[
https://issues.apache.org/jira/browse/FLINK-9735?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16532982#comment-16532982
]
Sihua Zhou commented on FLINK-9735:
-----------------------------------
I think this might be a trade-off of the design,
- 1) let the use create the option base on some basic configuration to get a
good performance in flink (e.g. setFsync(false) because Flink does not rely on
RocksDB data on disk for recovery).
- 2) give the chance for the use to create the option totally by themselves.
Given the design the user need to follow the rule it outlined, but I agree that
OptionFactory's api is a bit easy to overlook, and lead the currentOptions
parameter to be abandoned. Maybe a better interface could be provided to
address this.
> Potential resource leak in RocksDBStateBackend#getDbOptions
> -----------------------------------------------------------
>
> Key: FLINK-9735
> URL: https://issues.apache.org/jira/browse/FLINK-9735
> Project: Flink
> Issue Type: Bug
> Reporter: Ted Yu
> Assignee: vinoyang
> Priority: Minor
>
> Here is related code:
> {code}
> if (optionsFactory != null) {
> opt = optionsFactory.createDBOptions(opt);
> }
> {code}
> opt, an DBOptions instance, should be closed before being rewritten.
> getColumnOptions has similar issue.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
