OpenNMS 15: warm your postgres cache
OpenNMS 15 puts a much higher load on the database than previous versions.
Besides tuning postgres, the OS and perhaps splitting the app and the db on different boxes one aspect that I found to really make a difference is having a warm postgres cache.
Additional tip: if you haven't already put postgres on XFS. There is a reason that RH7 switched to XFS as the default fs and it is performance. You will also find that most postgres people recommend XFS instead of ext3/4.
If you followed the instructions on my previous post you should have a v_database_cache view in the opennms database. Soon after installing OpenNMS 15 I found that the events relation was not cached at all (less than 2% of it was cached after one day).
This is probably due to to various reasons, most likely queries have been improved to use indices instead of scanning the tables, but the UI performance suffers (it takes 1-2 seconds to display the node pages)[1].
To warm the datbase cache and improve general performance and responsiveness of the UI run this command as the postgres user:
Now check the database cache: percent_of_relation should show a larger value for the events relation. In my case it was 100% (shared_buffers=1GB , event table is ~180M) and I found the UI to be much much snappier.
Thanks for reading.
[1] yes I am running on somewhat aged hardware (Proliant DL580G5, RAID10 on 10K drives, 8GBRAM, 2 × XEON).
Besides tuning postgres, the OS and perhaps splitting the app and the db on different boxes one aspect that I found to really make a difference is having a warm postgres cache.
Additional tip: if you haven't already put postgres on XFS. There is a reason that RH7 switched to XFS as the default fs and it is performance. You will also find that most postgres people recommend XFS instead of ext3/4.
If you followed the instructions on my previous post you should have a v_database_cache view in the opennms database. Soon after installing OpenNMS 15 I found that the events relation was not cached at all (less than 2% of it was cached after one day).
This is probably due to to various reasons, most likely queries have been improved to use indices instead of scanning the tables, but the UI performance suffers (it takes 1-2 seconds to display the node pages)[1].
To warm the datbase cache and improve general performance and responsiveness of the UI run this command as the postgres user:
psql -A opennms -c "select * from events; " > /dev/nullIf you have a large events table consider adding a filter (ie: only events from the last week).
Now check the database cache: percent_of_relation should show a larger value for the events relation. In my case it was 100% (shared_buffers=1GB , event table is ~180M) and I found the UI to be much much snappier.
opennms=# select * from v_database_cache ;
relname | buffered | buffers_percent | percent_of_relation
-------------------------------+----------+-----------------+---------------------
events | 181 MB | 17.7 | 100.0
notifications | 44 MB | 4.3 | 84.6
outages | 16 MB | 1.6 | 100.0
events_ipaddr_idx | 4128 kB | 0.4 | 40.9
bridgemaclink | 4704 kB | 0.4 | 100.7
events_nodeid_idx | 4008 kB | 0.4 | 51.9
events_nodeid_display_ackuser | 4480 kB | 0.4 | 42.9
assets | 2848 kB | 0.3 | 101.1
snmpinterface | 1576 kB | 0.2 | 100.0
bridgemaclink_pk_idx2 | 2296 kB | 0.2 | 100.0
Thanks for reading.
[1] yes I am running on somewhat aged hardware (Proliant DL580G5, RAID10 on 10K drives, 8GBRAM, 2 × XEON).