[ https://issues.apache.org/jira/browse/HDDS-10043 ]
Tsz-wo Sze deleted comment on HDDS-10043:
-----------------------------------
was (Author: szetszwo):
[~weichiu] has found out that datanodes send entire chunk (i.e. 4MB) to the
client, even if the client reads only a few KB.
We may test writing smaller chunks, say 1MB instead of 4MB, by setting
- ozone.client.stream.buffer.size to 1M
Then, it will get 1MB chunks when reading back. Of course, this is just a
workaround but not a long term solution.
> java.lang.OutOfMemoryError on FSDataOutputStream write ops
> ----------------------------------------------------------
>
> Key: HDDS-10043
> URL: https://issues.apache.org/jira/browse/HDDS-10043
> Project: Apache Ozone
> Issue Type: Bug
> Components: Ozone Client
> Reporter: Pratyush Bhatt
> Priority: Major
>
> Seems like FSDataOutputStream utilise higher heap memory.
> Doing write Ops in ~300 FSDataOutputStream objects sequentially is causing
> {color:#FF0000}_java.lang.OutOfMemoryError: Java heap space_
> {color}{color:#FF0000}error. {color}
> While the FSDataOutputStream on HDFS doesn't throws heap issues, tried with
> ~3000 objects on same Environment, works fine.
> Note: Both the tests were performed in same environment.
> Client is a Kube container, specs:
> {code:java}
> limits:
> cpu: "1"
> ephemeral-storage: 5G
> memory: 300M
> requests:
> cpu: 200m
> ephemeral-storage: 1G
> memory: 200M {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
