Recently, I asked my subscribers what topics are interesting t | CatOps

Recently, I asked my subscribers what topics are interesting to them and a few people mentioned observability.

That’s funny, ‘coz yesterday I accidentally bumped into a great series of articles on setting SLAs for your products by Alex Ewerlöf!

- Calculating composite SLA - truly outstanding read!
- Some practical advice when setting SLA - notice, it says SLA, not SLO. So, there are some business related tips in this article as well. However, the core is technical, ofc.
- Calculating the SLA of a system behind a CDN - I haven’t read this one yet. But given the quality of previous two, I expect this one be great as well!

tl;dr for the first article in the list:

for serial, multiply availability; For parallels, multiply unavailability


I would personally also add that when you try to set a “full” SLO(A) for your service, that is also a composite SLO(A). You should treat it as a serial. For example, if you have 99.8% error rate SLO and 99.1% latency SLO, an “overall” SLO would be 0.998 0.991 100% = 98.9%

That’s not only good to know, but you may also want to write your marketing materials differently. There is a difference between:

> We guarantee 99.8% SLO on 5th error rate and 99.1% SLO on requests not taking longer than X milliseconds.

And

> We guarantee the 98.9% availability of our systems.

I’m not a marketing person, though. I don’t know what’s better. What I do know is that:”Nines doesn’t matter, if your users are unhappy”.

#observability #slo #sla