[Feature]: Support Lambda Labs backend type

[r4victor]

Problem

Lambda Labs is a GPU cloud provider with one of the best offerings on the market (https://fullstackdeeplearning.com/cloud-gpus/). We want to support Lambda Labs as a dstack backend.

Solution

Lambda Labs offers only compute; no object storage, secrets manager, logging needed for dstack current backend implementation to work. So we implement Lambda Labs + AWS backend: AWS is used for object storage, secrets manager, logging. And Lambda is used only for compute. Users have to configure both Lambda Labs and AWS, but they have all the dstack features working. Later, we can allow users to configure other clouds for object storage, etc.

Benefit

More backends is better!

Alternatives

Would you like to help contributing this feature?

Yes

[r4victor]

Another issue with Lambda Labs is that they don't support custom VM images, so we cannot bring docker and dstack images there – they have to be installed on the instance launch.

Lambda Labs say their instances are slow to launch, so a longer installation may not be such a big problem:

Single-GPU instances usually take 3-5 minutes to launch. Multi-GPU instances usually take 10-15 minutes to launch.