[topicmappr] handle larger metrics data in ZooKeeper
jamiealquiza opened this issue · comments
Currently, topicmappr operations that rely on metrics data (any rebalance command or rebuild with the storage placement option) read data from ZooKeeper. It's desirable to keep this input as the canonical source of this data for now.
One issue is that the current structure just blobs all of the topic/partition size data as a single json key, which can be quite large in clusters with many topics/partitions. We should version the data structure format and introduce a new type that utilizes compression or other ways to cope with very large inputs.
Init testing metrics data compression: https://github.com/DataDog/kafka-kit/compare/jamie/metrics-compression