You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently the restore portion of Guardian only uses a standard Kafka producer using Alpakka's Producer. While this is fine for standard scenarios most people when doing a restore would ideally want exactly once semantics for making a restore. This seems possible to do using the standard Producer with
However such a configuration has terrible throughput since you are essentially forcing the Kafka producer to send one message at a time with no batching
What is currently missing?
Currently the restore portion of Guardian only uses a standard Kafka producer using Alpakka's
Producer
. While this is fine for standard scenarios most people when doing a restore would ideally want exactly once semantics for making a restore. This seems possible to do using the standard Producer withHowever such a configuration has terrible throughput since you are essentially forcing the Kafka producer to send one message at a time with no batching
How could this be improved?
Support needs to be done in Alpakka Kafka so that its possible to create a
Transactional.source
from a non Kafka cluster source. The currentTransactional.sink
only works with aPartitionOffsetCommittedMarker
which is currently only created with aTransactional.source
. Trying to manually create aPartitionOffset
from a differentSource
just results in an exception being thrown at https://github.com/akka/alpakka-kafka/blob/1a1dab1e5168a829b7e76053375a364739043850/core/src/main/scala/akka/kafka/internal/TransactionalProducerStage.scala#L220 and its not possible to manually create aPartitionOffsetCommittedMarker
since its private/internal APIThere is currently an ongoing issue at Alpakka Kafka at akka/alpakka-kafka#1075 for this
Is this a feature you would work on yourself?
The text was updated successfully, but these errors were encountered: