[
https://issues.apache.org/jira/browse/HUDI-2348?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17405981#comment-17405981
]
ASF GitHub Bot commented on HUDI-2348:
--------------------------------------
pratyakshsharma commented on a change in pull request #3485:
URL: https://github.com/apache/hudi/pull/3485#discussion_r697657292
##########
File path: website/blog/2021-08-16-kafka-custom-deserializer.md
##########
@@ -0,0 +1,47 @@
+---
+title: "Schema evolution with DeltaStreamer using KafkaSource"
+excerpt: "Evolve schema used in Kafkasource of DeltaStreamer to keep data up
to date with business"
+author: sbernauer
+category: blog
+---
+
+The schema used for data exchange between services can change change rapidly
with new business requirements.
+Apache Hudi is often used in combination with kafka as a event stream where
all events are transmitted according to an record schema.
+In our case a Confluent schema registry is used to maintain the schema and as
schema evolves, newer versions are updated in the schema registry.
+<!--truncate-->
+
+## What do we want to achieve?
+We have multiple instances of DeltaStreamer running, consuming many topics
with different schemas ingesting to multiple Hudi tables. Deltastreamer is a
utility in Hudi to assist in ingesting data from multiple sources like DFS,
kafka, etc into Hudi. If interested, you can read more about DeltaStreamer tool
[here](https://hudi.apache.org/docs/writing_data#deltastreamer)
+Ideally every Topic should be able to evolve the schema to match new business
requirements. Consumers start producing data with a new schema version and the
DeltaStreamer picks up the new schema and ingests the data with the new schema.
For this to work, we run our DeltaStreamer instances with the latest schema
version available from the Schema Registry to ensure that we always use the
freshest schema with all attributes.
+A prerequisites it that all the mentioned Schema evolutions must be
`BACKWARD_TRANSITIVE` compatible (see [Schema Evolution and Compatibility of
Avro Schema
changes](https://docs.confluent.io/platform/current/schema-registry/avro.html).
This ensures that every record in the kafka topic can always be read using the
latest schema.
+
+
+## What is the problem?
+The normal operation looks like this. Multiple (or a single) producers write
records to the kafka topic.
+In regular flow of events, all records are in the same schema v1 and is in
sync with schema registry.
+<br/>
+Things get complicated when a producer switches to a new Writer-Schema v2 (in
this case `Producer A`). `Producer B` remains on Schema v1. E.g. a attribute
`myattribute` was added to the schema, resulting in schema version v2.
+Deltastreamer is capable of handling such schema evolution, if all incoming
records were evolved and serialized with evolved schema. But the complication
is that, some records are serialized with schema version v1 and some are
serialized with schema version v2.
+
+<br/>
+The default deserializer used by Hudi
`io.confluent.kafka.serializers.KafkaAvroDeserializer` uses the schema that the
record was serialized with for deserialization. This causes Hudi to get records
with multiple different schema from the kafka client. E.g. Event #13 has the
new attribute `myattribute`, Event #14 dont has the new attribute
`myattribute`. This makes things complicated and error-prone for Hudi.
+
+<br/>
+
+## Solution
+Hudi added a new custom Deserializer `KafkaAvroSchemaDeserializer` to solve
this problem of different producers producing records in different schema
versions, but to use the latest schema from schema registry to deserialize all
the records.<br/>
+As first step the Deserializer gets the latest schema from the Hudi
SchemaProvider. The SchemaProvider can get the schema for example from a
Confluent Schema-Registry or a file.
+The Deserializer then reads the records from the topic with the schema the
record was written. As next step it will convert all the records to the latest
schema from the SchemaProvider, in our case the latest schema. As a result, the
kafka client will return all records with a unified schema i.e. the latest
schema as per schema registry. Hudi does not need to handle different schemas
inside a single batch.
Review comment:
from the topic with the schema the record was written -> from the topic
using the schema the record was written with? IMHO this makes the statement
more clear :)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
> Publish a blog on schema evolution with KafkaAvroCustomDeserializer
> -------------------------------------------------------------------
>
> Key: HUDI-2348
> URL: https://issues.apache.org/jira/browse/HUDI-2348
> Project: Apache Hudi
> Issue Type: Improvement
> Reporter: sivabalan narayanan
> Assignee: sivabalan narayanan
> Priority: Major
> Labels: pull-request-available
> Fix For: 0.10.0
>
>
> Publish a blog on schema evolution with KafkaAvroCustomDeserializer
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
