Failing to identify leader when connecting to Azure EventHubs

[justinmchase]

Azure EventHubs has a "kafka projection" mode. It looks a lot like kafka but has some subtle differences I think.

When attempting to connect to eventhubs akhq has an error:

2023-02-03 22:53:57,298 [1 31mERROR[0 39m [35mr-thread-5[0 39m [36mi.m.h.server.RouteExecutor[0 39m Unexpected error occurred: Error encoding object [org.akhq.utils.ResultPagedList@691ab908] to JSON: Leader not found (through reference chain: org.akhq.utils.ResultPagedList["results"]->org.akhq.utils.PagedList[0]->org.akhq.models.Topic["partitions"]->java.util.ArrayList[0]->org.akhq.models.Partition["leader"])

Here is my configuration:

application.yaml

akhq:
  connections:
    kafka:
      properties:
        bootstrap.servers: "recovery-{{ .Values.global.envName }}.servicebus.windows.net:9093"
        security.protocol: SASL_SSL
        sasl.mechanism: PLAIN
        sasl.jaas.config: org.apache.kafka.common.security.plain.PlainLoginModule required username="$ConnectionString" password="{{ .Values.global.AZURE_EVENT_HUBS_ENDPOINT }}";

https://github.com/tchiotludo/akhq/blob/dev/src/main/java/org/akhq/models/Partition.java#L40-L46

I believe this is the problem code, I belive that EventHubs is reporting that there are no replicas therefore the code to identify the leader is failing.

    public Node.Partition getLeader() {
        return nodes
            .stream()
            .filter(Node.Partition::isLeader)
            .findFirst()
            .orElseThrow(() -> new NoSuchElementException("Leader not found"));
    }

Since there are no replicas, the .findFirst() finds nothing. I'm assuming the fix will be to either add the partition itself to the nodes list in the case where there are no replicas, or to return this.id in the getLeader() function instead of throwing an exception.

[tchiotludo]

Event hub support can be cool, but I think it will break on other edge case, PR are welcome, closing due to duplicate of #657

[justinmchase]

Can you elaborate on what other edge cases you're worried about? It seems like the fix to make it work with event hubs would be pretty simple and I can test it against local single-instance and also GCP cluster. But if you were to elaborate on edge cases I could probably try to take that into consideration.

[tchiotludo]

mostly I think on every specific feature like topic config / node config / compaction / etc ... that could not be implemented on event hub

[justinmchase]

Ok as in, if this were fixed it may still experience more issues with EventHubs? You're not necessarily worried about this fix breaking non-EH systems?

[tchiotludo]

maybe you will see other issues with EH, but I'm not worried at all for all non EH broker, exactly