Details
-
Technical task
-
Resolution: Fixed
-
Major
-
4.0.0
-
Security Level: Public
Description
Ran the following test and I could reproduce the problem easily.
1. Create 3 node cluster using cluster_run with one indexer node
cluster_run -n 3
cluster_connect -n 3 -T n0:index,n1:kv,n2:kv
2. Create 4 indexes
for i in
do
cbindex -type create -index "first_name$i" -bucket default -fields=personal_details.first_name
done
3. Insert 10 million docs
4. Drop the bucket to ensure that all indexes are deleted
Observed that total indexer process memory remains ~6G, out of that 4.5G is consumed by forestdb.
I have tried adding forestdb snapshot refcount stats in the indexer to verify that we are not holding any handle references.
Also tried linking forestdb with jemalloc to ensure that it is not caused by system malloc not releasing memory to OS even after 2 hours.
inmemory snapshot interval=200ms
commit interval=30mins
Memory growth is significantly less if you set commit interval to 3mins.
Attachments
Issue Links
- relates to
-
MB-15338 Make jemalloc the default allocator in forestdb sherlock build
- Closed