Support for multiple bootstrap_uri does not work

[nvp152]

What can we help you with?

When running kafka in high availability mode, one would typically set up one's clients with a list of bootstrap uri so that the client can cycle through them and find the first one that is willing to handover its broker topology info. It does not appear that Karapace supports this. I can see that if I pass a list karapace will split the string and operate on the first bootstrap uri, however, i don't see it cycling through the various entries.

Am I missing something?

[tvainika]

Karapace passes bootstrap_uri as a string to underlying libraries, ie aiokafka and kafka-python. Those support comma separated list e.g. server1:9092,server2:9092 as a list of bootstrap servers. However after that it is randomly picked which server to connect first. Does this help your case?

[nvp152]

@tvainika - so it does answer the question that multiple bootstrap_uris are supposed to work. I wanted to make sure that was the case before I reported a bug. So bug there is - if I send a list of 2 bootstrap uris:

- name: KARAPACE_BOOTSTRAP_URI
  value: mykafka-controller-1.mykafka-controller-headless.default.svc.cluster.local:9092,mykafka-controller-0.mykafka-controller-headless.default.svc.cluster.local:9092

and one of them is down, karapace keeps cycling and kafka clients that try to connect to it get this error

 io.confluent.kafka.schemaregistry.client.rest.exceptions.RestClientException: Error; error code: 50005

The log cycling i am referring to looks like this:

kafka.conn              schema-reader   WARNING         DNS lookup failed for mykafka-controller-1.mykafka-controller-headless.default.svc.cluster.local:9092, exception was [Errno -2] Name or service not known. Is your advertised.listeners (called advertised.host.name before Kafka 9) correct and resolvable?
kafka.conn              schema-reader   ERROR           DNS lookup failed for mykafka-controller-1.mykafka-controller-headless.default.svc.cluster.local:9092 (AddressFamily.AF_UNSPEC)
karapace.utils          kafka-python-producer-1-network-thread  INFO            Node id bootstrap-1 no longer in cluster metadata, closing connection and requesting update
kafka.conn              kafka-python-producer-1-network-thread  WARNING         DNS lookup failed for mykafka-controller-1.mykafka-controller-headless.default.svc.cluster.local:9092, exception was [Errno -2] Name or service not known. Is your advertised.listeners (called advertised.host.name before Kafka 9) correct and resolvable?
kafka.conn              kafka-python-producer-1-network-thread  ERROR           DNS lookup failed for mykafka-controller-1.mykafka-controller-headless.default.svc.cluster.local:9092 (AddressFamily.AF_UNSPEC)
karapace.utils          schema-reader   INFO            Node id bootstrap-1 no longer in cluster metadata, closing connection and requesting update