Details
-
Bug
-
Resolution: Incomplete
-
Critical
-
None
-
Cheshire-Cat
-
Untriaged
-
1
-
Unknown
Description
When we run N1QL YCSB tests on aws, we hit out of memory issue when loading docs with indexes created.
Hard out-of-memory error: Bucket "bucket-1" on node XXX is full. No memory currently allocated to this bucket can be easily released.
The AWS instance we use for serve nodes is t3.large.
OS: CentOS 7, CPU: t3.large (2 vCPU), Memory: 8 GB, Disk: EBS 40GB
In our test configurations, there are 3 server nodes, and we run multiple services (kv/index/n1ql/fts) on each node. The memory quota of each service on a node is listed below.
mem_quota = 256
index_mem_quota = 1024
fts_index_mem_quota = 256
There are 1 bucket and 40 collections. When we tried to load 150K docs with 100 workers, we hit the out-of-memory issue. This issue is reproducible.
Build: 7.0.0-5219
Job: http://perf.jenkins.couchbase.com/job/Cloud-Tester/64/
Logs:
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-Cloud-Tester-64/ec2-34-217-103-182.us-west-2.compute.amazonaws.com.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-Cloud-Tester-64/ec2-34-218-237-13.us-west-2.compute.amazonaws.com.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-Cloud-Tester-64/ec2-52-40-243-48.us-west-2.compute.amazonaws.com.zip
When we reduced the number of workers from 100 to 10, we got a good run. It looks like the cluster can't handle loading docs correctly when there are many workers.
Job: http://perf.jenkins.couchbase.com/job/Cloud-Tester/68/
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-Cloud-Tester-68/ec2-18-236-72-97.us-west-2.compute.amazonaws.com.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-Cloud-Tester-68/ec2-34-215-114-206.us-west-2.compute.amazonaws.com.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-Cloud-Tester-68/ec2-54-212-38-209.us-west-2.compute.amazonaws.com.zip