-
Type:
Problem report
-
Resolution: Unresolved
-
Priority:
Minor
-
Affects Version/s: 7.0.32
-
Component/s: Agent (G)
-
None
-
Sprint candidates
Steps to reproduce:
- Windows Server 2019 with SQL Server, Zabbix agent 7.0.30, about 400 passive perf_counter_en items on SQLServer:Databases(*) counters with 1m interval.
- Reboot the host, so the agent starts before the SQL Server counter instances exist.
Result:
In every cycle the collector adds 162 counters again and fails to read them. It does not recover until agent is restarted.
10236:20261120:132000.628 collect_perfstat(): PerfCounter '\SQLServer:Databases(DB2)\Log Flush Wait Time' successfully added 10236:20261120:132002.312 collect_perfstat(): cannot get counter value '\SQLServer:Databases(DB2)\Log Flush Wait Time': [0x800007D1] The specified instance is not present. 10236:20261120:132002.317 deactivate_perf_counter() counterpath:'\SQLServer:Databases(DB2)\Log Flush Wait Time'
A collector cycle takes 4 to 10 seconds instead of 0.02 seconds.
10236:20261120:132005.839 In collect_perfstat() 10236:20261120:132015.495 End of collect_perfstat()
Listeners still return correct values for the same counters, but perf_counter_en takes about 5 seconds instead of 0.6 seconds. All listeners stay busy and most incoming connections are accepted only after proxy has closed them.
92:20261120:132000.820 failed to accept an incoming connection: from 192.0.2.10: unspecified certificate verification error: TLS handshake set result code to 1: file ssl\record\rec_layer_s3.c line 701 func : error:0A000126:SSL routines::unexpected eof while reading: TLS write fatal alert "decode error"
After agent restart, collector has no such errors.
All passive items on the host have gaps until agent restart (01:58 PM in the graph):

Expected:
The collector recovers when instances appear and it does not slow down other passive checks.