Speculative Execution Metrics

[fee-mendes]

Currently, Speculative Execution can be observed in two ways:

• Query Tracing,
• speculative execution fibers via Query History

It would be nice to also have Speculative Execution numbers also exposed in Metrics. For example, the CPP driver increments a counter each time a new fiber is started, and from there allows one to infer the percentage of total queries which triggered speculative execution.

For each Speculative Execution, it would also be nice to keep track of their latencies in a separate histogram, similarly as what CPP does in Histogram speculative_request_latencies;

[siningma]

Would be nice if the counter count be split by read and writes queries.