Details
-
Bug
-
Resolution: Fixed
-
Blocker
-
2.2.0
-
Security Level: Public
-
None
-
Centos 64-bit
-
Description
Some test details:
2 -> 2 unidirectional replication, 25M items, ~60% resident ratio, 10K mixed front-end ops/sec.
https://raw.github.com/pavel-paulau/perfrunner/master/tests/xdcr_2x2_unidir_25M_dgm.test
On destination nodes memory utilization is pretty low (600-700 MB), on source node it's about 3GB and tends to grow. No surprisingly we have troubles with XDCR on Windows.
Problem is even more noticeable in init. replication tests. RSS reaches ~5GB level. Regression is pretty obvious.
Memory utilization becomes lower when there is no XDCR activity.
Attachments
Issue Links
- relates to
-
MB-9084 Source node went down multiple times after XDCR initialization [{nodedown_reason, net_tick_timeout}]
- Closed
-
MB-9497 Rebalance fails while loading data simultaneously onto 2 active clusters with bidirectional XDCR
- Closed
-
MB-9097 On a sufficiently sized cluster, seeing erl_crash dumps w/ "Cannot allocate 306104640 bytes of memory (of type "heap")"
- Closed