beam-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jozef Vilcek (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (BEAM-5180) Broken FileResultCoder via parseSchema change
Date Mon, 20 Aug 2018 12:11:00 GMT

     [ https://issues.apache.org/jira/browse/BEAM-5180?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Jozef Vilcek updated BEAM-5180:
-------------------------------
    Description: 
Recently this commit

https://github.com/apache/beam/commit/3fff58c21f94415f3397e185377e36d3df662384

introduced more strict schema parsing which is breaking the contract between _FileResultCoder_ and
_FileSystems.matchNewResource()_.

Coder takes _ResourceId_ and serialize it via `_toString_` methods and then relies on filesystem
being able to parse it back again. Having strict _scheme://_ breaks this at least for Hadoop
filesystem which use _URI_ for _ResourceId_ and produce _toString()_ in form of `_hdfs:/some/path_`

I guess the _ResourceIdCoder_ is suffering the same problem.

Either scheme parsing should be less strict or ResourceId.toString() for hadoop fixed

  was:
Recently this commit introduced more strict schema parsing which is breaking the contract
between `FileResultCoder` and `FileSystems.matchNewResource()`.

Coder takes `ResourceId` and serialize it via `toString` methods and then relies on filesystem
being able to parse it back again. Having strict `scheme://` breaks this at least for `Hadoop`
filesystem which use `URI for `ResourceId` and produce `toString()` in form of `hdfs:/some/path`

I guess the `ResourceIdCoder` is suffering the same problem.

Either scheme parsing should be less strict or `ResourceId.toString()` for `hadoop` fixed


> Broken FileResultCoder via parseSchema change
> ---------------------------------------------
>
>                 Key: BEAM-5180
>                 URL: https://issues.apache.org/jira/browse/BEAM-5180
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-core
>    Affects Versions: 2.6.0
>            Reporter: Jozef Vilcek
>            Assignee: Kenneth Knowles
>            Priority: Blocker
>
> Recently this commit
> https://github.com/apache/beam/commit/3fff58c21f94415f3397e185377e36d3df662384
> introduced more strict schema parsing which is breaking the contract between _FileResultCoder_ and
_FileSystems.matchNewResource()_.
> Coder takes _ResourceId_ and serialize it via `_toString_` methods and then relies on
filesystem being able to parse it back again. Having strict _scheme://_ breaks this at least
for Hadoop filesystem which use _URI_ for _ResourceId_ and produce _toString()_ in form
of `_hdfs:/some/path_`
> I guess the _ResourceIdCoder_ is suffering the same problem.
> Either scheme parsing should be less strict or ResourceId.toString() for hadoop fixed



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message