Response code for latency based SLO? | Voters

Response code for latency based SLO?

Jerome Comptdaer

What are the http response codes considered by the latency-based SLO ? Does the latency-based SLO consider successful responses only (< 500) or any response code?

Also, how the SLI is affected by non-running service ? (the service is down and not responding to requests)

November 13, 2020

Andrew Seigner

Currently latency-based SLOs are calculated irrespective of http response code or success status. We'd like to make this more flexible. Are there particular queries or options you'd like to see?

Re: a service being down, it depends. If it's still reporting metrics, and it is receiving traffic but just failing requests, we will show 0% success rate. If it's receiving 0 requests, or if metrics are not being collected, we will show "--" for success rate.

For reference, Success Rate SLI calculations are essentially:

sum(
	rate(
		response_total{direction="inbound", classification="success"}[1m]
	)
)
/
sum(
	rate(
		response_total{direction="inbound"}[1m]
	)
)

...and for Latency:

sum(
	rate(
		response_latency_ms_bucket{direction="inbound"}[1m]
	)
) by (le)

Jerome Comptdaer

Andrew Seigner: 
Thank you for the provided details.
Overall I would like the SLO to capture the actual user experience as close as possible. For my use case, I see this as a combination of both successful responses (<500) and latency threshold.

Jean Sandberg

Jerome Comptdaer: I think you're saying you want to know the % of requests that are both successful and meet a latency threshold?