Here is screenshots of stats from one of our servers:
From time to time one of the servers starts responding with temporary OOM errors.
First i thought that i can change mem_high_wat/mem_low_wat thresholds so eviction process start earlier, and server will have more RAM, even if it get more IO load. But setting low mem_high_wat not really affect us, as you can see on graph “Memory used” graph, not changing at all. And on second graph you can see, that “evictions” happen very rarely, despite the fact that “Memory used” much higher then mem_high threshold. Also you can see that “Active user_data in RAM” + “Replica user_data in RAM” do not reconcile with totals. So i can assume it can be some sort of memory leak?
Problem fixes only by restarting couchbase server on that machine. Thanks.