Steve,
thanks a lot for looking into this closer!
Let's discuss the resolution of the issue in the ticket Dawid has created:
https://issues.apache.org/jira/browse/FLINK-15941
Best,
Robert
On Thu, Feb 6, 2020 at 6:59 PM Steve Whelan wrote:
> Robert,
>
> You are correct that it is using a *Cache
Robert,
You are correct that it is using a *CachedSchemaRegistryClient* object.
Therefore, *schemaRegistryClient.*register() should be checking the cache
first before sending a request to the Registry. However, turning on debug
logging of my Registry, I can see a request being sent for every seria
Hi Steve,
I think your observation is correct. If I am not mistaken we should use
*schemaRegistryClient.getId(subject, schema); *instead of**
**schemaRegistryClient.register(subject, schema);. **The former should
perform an http request only if the schema is not in the cache.
I created an issue t
Hi,
thanks a lot for your message. It's certainly not intentional to do a HTTP
request for every single message :)
Isn't the *schemaRegistryClient *an instance of CachedSchemaRegistryClient,
which, as the name says, caches?
Can you check with a debugger at runtime what registry client is used, and
Hi,
I'm running Flink v1.9. I backported the commit adding serialization
support for Confluent's schema registry[1]. Using the code as is, I saw a
nearly 50% drop in peak throughput for my job compared to using
*AvroRowSerializationSchema*.
Looking at the code, *RegistryAvroSerializationSchema.se
