Enable pausing the workpool

[ianmacartney]

When an external provider is down, allow stopping processing manually and re-starting it later.

[ianmacartney]

In the keyed workpool world, you may want to pause just certain keys.

[danthegoodman1]

To add context: This is useful when you have workpools that are either singletons (e.g. parallelism = 1), or dedicated to a specific purpose (e.g. a specific agent).

In our use case, we have many of both of these cases. When there is a provider having downtime or increased errors, it is more useful to us to stop processing than to spin with lots of error logs. This is because we have our workpools retry effectively indefinitely, as they are infallible workflows.

In the short term, we could conditionally check whether there is a known error caught, and reschedule the function for some time in the future. However this would then violate any ordering that already exists (assuming that there is at least loose ordering with workpools).

Instead of building wrappers around all of our actions, it would be much easier if we could just (from an external control plane, like feature flags) just push out pauses to workpools by calling authenticated mutations that instruct the workpool to stop scheduling. Then, once the incident is over, we can push an update to raise max parallelism to above 0 for the respective workpools.

[danthegoodman1]

I had a look through the code and it seems like this change would actually be really easy. There's nothing that breaks this other than some checks:

getOrUpdateGlobals():

if (
      config.maxParallelism &&
      config.maxParallelism !== globals.maxParallelism
    ) {
      globals.maxParallelism = config.maxParallelism;
      updated = true;
    }

Would have to be updated to check for undefined instead of falsey

validateConfig():

} else if (config.maxParallelism < 1) {
    throw new Error("maxParallelism must be >= 1");
  }

just need to be changed to 0