[SUPPORT] HoodieStreamer: HoodieKeyException error when using SQL Transformer about hudi

[YousifS7]

Hi @wombatu-kun,

I wanted to provide an update on the issue. After some experimentation with the hoodie.streamer.schemaprovider.registry.targetUrl configuration, I was able to resolve the problem. Initially, I assumed that Spark would infer the schema from the transformed data. However, I registered a custom flattened schema in my schema registry to match the desired output and utilized the hoodie.streamer.schemaprovider.registry.targetUrl setting. Once I ran the job with these configurations, it worked.

Thank you

[wombatu-kun]

Try CACHE TABLE dbz_fields AS SELECT ts_ms, source_fields FROM dbz_events;. (source_fields without .*)

[YousifS7]

Hi @wombatu-kun, Thank you for getting back to me.

I have modified the SQL File to the following:

CACHE TABLE dbz_filtered AS
SELECT ts_ms, op, before, after FROM <SRC> WHERE op IN ('d', 'u', 'c', 'r');

CACHE TABLE dbz_events AS
SELECT ts_ms, CASE WHEN op = 'd' THEN before ELSE after END AS source_fields, CASE WHEN op = 'd' THEN true ELSE false END AS is_deleted FROM dbz_filtered;

CACHE TABLE dbz_fields AS
SELECT ts_ms, source_fields FROM dbz_events;

SELECT source_fields.*, Concat(source_fields.col1, source_fields.col2) AS hudi_key, YEAR(FROM_UNIXTIME(source_fields.col2 / 1000)) AS partition_path FROM dbz_fields;

I'm still getting the same error:

HoodieKeyException: recordKey value: "null" for field: "hudi_key" cannot be null or empty.

Not sure if I'm missing something else.

Thank you

[wombatu-kun]

what if you try this?

CACHE TABLE dbz_fields AS
SELECT ts_ms, source_fields FROM dbz_events;

SELECT s.*, Concat(s.col1, s.col2) AS hudi_key, YEAR(FROM_UNIXTIME(s.col2 / 1000)) AS partition_path FROM dbz_fields s;

[YousifS7]

Trying the suggested SQL is giving the following error:

Exception in thread "main" org.apache.spark.sql.AnalysisException: [UNRESOLVED_COLUMN.WITH_SUGGESTION] A column or function parameter with name `s`.`col1` cannot be resolved. Did you mean one of the following? [`s`.`ts_ms`, `s`.`source_fields`].

[YousifS7]

Following up on this, I was able to run the spark app without errors, however, even when specifying columns that are in the 'before' and 'after' Struct type extracted from Kafka, only the following columns were written to the Hudi table:

ts_ms, op, before, after, source

Transformer File Query

CACHE TABLE dbz_filtered AS
SELECT a.ts_ms, a.op, a.before, a.after, a.source FROM <SRC> a WHERE a.op IN ('d', 'u', 'c', 'r');

SELECT b.before.col1 AS before_col1, b.after.col1 AS after_col1, b.ts_ms, b.op, b.before, b.after, b.source FROM dbz_filtered b;

It seems the additional columns extracted from the envelopes 'before' and 'after' are being ignored for some reason. I even tried to specify a static columns like: 'test' AS test_columns, it was not written on the Hudi tables.

Properties File

hoodie.datasource.write.keygenerator.class=org.apache.hudi.keygen.ComplexKeyGenerator
hoodie.datasource.write.recordkey.field=op
hoodie.datasource.write.partitionpath.field=op
hoodie.datasource.write.precombine.field=ts_ms
hoodie.streamer.transformer.sql.file=s3://some_bucket/configs/some_table.sql
hoodie.streamer.schemaprovider.registry.url=http://localhost:8081/subjects/test.dbo.some_table-value/versions/latest
schema.registry.url=http://localhost:8081
hoodie.streamer.source.kafka.topic=test.dbo.some_table
bootstrap.servers=localhost:9092
auto.offset.reset=earliest

Spark Submit

spark-submit 
--packages org.apache.hudi:hudi-spark3.5-bundle_2.12:0.15.0,org.apache.hudi:hudi-aws-bundle:0.15.0
--conf spark.streaming.kafka.allowNonConsecutiveOffsets=true 
--conf spark.serializer=org.apache.spark.serializer.KryoSerializer 
--class org.apache.hudi.utilities.streamer.HoodieStreamer s3://some_bucket/jars/hudi-utilities-slim-bundle.jar 
--schemaprovider-class org.apache.hudi.utilities.schema.SchemaRegistryProvider 
--props s3://some_bucket/configs/some_table.properties 
--source-class org.apache.hudi.utilities.sources.AvroKafkaSource 
--enable-sync 
--source-ordering-field ts_ms 
--target-base-path s3://some_bucket/some_folder/some_table 
--target-table dbo.some_table 
--table-type COPY_ON_WRITE 
--sync-tool-classes org.apache.hudi.aws.sync.AwsGlueCatalogSyncTool 
--op BULK_INSERT 
--transformer-class org.apache.hudi.utilities.transform.SqlFileBasedTransformer

Is there something else that I might be missing?

Thank you

[wombatu-kun]

As I see from documentation here https://hudi.apache.org/docs/hoodie_streaming_ingestion#schema-providers

By default, Spark will infer the schema of the source and use that inferred schema when writing to a table...

May be you should add target schema with additional fields to schema registry and provide Url to that schema in config hoodie.streamer.schemaprovider.registry.targetUrl

[YousifS7]

@wombatu-kun Couldn't that be extracted from the Transformer?

SqlFileBasedTransformer

As I understand it, the Transformer sits between the Source and Target. For some reason my additional fields are being ignored.
I included several fields into my SQL Transformer but none of the additional fields were written to the Hudi table.

Thank you