You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@nifi.apache.org by "Mike Thomsen (Jira)" <ji...@apache.org> on 2022/03/25 13:43:00 UTC

[jira] [Resolved] (NIFI-6204) Avro with a default field of type long fails with ClassCastException

     [ https://issues.apache.org/jira/browse/NIFI-6204?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Mike Thomsen resolved NIFI-6204.
--------------------------------
    Resolution: Fixed

Ran a simple flow based on this schema, and it worked just fine.

> Avro with a default field of type long fails with ClassCastException
> --------------------------------------------------------------------
>
>                 Key: NIFI-6204
>                 URL: https://issues.apache.org/jira/browse/NIFI-6204
>             Project: Apache NiFi
>          Issue Type: Bug
>            Reporter: Lars Francke
>            Priority: Minor
>         Attachments: Avro-issue.xml, Screen Shot 2019-04-11 at 00.33.51.png, Screen Shot 2019-04-11 at 00.38.31.png
>
>
> I get this error:
> {code}org.apache.avro.file.DataFileWriter$AppendWriteException: java.lang.ClassCastException: java.lang.Integer cannot be cast to java.lang.Long{code}
> I've attached an example workflow plus screenshots.
> Basically: Converting a schema like this
> {code}
> {
>  "name": "foobar",
>  "type": "record",
>  "fields": [
>    { "name": "a", "type": "string" }
>  ]
> }
> {code}
> into this
> {code}
> {
>  "name": "foobar",
>  "type": "record",
>  "fields": [
>    { "name": "a", "type": "string" },
>    { "name": "b", "type": "long", "default": 100 }
>  ]
> }
> {code}
> fails because of this snippet in {{AvroTypeUtil}}:
> {code}
>         // see if the Avro schema has any fields that aren't in the RecordSchema, and if those fields have a default
>         // value then we want to populate it in the GenericRecord being produced
>         for (final Field field : avroSchema.getFields()) {
>             final Optional<RecordField> recordField = recordSchema.getField(field.name());
>             if (!recordField.isPresent() && rec.get(field.name()) == null && field.defaultVal() != null) {
>                 rec.put(field.name(), field.defaultVal());
>             }
>         }
> {code}
> This relies on {{field.defaultVal()}} returning the correct type.
> The field that this relies on is a {{JsonNode}} from Jackson. So it parses the Schema and the "default" field is parsed into an {{IntNode}} which is then put into the record and that is not a Long so the cast fails when Avro tries to write it later.
> Now Jackson has no way of knowing that this is supposed to be a LongNode instead of an IntNode. I'm not sure if Avro should catch this or NiFi and I also have no good idea on how to fix this.



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