Active items been removed from cache unexpectedly.

Hi Folks,

We have a 4 node cluster and we are experiencing a strange behavior which it doesn’t seem to be normal and we would like to know your opinion on the subject.

The fact is that the “Active Items” stats are extremely variable and it looks like the active items do not stay active the time it is expected. A high level stats monitoring shows that the number of active items suffer continuous variances growing up to several thousands of items and a couple seconds after it goes down to zero. The point is we are storing items which are share across users with an expiration time of 12 hours. We would expect to see at least all these items active for a long time and the “Active” stats could never be 0 (or at least wouldn’t be likely). However it seems that our items have a really short life in our Membase cluster.

Suspecting that this issue was related with the item’s expiration, we created a cluster in our testing environment and the active stats look s consistent. The number of active items grows up in a continuous way and they are not immediately removed (like it occurs with our Prod cluster).

Additionally, as far as we’ve read there is a job which is executed once per hour to clean up the active items that have already expired. Taking this into account, even if the items would be expiring before expected (which does not seem to be the case based in our test with another cluster), it should be expected that the items would remain at least one hour, shouldn’t it ?

All nodes are Windows Server 2008 R2 SP1 and membase version is 1.7.2r-20-g6604356.

Any idea that you could share with us, it would be more than welcome.

Thanks in advance !,
Ale