[
https://issues.apache.org/jira/browse/NIFI-16074?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18121965#comment-18121965
]
Henri Saksi commented on NIFI-16074:
------------------------------------
The timestamp, list and struct parts of this issue were addressed in NIFI-16069
(2.12.0). Byte values are still passed to Iceberg unconverted: writing to a
{{binary}} column fails with {{ClassCastException: [Ljava.lang.Object; cannot
be cast to java.nio.ByteBuffer}} (seen on 2.12.0 with JsonTreeReader and an
Avro {{bytes}} field), and {{fixed}} columns fail similarly. On main (Iceberg
1.12), String values for {{uuid}} columns also fail in the Parquet writer with
{{String cannot be cast to UUID}}, unless the reader already provides
{{java.util.UUID}} values.
The same conversion is now needed for {{geometry}} and {{geography}} columns:
Iceberg 1.12
([apache/iceberg#16982|https://github.com/apache/iceberg/pull/16982], upgraded
in NIFI-16407) writes both from WKB in a {{ByteBuffer}}.
I opened [PR #11749|https://github.com/apache/nifi/pull/11749], which converts
{{binary}}, {{fixed}}, {{uuid}}, {{geometry}} and {{geography}} values by the
target column type in {{RecordConverter}}, including values nested in structs,
lists and maps. Its tests include writing through the Parquet writer to a
format-version 3 table with geometry and geography columns. If you'd prefer the
geospatial types under a separate issue, I can split them out.
> PutIcebergRecord missing convertors
> -----------------------------------
>
> Key: NIFI-16074
> URL: https://issues.apache.org/jira/browse/NIFI-16074
> Project: Apache NiFi
> Issue Type: Bug
> Reporter: Nir Yanay
> Priority: Major
> Time Spent: 20m
> Remaining Estimate: 0h
>
> The following Iceberg types currently don't appear to be handled correctly: *
> {{timestampz}}
> * {{bytes}}
> * {{list}}
> * {{struct}}
> None of these have conversion logic implemented, which results in
> {{ClassCastException}} errors.
> Additionally, the UUID conversion should use {{{}UUID.fromString(){}}}, but
> it currently doesn't have its own converter, although it does not fail like
> the others.
>
> I opened a PR:
> [https://github.com/apache/nifi/pull/11423]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
