[
https://issues.apache.org/jira/browse/FLINK-17312?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17092441#comment-17092441
]
lun zhang commented on FLINK-17312:
-----------------------------------
[~jark] support DML hint is awesome.But it'll break the ANSI _SQL rule. Can we
try another way like add a option *restore-savepoint-path* config in the sink
table source. So we can support each insert DML has the own_
_*restore-savepoint-path* ._
_for eg:_
_in env.yml_
_tables:_
_{{tables:}}_
_{{- name: sink_table}}_
_{{type: sink}}_
_{{update-mode: append}}_
_{{connector: --}}_
_{{restore-savepoint-path: xxx}}_
_{{when you try "insert into sink_table select * from xxx" it will support read
restore-savepoint and restart the job.}}_
__
> Support sql client start with restore savepoint
> -----------------------------------------------
>
> Key: FLINK-17312
> URL: https://issues.apache.org/jira/browse/FLINK-17312
> Project: Flink
> Issue Type: Improvement
> Components: Table SQL / Client
> Affects Versions: 1.10.0, 1.11.0
> Reporter: lun zhang
> Priority: Major
> Labels: pull-request-available
>
> Now flink support stop or cancel a flink job by *flink cancel -s
> savepointpath* . And you can restart the jar flink job from *flink run -s
> savepointpath* .
> But in sql client, we can't specific the restart savepoint .
> By help of [~ykt836] and [~lzljs3620320] we have two way to go.
> # the most easy way, set in envirment.yml. I've pull a request for it.It's
> add *restore-savepoint-path prop* in Excution part.You can set
> *restore-savepoint-path* in envirment.yml or just in command line( by *_set
> execution.restore-savepoint-path=xxx_*). When we update a "insert" DML.It'll
> add restore-savepoint-path in job graph.And restart with the last
> restore-savepoint.But in this way we can only unset
> *_execution.restore-savepoint-path_* _when you update two or more *insert*_
> DML.
> # support by DML hit(for eg: "*insert into tb select * from tb2 with
> (restore-savepoint-path = 'xxx')*". It's very useful if you want insert two
> or more DML in one sql client session. But It's may token more work to go.We
> need support a new DML hit.
> In my opinion, the way 2 is most elegant to go.But it'll take much more
> working on it.Restart two or more stop with savepoint job is not usual in
> real world.So it's better to just add this in envirment variable.When we
> submit our job only need check the env variable and set the restore-savepoint
> in job graph.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
