Hi Nathan (& everyone), as discussed on IRC, here's the example PMDA we created
copied, cleaned, scaled, and gutted from our activemq which shows what we're
seeing the issue where the PMDA crashes when the refresh decides to inform PMCD
that all of the instances in the indom have disappeared (I always think of the
canonical case of instance domains of 'disks' where one has ejected all the
disks leaving nothing, which should be ok.. as you say, not usual, but should
work).
Here's also the attached output from running 'pcp' after the PMDA has died.
Performance Co-Pilot configuration on atrigg-T1700:
platform: Linux atrigg-T1700 3.13.0-40-generic #69-Ubuntu SMP Thu Nov 13
17:53:56 UTC 2014 x86_64
hardware: 8 cpus, 2 disks, 1 node, 32119MB RAM
timezone: AEDT-11
pmcd: Version 3.8.12-1, 7 agents, 4 clients
pmda: pmcd proc xfs linux mmv jbd2 activemq failingpmda[720904]
pmlogger: atrigg-T1700: /var/log/pcp/pmmgr/atrigg-T1700/archive-20141203.220600
primary logger: atrigg-T1700/20141204.00.10
pmie: atrigg-T1700: /var/log/pcp/pmie/atrigg-T1700/pmie.log
atrigg-T1700: /var/log/pcp/pmmgr/atrigg-T1700/pmie.log
failingpmda.tar.gz
Description: GNU Zip compressed data
|