We have our Tribe node (v5.3.0) that crashed recently. The hprof analysis showed that it was full of old statistics.
It is connected to 3 clusters. For two of them, there were statistics 10 minutes old. In the following images, the retained size is the size for a single NodeStats object (for a given IP, and a given timestamp).
Here is the path to GC roots for one of those 10 minutes old NodeStats:
Is having old statistics an expected behavior ? Are there any tribe specific configuration to prevent from keeping them for more than few minutes (1 or 2) ?