[
https://issues.apache.org/jira/browse/FLINK-9423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16490431#comment-16490431
]
ASF GitHub Bot commented on FLINK-9423:
---------------------------------------
Github user StefanRRichter commented on the issue:
https://github.com/apache/flink/pull/6062
@sihuazhou This is exactly my plan on how introduce a "timer state" for the
keyed state backends. In a nutshell, time timer service can just register and
operator on those states, where the basic operations that we expose are
somewhat along the lines of `scheduleTimer`, `stopTimer`, `peekNextTimer`,
`pollNextTimer`. I am currently working on a small design document to outline
this integration. For RocksDB we can implementation of the state can be loosely
based on some of the ideas from the class `RocksDBTimerHeap` of PR #3359.
> Implement efficient deletes for heap based timer service
> --------------------------------------------------------
>
> Key: FLINK-9423
> URL: https://issues.apache.org/jira/browse/FLINK-9423
> Project: Flink
> Issue Type: Improvement
> Components: Streaming
> Affects Versions: 1.6.0
> Reporter: Stefan Richter
> Assignee: Stefan Richter
> Priority: Major
>
> The current data structures in the `HeapInternalTimerService` are not able to
> support efficient timer deletes, the complexity is currently O\(n\), where n
> is the number of registered timers.
>
> We can keep track of timer's positions in the priority queue and (in
> combination with the already existing set/map) have a more efficient
> algorithm for deletes.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
