Details
-
Improvement
-
Resolution: Unresolved
-
Major
-
6.5.0
-
A one (1) node quiescent system 8 core Xenon VM runing Debian 8 the VM has 24MB of RAM
uname -a
Linux couch01 3.16.0-10-amd64 #1 SMP Debian 3.16.70-1 (2019-07-22) x86_64 GNU/Linux
Couchbase Server 6.5.0-0000 (CE) just built from git today on Thu Sep 12 22:06:32 PST 2019
The hypervisor running the above VM is Joyent's SmartOS
uname -a
SunOS smart130 5.11 joyent_20190815T002608Z i86pc i386 i86pcA one (1) node quiescent system 8 core Xenon VM runing Debian 8 the VM has 24MB of RAM uname -a Linux couch01 3.16.0-10-amd64 #1 SMP Debian 3.16.70-1 (2019-07-22) x86_64 GNU/Linux Couchbase Server 6.5.0-0000 (CE) just built from git today on Thu Sep 12 22:06:32 PST 2019 The hypervisor running the above VM is Joyent's SmartOS uname -a SunOS smart130 5.11 joyent_20190815T002608Z i86pc i386 i86pc
Description
As I add more and more empty buckets to an idle Couchbase system built from just one (1) node with an 8 CPU VM and 24MB of RAM doing nothing the CPU continues to climb for both memcached and beam.smp at 28 buckets (26 empty) we use almost 1/2 of the cores in the 8 CPU VM
I start with no buckets and the top output is relatively low (top -n 4 user linuxbrew)
{{PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
2401 linuxbr+ 20 0 4182012 180416 7128 S 8.0 0.7 0:16.65 beam.smp
2741 linuxbr+ 20 0 8917820 484636 21860 S 2.0 2.0 0:18.64 java
2541 linuxbr+ 20 0 743128 32984 19756 S 0.8 0.1 0:01.46 indexer
2594 linuxbr+ 20 0 916168 54980 39020 S 0.8 0.2 0:00.93 cbq-engine}}
I add 4 buckets 2 with data (32K items and 5K items) (top -n 4 user linuxbrew)
{{PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
2401 linuxbr+ 20 0 4704780 599744 7192 S 61.3 2.4 6:34.93 beam.smp
2669 linuxbr+ 20 0 2465196 194636 24328 S 51.0 0.8 2:56.27 memcached
2541 linuxbr+ 20 0 1389776 335672 21896 S 16.4 1.4 2:31.17 indexer
2553 linuxbr+ 20 0 851196 307380 15856 S 2.4 1.2 0:34.77 projector}}
(more details in attachment)
I add 24 more empty buckets (28 total) (top -n 4 user linuxbrew)
{{PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
2669 linuxbr+ 20 0 3341740 906616 24328 S 191.8 3.7 30:40.84 memcached
2401 linuxbr+ 20 0 7157152 1.784g 7192 S 144.7 7.6 36:47.43 beam.smp
2541 linuxbr+ 20 0 1389968 325808 21896 S 12.8 1.3 6:08.28 indexer
2459 linuxbr+ 20 0 4222184 315872 16748 S 2.3 1.3 0:48.81 beam.smp}}
Is this really the expected CPU usage behavior for a one (1) node system doing
nothing e.g. the final CPU for the Couchbase user "linuxbrew" CPU sums to 351.6%
this is about 3.5 cores or almost 1/2 of my available compute.
I have attached an lsof output of running 28 bucket system and in addition some strace outputs.