fallback if Spot request cannot be fullfilled

[universam1]

We are facing quite often that AWS is not able to fullfill the spot request - leaving the setup in a broken state.

Status: capacity-not-available: There is no Spot capacity available that matches your request.

Now docker-machine should but does not provide a logic to fall back to on-demand instances in that case.

Please consider this feature as it renders spotting impossible to use in production grade systems

[dhohengassner]

Would love to see such a feature - spot instances could be used for much more use cases

[icereed]

Facing same issue. Would be a cool enhancement!

[buholzer]

+1 - Have the exact same use case. Would be very helpful.

[Adriks976]

There's also the ability to use aws fleet: a list of machine types (m4, m5) with a bid price and have one of this instance type if one is not available.
--machine-machine-options "amazonec2-instance-type=m4.2xlarge,m5.2xlarge"` for example.

[universam1]

@Adriks976 Can you please shed more insight how you are able to make that work?
On my setup this is creating following error:

Invalid value 't3.large,t3.medium' for InstanceType....

[Adriks976]

No I was suggesting to have this feature with a list of instance type. There’s no way for now to handle this. Let’s hope that docker machine will soon deal with this feature.

[fransgidlof]

A proper EC2 Spotfleet is actually a spotfleet request ID: sfr-XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX Which corresponds to a Spotfleet, which dictates VPC, subnet, IAM role etc and instance types.

You can even mix and match on-demand and spot in the same fleet, it would solve this original issue's problem of exhausted spot-instances. (Or try to anyway)

https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/spot-fleet.html

[spider-network]

You can even mix and match on-demand and spot in the same fleet, it would solve this original issue's problem of exhausted spot-instances. (Or try to anyway)

https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/spot-fleet.html

+1