[jira] [Updated] (BEAM-12753) Flink Runner fails with memory issue when creating first TestStream instance

["Danny McCormick (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/BEAM-12753?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Danny McCormick updated BEAM-12753:
-----------------------------------
    Fix Version/s: Not applicable
       Resolution: Fixed
           Status: Resolved  (was: Open)

> Flink Runner fails with memory issue when creating first TestStream instance
> ----------------------------------------------------------------------------
>
>                 Key: BEAM-12753
>                 URL: https://issues.apache.org/jira/browse/BEAM-12753
>             Project: Beam
>          Issue Type: Bug
>          Components: runner-flink, sdk-go
>            Reporter: Jack McCluskey
>            Assignee: Danny McCormick
>            Priority: P3
>             Fix For: Not applicable
>
>          Time Spent: 14h 10m
>  Remaining Estimate: 0h
>
> Running Go SDK TestStream integration tests on the Flink runner consistently results in the following error on the first test run regardless of the test itself:
>  
> {code:java}
> 2021/08/11 21:12:47  (): java.io.IOException: Insufficient number of network buffers: required 17, but only 14 available. The total number of network buffers is currently set to 2048 of 32768 bytes each. You can increase this number by setting the configuration keys 'taskmanager.memory.network.fraction', 'taskmanager.memory.network.min', and 'taskmanager.memory.network.max'.
> {code}
> This error is thrown on the first test run on a fresh instance of the Flink job server; subsequent test runs of the same suite pass. Stopping the job server and starting a new one results in the first test run failing again. The memory error is thrown for int64, float64, and boolean data types; however, if the first TestStream contains strings or byte slices the memory error does not occur. It should be noted that strings and byte slices are getting appended with an extra length prefix on the Flink side, although it is unclear if that is related. 
>  
> On further investigation, the bug appears to be environment-specific, with many machines unable to reproduce the issue; however, the error still occurs on Jenkins for test runs.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)