[
https://issues.apache.org/jira/browse/FLINK-7750?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16187875#comment-16187875
]
Razvan edited comment on FLINK-7750 at 10/2/17 11:47 AM:
---------------------------------------------------------
Sure, the application is reading messages from Kafka then calls a service via
HTTP. The state is made up of counters. I tried to use S3 (Configured
state.savepoints.dir: [S3]:///flink/savepoints) and this works fine.
Using RocksDB as state backend and S3 location.
But if I specify folder location from JobManager it splits the files as I wrote
in description.
The job is dependant on Apache Flink version 1.2.0 (Maven dependency) , should
this be an issue?
was (Author: razvan):
Sure, the application is reading messages from Kafka then calls a service via
HTTP. The state is made up of counters. I tried to use S3 (Configured
state.savepoints.dir: [S3]:///flink/savepoints) and this works fine.
But if I specify folder location from JobManager it splits the files as I wrote
in description.
The job is dependant on Apache Flink version 1.2.0 (Maven dependency) , should
this be an issue?
> Strange behaviour in savepoints
> -------------------------------
>
> Key: FLINK-7750
> URL: https://issues.apache.org/jira/browse/FLINK-7750
> Project: Flink
> Issue Type: Bug
> Components: Core
> Affects Versions: 1.3.2
> Reporter: Razvan
> Priority: Blocker
>
> I recently upgraded from 1.2.0 and savepoint creration behaves strange.
> Whenever I try to create a savepoint with specified directory Apache Flink
> creates a folder on the active JobManager (even if I trigger savepoint
> creation from a different JobManager) which contains only _metadata. And
> another folder on the TaskManager where the job is running which contains the
> actual savepoint.
> Obviously if I try to restore it says it can't find the savepoint.
> This worked in 1.2.0.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
