kafka-cluster partitions don't necessarily match kafka-mdm-in partitions

[shanson7]

Data comes into Metrictank on the data topic (let's call that mdm) and when chunks are persisted to cassandra, summaries of the chunk are sent to the persist topic (let's call that persist). On start up, MT uses persist to avoid overwriting chunks in cassandra that were already persisted. This means that it is important that the summaries in persist line up with the data in mdm for the instance handling a given partition.

1. MT doesn't really enforce how the data in mdm is partitioned, just that partitioning is consistent.
2. Partitioning in persist is either "ByOrg" or "BySeries" (see here and here). If this isn't how data in mdm is partitioned, there is trouble.

It seems to me that the simpler solution is to simply use def.Partition to put the summaries into persist.

I am rolling out this change on our side and can submit a cleanup PR. I'm not sure if removing the parameter entirely is backwards compatible or not, however.

[Dieterbe]

It seems to me that the simpler solution is to simply use def.Partition to put the summaries into persist

+1. more robust and more efficient probably

I'm not sure if removing the parameter entirely is backwards compatible or not, however.

the parameters purpose is to try to make sure partitioning is consistent. your approach does that more accurately, so is better. parameter can be dropped.

[woodsaj]

+1 this seems like a much better approach.

I think the current approach was used as we were not storing the partition info at the time the feature was added.