kporter
January 14, 2015, 5:05am
21
Your case is abnormal and we will be investigating the problem further on our side. You may try setting a 1000 microsecond sleep between nsup deletes to see if that improves the latencies without needing to resort to running it via cron.
You can dynamically set nsup-delete-sleep from asadm using:
asadm -e "asinfo -v 'set-config:context=service;nsup-delete-sleep=1000'"
hi, kporter
here is the info after set nsup-delete-sleep
Monitor> printconfig -h 192.168.3.61 -v nsup-period
====192.168.3.61:3000====
nsup-period 120
Monitor> printconfig -h 192.168.3.61 -v nsup-delete-sleep
====192.168.3.61:3000====
nsup-delete-sleep 1,000
Monitor> latency -k reads
====reads====
timespan ops/sec >1ms >8ms >64ms
192.168.3.52:3000 06:04:52-GMT->06:05:02 11769.6 0.08 0.00 0.00
192.168.3.53:3000 06:04:53-GMT->06:05:03 11678.8 0.03 0.00 0.00
192.168.3.54:3000 06:04:48-GMT->06:04:58 10889.8 0.02 0.00 0.00
192.168.3.55:3000 06:04:57-GMT->06:05:07 11952.7 0.09 0.00 0.00
192.168.3.56:3000 06:04:58-GMT->06:05:08 11465.8 0.06 0.00 0.00
192.168.3.57:3000 06:04:56-GMT->06:05:06 12828.5 0.02 0.00 0.00
192.168.3.58:3000 06:04:52-GMT->06:05:02 12222.3 0.02 0.00 0.00
192.168.3.59:3000 06:04:54-GMT->06:05:04 11986.8 0.03 0.00 0.00
192.168.3.60:3000 06:04:55-GMT->06:05:05 11495.1 0.02 0.00 0.00
192.168.3.61:3000 06:04:56-GMT->06:05:06 10888.8 0.25 0.16 0.14
kporter
January 14, 2015, 6:51am
23
Could you run this again:
tail -n 2000 /var/log/aerospike/aerospike.log | grep "{cmmapping}"
And provide a another latency output afterwards.
hi, kporter
i only set nsup-delete-sleep for 5 minute, and than i set it back.
do i need set it and run for hours ?
kporter
January 14, 2015, 4:42pm
25
No not neccesary, after second inspection I realized the setting was only on 192.168.3.61 which lost performance.
i am sorry i didn’t say it clearly.
We look forward to your good news.
S3500’s latency is pretty high we fighting with this issue too
@blonkel what version of Aerospike are you running.
@QLeelulu , @blonkel this issue may have been resolved in 3.5.2 we improved our partition tree locking strategy and have seen a reduction in long tail latencies.
1 Like
@QLeelulu , @blonkel - the issue @kporter is referring to is the following, which was fixed in server release 3.5.2:
KVS - AER-3250 - Improved latencies during nsup - use dual tree locks so reads and overwrites are not blocked while reducing a partition tree.
1 Like