[
https://issues.apache.org/jira/browse/BEAM-8442?focusedWorklogId=341659&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-341659
]
ASF GitHub Bot logged work on BEAM-8442:
----------------------------------------
Author: ASF GitHub Bot
Created on: 12/Nov/19 04:42
Start Date: 12/Nov/19 04:42
Worklog Time Spent: 10m
Work Description: robertwb commented on issue #10004: [BEAM-8442] Unify
bundle register in Python SDK harness
URL: https://github.com/apache/beam/pull/10004#issuecomment-552731641
Actually, looking at the code, if SdkWorker.register (which is now
asynchronous) does not complete before SdkWorker.process_bundle
starts, process_bundle fails. We could be getting lucky as register is
fast, but there's a race here now that it's not synchronous.
https://github.com/sunjincheng121/beam/blob/79d21b11ca31963bd1be84a26502e5577bbefbdf/sdks/python/apache_beam/runners/worker/sdk_worker.py#L371
On Mon, Nov 11, 2019 at 7:24 PM Jincheng Sun <[email protected]>
wrote:
> @robertwb <https://github.com/robertwb>
> Thanks a lot for comments. After this change, if the registration failed,
> the data channel between the runner and the SDK harness will not be
> established and so the data transmission from the runner to the SDK harness
> will be timedout
>
<https://github.com/apache/beam/blob/74ea7496ff486640affbf2e6ab4a45d3ea25efc8/runners/java-fn-execution/src/main/java/org/apache/beam/runners/fnexecution/data/GrpcDataService.java#L170>.
> At that time, the registration failure could be reported. This is also the
> same case for the process bundle instruction before and after this change.
> Are you concerning that we should detect the registration failure at the
> early time? If it's the case, how do we deal with the process bundle
> instruction, should we keep the same behavior for both the registration and
> process bundle instruction.
> What do you think?
> Best, Jincheng
>
> —
> You are receiving this because you were mentioned.
> Reply to this email directly, view it on GitHub
>
<https://github.com/apache/beam/pull/10004?email_source=notifications&email_token=AADWVAO3L4WQ7QEWLIJ3K4DQTIOWNA5CNFSM4JJPDHSKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDY4X7I#issuecomment-552717309>,
> or unsubscribe
>
<https://github.com/notifications/unsubscribe-auth/AADWVAMRFWNVPIBS5UBISE3QTIOWNANCNFSM4JJPDHSA>
> .
>
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 341659)
Time Spent: 5h (was: 4h 50m)
> Unify bundle register in Python SDK harness
> -------------------------------------------
>
> Key: BEAM-8442
> URL: https://issues.apache.org/jira/browse/BEAM-8442
> Project: Beam
> Issue Type: Sub-task
> Components: sdk-py-harness
> Reporter: sunjincheng
> Assignee: sunjincheng
> Priority: Major
> Fix For: 2.18.0
>
> Time Spent: 5h
> Remaining Estimate: 0h
>
> There are two methods for bundle register in Python SDK harness:
> `SdkHarness._request_register` and `SdkWorker.register.` It should be unfied.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
