request: process groups in a single yaml file

[colemickens]

Feature Request

We have different scenarios for which we run our app:

• basic setup
• load balanced setup with two app instances, and a reverse proxy

For now, we add in default-disabled commands, or copy/paste/edit the process-compose file to enable these various scenarios.

However, I can imagine a concept of "process groups" (a group of pre-declared tasks maybe), that would allow us to DRY and have these scenarios/configs declared in a single YAML file

Use Case:

• Single-file config, removing duplication from similar configurations

Proposed Change:

• Add another layer/concept to enable named groups of processes/tasks inside the yaml file to be run by user choice.

Who Benefits From The Change(s)?

Users

Alternative Approaches

Lots of separate files, using advanced yaml features inclusions, etc. All of these are subpar options, in my view.

[F1bonacc1]

Hi @colemickens,

Have you considered using the merge functionality of process compose? https://f1bonacc1.github.io/process-compose/merge/

[thenonameguy]

Agreed with @F1bonacc1, that should be the correct way to do it. So in the end you do something like:

process-compose --tui=false up -f process-compose.yml -f process-compose.load-balanced.yml

[colemickens]

While I appreciate the answer, the UX for all of this feels... undesirable, compared to what I was suggesting. I'm thinking about hacking on this myself and seeing what I come up with. Or maybe an alternative approach as a zellij plugin. I really don't mean to sound ungrateful and I wish I could better articulate why I feel the way I do about the UX. Maybe I'll come up with some more examples and try out how it feels.

[stijnveenman]

i think I have a similar situation. I have multiple vertical slices that we run separately (we use process-compose for our local dev env). So we define multiple tasks for each vertical slice. With an 'empty' command, which could just be echo "done", or in our case, open the web page of the vertical slice when startup is complete.

  group1:
    command: open http://localhost:1080/group1
    depends_on:
      process1: { condition: process_started }
      process2: { condition: process_started }

  group2:
    command: open http://localhost:1080/group2
    depends_on:
      process1: { condition: process_started }
      process3: { condition: process_started }

could this be an alternative approach? this allows for a simple process-compose up group1 or both process-compose up group1 group2 while allowing processes to still be used in multiple groups (unlike namespaces for example)

[F1bonacc1]

Another option that might work for you, in case you prefer to use a single file, is to use the --namespace flag that will start only the specified list of namespaces.

process-compose -n basic -n common

And for the LB configuration:

process-compose -n load-balancer -n common

[thenonameguy]

While I appreciate the answer, the UX for all of this feels... undesirable, compared to what I was suggesting. I'm thinking about hacking on this myself and seeing what I come up with. Or maybe an alternative approach as a zellij plugin. I really don't mean to sound ungrateful and I wish I could better articulate why I feel the way I do about the UX. Maybe I'll come up with some more examples and try out how it feels.

No worries :) This approach was recommend mainly as it's how these kind of service composition is done in docker-compose.

If you talk about end-user UX, I almost always create some kind of aliases for these invocations, as there is no problem that one more layer of abstraction doesn't solve :)

See my ci-test invocation @ https://github.com/schemamap/schemamap#day-to-day-operations

[colemickens]

We're already approaching 3 duplicated process-compose files with very tiny adjustments between them, and then layering on more complexity with optionally disabled commands. This is quickly becoming untenable.

@F1bonacc1 can you give an indication if you'd be open to merging this feature if I take a weekend and write it? (Not committing to this, don't let me stop someone else, I'm very busy for the next couple of months.)

[adrian-gierakowski]

I’d suggest that helping to tame duplication in config files should be out of scope of process-compose. Instead of cramming more features into single tool, it’s better to delegate to other tools, more suitable for the job.

One such tool is https://github.com/Platonic-Systems/process-compose-flake which uses nix language for managing PC configurations. If you’re into nix then next step up would be: https://github.com/juspay/services-flake

If you don’t like nix you could try cue or jssonnet

[colemickens]

@adrian-gierakowski great suggestions, including at least one I know I haven't seen before. Indeed, we're a Nix shop, so I've been considering looking into devenv.sh which I think exactly uses nix to abstract/generate process-compose files.

Maybe throwing Nix at it is the better direction, given that there's other "stuff" we want to configure that could be driven from a singular Nix config too. Appreciate the suggestion, I'll definitely look into those more before adding features to p-c.

[thenonameguy]

An example of how devenv.sh-based processes look like:

https://github.com/schemamap/schemamap/blob/main/devenv.nix#L74-L90 Just have to set this config value: https://github.com/schemamap/schemamap/blob/main/devenv.nix#L23