RFC: support sampling rate for querylog

[timvaillancourt]

The querylog feature can generate a lot of logs. Today, my understanding is the only knob to control how often a query is logged is this flag:

--querylog-filter-tag string string that must be present in the query for it to be logged; if using a value as the tag, you need to disable query normalization

This means a user that wants to log X % of query traffic needs to handle the sampling rate from the client side, by adding a filter tag to X % of queries

In practice this can mean a team supporting Vitess in production may need to ask an application team to change configs/code, just to modify the sampling rate of queries. Finally, what if an app doesn't/forgets-to set the filter tag? This would mean no samples for some use cases

This RFC proposes a new querylog flag (vtgate and vttablet) such as --querylog-sample-rate 0.0 or --querylog-sampling-rate 0.0 to specify a rate vtgate/vttablet should log queries to querylog, regardless of if the --querylog-filter-tag was provided in the query. This would work in addition to the --querylog-filter-tag functionality, ie: queries with X tag are still logged if both flags are set

This would provide:

1. Full control of the sampling rate of querylog from the Vitess-side
2. More certainty the querylog contains all workloads (not just workloads adding a filter tag)

Use Case(s)

• A Vitess user that would like to capture a sample of all workloads hitting vtgate/vttablet
• A Vitess user that would like to adjust the rate of querylog samples without having to make client-side changes to the --querylog-filter-tag