[
https://issues.apache.org/jira/browse/FLINK-15368?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17002693#comment-17002693
]
Yun Tang commented on FLINK-15368:
----------------------------------
The basic idea of end-to-end test for controlling RocksDB memory usage is to
expose RocksDB native metrics and print in logs to track the memory usage just
like {{wait_oper_metric_num_in_records}} bash function used in many end-to-end
tests. There still needs something else to do like introducing [block cache
properties|https://github.com/facebook/rocksdb/blob/afa2420c2bf0304a4b8796cab219e859146cc031/include/rocksdb/db.h#L790]
into Flink RocksDB native metrics.
I have implemented to expose block cache metrics in my private branch. However,
I found with a simplified version of {{DataStreamAllroundTestProgram}}, the
block cache usage would easily exceed the capacity due to the large pinned
usage.
I have tried to avoid to pin L0 and top level index & filter but it still
existed. Then I try to allocate the LRUCache with {{strictCapacityLimit=true}}
property. However, task manager would easily crash due to core dump of RocksDB.
Still Investigating.
> Add end-to-end test for controlling RocksDB memory usage
> --------------------------------------------------------
>
> Key: FLINK-15368
> URL: https://issues.apache.org/jira/browse/FLINK-15368
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / State Backends
> Affects Versions: 1.10.0
> Reporter: Yu Li
> Assignee: Yun Tang
> Priority: Critical
> Fix For: 1.10.0
>
>
> We need to add an end-to-end test to make sure the RocksDB memory usage
> control works well, especially under the slot sharing case.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
