A recent change to some the cirrusSearchElasticaWrite caused it to submit a LOT more jobs at once. EventGate and Kafka are handling receiving the messages fine, but MirrorMaker is lagging when these jobs get submitted.
For high volume topics like this, we need more partitions. I believe EventGate's node-rdkafka prodcuer will just do the right thing if we increase the partitions, but I want to check with JobQueue folks (@Pchelolo) first.
Additionally, Erik told me that change-prop is repartitioning this and other high volume topics by some key. Could we just partition by key in the first place? We'd need EventGate and stream config to support specifying the partition key from the event data, but I think we could do this!