How to deploy code changes across microservices?

Question

We have 10 instances of same microservice identified via eureka service discovery and calls being routed to them through gateway. We want to deploy code changes across these 10 instances but the code changes should be atomic. Meaning at no point of time, 2 instances be running different code. The simple strategy could be to bring down 9 of the instances--> deploy changes on them --> bring them up --> bring down remaining one instance and after deployment change, bring it up again.

Is this the ideal strategy to be followed on production environment or are there specific patterns to be followed? The answers on blogs seems to discuss the microservices pattern but none talk about the scenario when some of the instances have newer code version and others yet to be updated.

Answer 1

The ideal strategy is to spin up a few new instances and start balancing requests to them progressively. The load balancer can do IP address pinning so that starting at a particular point in time, an IP address only gets replies from the new instances.

Answer 2

In ideal production world; your atomic requirement is NOT there... Generally we deploy new code on suppose 10% on servers.. see how it is performing in terms of exceptions, latency numbers ..and if all good we keep increasing this percentage..

But I completely understand for some releases ( for example some DB changes though there is even solution for that but that is for another what if ) or for some scenarios we CANNOT have multiple code bases running. First question to be asked for any deployment is "allowed downtime".

Let us assume u need minimum downtime... then solution is that u deploy on another 10 servers; test them out .. and once all is ok , then point your ELB to new servers.. Note that there will be few minutes downtime here..as we have atomic requirement.