GitHub actions downtime is becoming painful for us. Having been lured on there with 10,000 included minutes which they shortly thereafter dropped to 3,000 I feel aggrieved paying for overages incurred from actions regularly shitting the bed.
It's gotten more reliable over time (especially selenium events being dropped on the floor causing tests to stall and fail), but I used to have to babysit it quite a bit and there were quite a number of times where IE instances just would not spool up (with a multiple minute timeout set). Sometimes it was a one-shot thing, other times it went on for hours.
During these incidents the average allocation times listed on their status page would double for Windows VMs (I don't recall the exact numbers but they were on the order of 10 seconds vs 5) but nothing would be red, and most of the time nothing ever did go red.
And that's what you get for using averages for things and divide infinity by n improperly.
The only reason we're using it is because it's free..