Save .hydra/*.yaml before task execution

[binshengliu]

🐛 Bug

Saving the configurations is a really cool idea that helps me check what I actually did. I noticed the save timing is delayed after task execution in the 1.0.0rc1. I have encountered some scenarios that I think it would be better if the configs are saved earlier.

For example,

1. Tasks may exit accidentally due to bugs. Knowing detailed configurations would be helpful.
2. For time-consuming tasks like ML training which may last for several days, it would be helpful to be able to check the config.yaml. This helps me a lot when I have multiple tasks running but I can't remember the real difference. I do diff config.yaml.
3. Some tasks don't need to be finished. The full training may be designed to run 100 epochs but it also happens it's aborted/killed in the middle.

To reproduce

Default behavior in 1.0.0rc1.

** Minimal Code/Config snippet to reproduce **
NA

** Stack trace/error message **
NA

Expected Behavior

NA

System information

• Hydra Version : 1.0.0rc1
• Python version : NA
• Virtual environment type and version : NA
• Operating system : NA

[omry]

Is this a change in 1.0?

[binshengliu]

It seems this change was introduced in 98128a4

[omry]

Interesting, not sure why I moved it down.
can you send a PR fixing it?

[binshengliu]

Yeah sure.