You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Sebastian Klemke (JIRA)" <ji...@apache.org> on 2017/07/02 00:25:00 UTC

[jira] [Commented] (FLINK-6965) Avro is missing snappy dependency

    [ https://issues.apache.org/jira/browse/FLINK-6965?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16071454#comment-16071454 ] 

Sebastian Klemke commented on FLINK-6965:
-----------------------------------------

flink-avro connector for DataSet API is also affected: Snappy-encoded avro datasets can neither be read nor written using AvroInputFormat/AvroOutputFormat. Flink runtime built from PR also fixes this.

> Avro is missing snappy dependency
> ---------------------------------
>
>                 Key: FLINK-6965
>                 URL: https://issues.apache.org/jira/browse/FLINK-6965
>             Project: Flink
>          Issue Type: Bug
>          Components: Type Serialization System
>    Affects Versions: 1.3.0
>            Reporter: Chesnay Schepler
>            Assignee: Chesnay Schepler
>            Priority: Blocker
>             Fix For: 1.3.2
>
>
> The shading rework made before 1.3 removed a snappy dependency that was accidentally pulled in through hadoop. This is technically alright, until class-loaders rear their ugly heads.
> Our kafka connector can read avro records, which may or may not require snappy. Usually this _should_ be solvable by including the snappy dependency in the user-jar if necessary, however since the kafka connector loads classes that it requires using the system class loader this doesn't work.
> As such we have to add a separate snappy dependency to flink-core.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)