize without mention of "ERROR."



Here is the rest of my oracle configuration for systcl.conf

#Oracle
fs.file-max = 6553600
kernel.shmmni = 4096
kernel.sem = 250 32000 100 128
net.ipv4.ip_local_port_range = 1024 65000
net.core.rmem_default = 4194304
net.core.rmem_max = 4194304
net.core.wmem_default = 262144
net.core.wmem_max = 262144

[jansonz]

Sadly, but shis solves nothing...

[jansonz]

Ok, it seems, that there will be no solution from Zabbix developers. Is there some solution to downgrade Zabbix server from 1.8 to 1.6.8.

[umasse]

Ubuntu server settings

Ran into this issue with an Ubuntu 10.04 installation, running Zabbix 1.8.1 from the repositories.

I had to add this line to "/etc/zabbix/zabbix_server.conf"

CacheSize = 64M

Then I added a file called "60-zabbixshmem.conf" to the folder "/etc/sysctl.d/"
with this in it:

# Controls the maximum shared segment size, in bytes
kernel.shmmax = 134217728
kernel.shmall = 8388608
kernel.shmmni = 4096

This has made the error go away, Zabbix Server now runs, and is receiving values again.