Riverqueue

Fixed

- River CLI `go.sum` entries fixed for 0.0.13 release.

Added

- Added `riverdriver/riverdatabasesql` driver to enable River Go migrations through Go's built in `database/sql` package. [PR 98](https://github.com/riverqueue/river/pull/98).

Changed

- Errored jobs that have a very short duration before their next retry (<5 seconds) are set to `available` immediately instead of being made `scheduled` and having to wait for the scheduler to make a pass to make them workable. [PR 105](https://github.com/riverqueue/river/pull/105).
- `riverdriver` becomes its own submodule. It contains types that `riverdriver/riverdatabasesql` and `riverdriver/riverpgxv5` need to reference. [PR 98](https://github.com/riverqueue/river/pull/98).
- The `river/cmd/river` CLI has been made its own Go module. This is possible now that it uses the exported `river/rivermigrate` API, and will help with project maintainability. [PR 107](https://github.com/riverqueue/river/pull/107).

Added

- Added `river/rivermigrate` package to enable migrations from Go code as an alternative to using the CLI. PR 67.

Added

- `Stop` and `StopAndCancel` have been changed to respect the provided context argument. When that context is cancelled or times out, those methods will now immediately return with the context's error, even if the Client's shutdown has not yet completed. Apps may need to adjust their graceful shutdown logic to account for this. PR 79.

Changed

- `NewClient` no longer errors if it was provided a workers bundle with zero workers. Instead, that check's been moved to `Client.Start` instead. This allows adding workers to a bundle that'd like to reference a River client by letting `AddWorker` be invoked after a client reference is available from `NewClient`. PR 87.

Added

- Added `Example_scheduledJob`, demonstrating how to schedule a job to be run in the future.
- Added `Stopped` method to `Client` to make it easier to wait for graceful shutdown to complete.

Fixed

- Fixed a panic in the periodic job enqueuer caused by sometimes trying to reset a `time.Ticker` with a negative or zero duration. Fixed in PR 73.

Changed

- `DefaultClientRetryPolicy`: calculate the next attempt based on the current time instead of the time the prior attempt began.

Fixed

- **DATABASE MIGRATION**: Database schema v3 was introduced in v0.0.8 and contained an obvious flaw preventing it from running against existing tables. This migration was altered to execute the migration in multiple steps.