Details
-
Type:
Incident report
-
Status: Closed
-
Priority:
Critical
-
Resolution: Fixed
-
Affects Version/s: 1.8.15
-
Component/s: Proxy (P), Server (S)
-
Labels:
-
Environment:Zabbix 1.8.15, 10000 dynamic index items
Description
Poller stick to 100% busy after restarting Zabbix server if there are a lot of dynamic index items. According to attached screenshot, there are 10000 dynemic index items, poller stick to 100% busy for 20-25 minutes.
Poller does not have shared cache for dynamic index at this moment, so there is no way to avoid this problem for large environment.
Activity
Problem reproduced with 1-2 poller processes and 1000 dynamic indexes. Pollers are 100% busy for 15-20 minutes after server start.
Fixed in development branch svn://svn.zabbix.com/branches/dev/ZBX-5973
Searching for one SNMP dynamic index inserts all other indexes from the same OID table into the cache in the same pass. The cache builds up much faster, number of SNMP queries is reduced.
The assumption is that most of cache entries will be used and unnecessary elements in the cache is not a problem.
Successfully tested! The performance improved dramatically. See my small change in r32682.
Andris Mednis Thanks, reviewed. Agree.
Fixed in versions pre-1.8.17 r32832, pre-2.0.5 r32863.
For trunk fixed in development branch svn://svn.zabbix.com/branches/dev/ZBX-5973-trunk
Fixed in version pre-1.8.17 r33135.
Fixed in version pre-2.0.5 r33192.
For trunk fixed in development branch svn://svn.zabbix.com/branches/dev/ZBX-5973-trunk, r33235.
Alexander Vladishev Successfully tested! Please review my changes in r33240:33245.
Andris Mednis Reviewed. Agree.
Fixed in version pre-2.1.0 rev.33254.
OK we decided to fix this for 1.8, 2.0 and trunk (2.2). Currently we plan to improve caching of dynamic indexes by implementing 2 things: