Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Fixed
-
4.0.0
-
llap
Description
The LRFU cache policy for the LLAP data cache doesn't provide enough insight to figure out, what is cached and why something might get evicted. This ticket is used to add Hadoop metrics 2 information (accessible via JMX) to the LRFU policy, providing following information:
- How much memory is cached for data buffers
- How much memory is cached for meta data buffers
- How large is the min-heap of the cache policy
- How long is the eviction short list (linked list)
- How much memory is currently "locked" (buffers with positive reference count) and therefore in use by a query
These new counters are found in the MX bean, following this path:
Hadoop/LlapDaemon/LowLevelLrfuCachePolicy-<hostname>