[ 
https://issues.apache.org/jira/browse/FLINK-8716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16370052#comment-16370052
 ] 

Arvid Heise commented on FLINK-8716:
------------------------------------

In the email thread, [~till.rohrmann] also pointed out that 
cancel-with-snapshot always produces full checkpoints. So within a snapshot 
only one serializer version is used.

The question should be: what happens when we do an incremental checkpoint 
afterwards. Then we probably need two different serializer configurations for 
the same descriptor. That sounds doable (use descriptor name/checkpoint name as 
the key of the meta information instead of just the descriptor name) but may 
require larger refactoring.

> AvroSerializer does not use schema of snapshot
> ----------------------------------------------
>
>                 Key: FLINK-8716
>                 URL: https://issues.apache.org/jira/browse/FLINK-8716
>             Project: Flink
>          Issue Type: Bug
>    Affects Versions: 1.4.0
>            Reporter: Arvid Heise
>            Priority: Major
>
> The new AvroSerializer stores the schema in the snapshot and uses it to 
> validate compability.
> However, it does not use the schema of the snapshot while reading the data. 
> This version will fail for any change of the data layout (so it supports more 
> or less only renaming currently).
>  
> [https://github.com/apache/flink/blob/f3a2197a23524048200ae2b4712d6ed833208124/flink-formats/flink-avro/src/main/java/org/apache/flink/formats/avro/typeutils/AvroSerializer.java#L265]
>  needs to use the schema from
>  
> [https://github.com/apache/flink/blob/f3a2197a23524048200ae2b4712d6ed833208124/flink-formats/flink-avro/src/main/java/org/apache/flink/formats/avro/typeutils/AvroSerializer.java#L188]
>  as the first parameter. Accordingly, a readSchema field need to be set
>  in #ensureCompatibility and relayed in #duplicate. Note that the readSchema 
> is passed as the write schema parameter to the DatumReader, as it was the 
> schema that was used to write the data.



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

Reply via email to