runtask failure : InvalidFleetConfig about tibanna HOT 6 CLOSED

It will always launch only one instance. If you are not using spot, it will automatically launch the cheapest instance type from the list that satisfies your requirements. If you are using spot, it will automatically choose the instance type with the highest spot availability (and therefore the one with the best chances to successfully run the job)

from tibanna.

Hm... interesting. It has never failed at this step before. Unfortunately, we are not logging the fleet configuration at this step (we should). Is it using Benchmark to get a possible list of EC2s? Which instance types is it trying to launch? Can you set one manually?

from tibanna.

I think I found the issue. There are duplicate entries in our Benchmark EC2 list. See, e.g., r6a.32xlarge here:
https://github.com/4dn-dcic/Benchmark/blob/master/src/Benchmark/aws/Amazon%20EC2%20Instance%20Comparison.csv
@clarabakker Please have a look as soon as possible.

from tibanna.

I believe it is using benchmark. I'm not running it with a specific instance_type, just a mem and cpu requirement. I am able to launch 1 TB nodes if I specify the instance type. I'm not sure exactly what instance_type the benchmark system identifies and tries to use.

from tibanna.

Thanks for clarifying. This confirms the issue. If you set mem and cpu, Benchmark will pick all instances that fit your requirements from the list above. But of course the list should not contain duplicate entries or the create_fleet command will fail. As a workaround you can set the instance type directly (as you did). We will clean up the list and release a new version as soon as possible.

from tibanna.

Does create_fleet spawn all ec2 instances that satisfy requirements? If I only have one job, how does it pick which node in that 'fleet' it gets sent to? What happens to the nodes that don't receive the job?

from tibanna.