I noticed that when a new instance is spawning some request are served very slowly. This is a sign that the starting instance is receiving requests. Shouldn’t an instance be considered as good to handle requests as soon as its status move to healthy? Is my supposition correct? This can be a huge problem because, if an instance is starting because of load, lot of user will hit it having slow response times
This is related to a larger problem we have with service discovery: how do fly.io deploys work? - #2 by kurt
fly scale count 3, deploys should have no slow requests. Single VM deploys should work better but we haven’t solved that issue yet.
Ok, thanks for th the reply. I’m going to increase the count to have a minimum of 3 instances and check again. How do I know when this is solved? Do you have any official channel? should i subscribe to the other thread’
Also, you mention the usage of some
fly deploy --strategy canary. Is this something possible? Would it mitigate the problem?
canary deploy should help a little. A canary deploy starts a brand new VM before shutting down any old ones. This buys a little extra time for service discovery to catch up.