Description
At the moment, system reads are counted as regular reads and displayed in the metrics accordingly.
While this is technically correct (these are CQL reads after all), it can result in some confusion when looking at the metrics - especially due to the fact that system reads are always routed to Origin regardless of which cluster is set as primary. This means that, for example, a user who configures the proxy to use Target as primary and then issues some reads will still see read activity going to Origin due to metadata reads etc.
To improve this we have a few options:
- Not track system reads at all (although this is not really accurate - system reads are CQL reads that are handled by the proxy).
- Track them into a separate time series that is still represented in the Reads graphs but is separated out from the reads to regular tables. This would definitely be a clearer way to represent them.
- Create a new type of metrics specifically for system reads and represent them on dedicated graphs. This would probably be the most correct solution but requires a bit more work.
Whichever option we go for, we should also add some information (perhaps in the Troubleshooting section) about system reads and how they are reflected in the metrics.
┆Issue is synchronized with this Jira Task by Unito
┆Components: Proxy
┆Issue Number: ZDM-542