Passing a riverClient to a worker?

[chrboe]

I am trying to trigger a job from within another job (think: send a notification when a long-running job is complete). This requires that I have access to the riverClient in the worker context.

I could pass the client to the worker, like

river.AddWorker[action.SomeArgs](workers, &action.SomeWorker{DbPool: dbPool, RiverClient: riverClient})

but that creates a chicken-egg problem because I need an initialized Workers to create the riverClient.

Adding the worker after initialization does not work either because NewClient does not accept an empty Workers:

workers := river.NewWorkers()

riverClient, err := river.NewClient[pgx.Tx](riverpgxv5.New(dbPool.Pool), &river.Config{
    Workers: workers,
})
if err != nil {
    panic(err)
}

river.AddWorker[action.SomeArgs](workers, &action.SomeWorker{DbPool: dbPool, RiverClient: riverClient})

// panic: at least one Worker must be added to the Workers bundle

Adding some worker that does not depend on the client does actually work, but is kind of awkward:

workers := river.NewWorkers()
river.AddWorker[action.DummyArgs](workers, &action.DummyWorker{})

riverClient, err := river.NewClient[pgx.Tx](riverpgxv5.New(dbPool.Pool), &river.Config{
    Workers: workers,
})
if err != nil {
    panic(err)
}

river.AddWorker[action.SomeArgs](workers, &action.SomeWorker{DbPool: dbPool, RiverClient: riverClient})

// -> both workers work as expected

Is there any other "pattern" to achieve this? If not, I think the status quo should at least be documented somewhere.

[brandur]

Hey @chrboe, thanks for the detailed report.

Hm, this indeed seem like a problem. I talked it over with Blake and we think that there's not really that much good rationale behind ensuring that there's a non-zero worker bundle in NewClient — we in fact add workers all the time after NewClient in our test suite even.

I opened #88 proposing that we move the zero worker check from NewClient to Start instead, meaning that it should be easy for you to create a client, then add workers that depend on the client, then after doing to start the client.

@bgentry also proposed we provide an easy way to make a client available via a context key passed to each work function. We'll put that on the backburner for now, but it could be a nice improvement if this becomes an obviously common pattern that people are using.

[brandur]

We just cut a new release for 0.0.11 that includes the changes in #88.