I’m running a Couchbase cluster, in 3.0.1 community edition, where I see the memory usage growing every day, over the quota defined per server. I tried to find out which part is eating the memory, but any Couchbase graph reports a so high usage. Here is an example of what I’m talking about. In my cluster, the per RAM server quota is set to 71Go.
On a particular node, I see this reports, from Couchbase graphs:
- Meta data usage: 24 Go
- User RAM usage: 45 Go
- Global RAM usage: 51 Go
And from the system, a pmap -d on memcached process report 77 Go used by this processus:
> pmap -d 1412 > ... > 00000000016de000 77381432 rw--- 0000000000000000 000:00000 [ anon ] > ... > mapped: 77799464K writeable/private: 77710396K shared: 0K
There is 6 Go difference between the quota and the memory really used, not reported by Couchbase, but only by the system, and it’s still growing. The system is running a Linux 3.16.0-4 under Debian. I disabled transparent huge pages, and set the swapiness to 0.
Have you any idea of the origin of the problem ?
Thanks in advance for the help.