v0.12.0 all tasks go to the status discarded with InsertTx

[sxwebdev]

After upgrading to version 0.12.0 all my new tasks created in the following way:

riverClient.InsertTx(ctx, dbTx,
	taskmanager.SomeArgs{
		Key: value,
	},
	&river.InsertOpts{
		UniqueOpts: river.UniqueOpts{
			ByArgs: true,
		},
		ScheduledAt: time.Now().Add(time.Second*30),
	},
)

Began to be marked in the database with the status "discarded"

This problem was solved only in this way:

func (s *Worker) InsertOpts() river.InsertOpts {
	return river.InsertOpts{
		UniqueOpts: river.UniqueOpts{
			ByArgs: true,
		},
	}
}

It looks like unique parameters are ignored at the task creation level.

Before updating to version 0.12.0, I rolled out migrations from scratch

[bgentry]

I’m not clear on what’s happening or what actually fixed it. You’re saying that you were using insert time UniqueOpts on a prior version with only ByArgs with no problems? And after upgrading, your jobs are somehow being inserted directly into a discarded stste?

I don’t think the discarded state can ever be reached at insert time, or really ever, without the job executing first and having errors, which is why this confuses me. However I don’t doubt that something broke here and It’s likely pretty important. Are there any other logs you can provide, or a minimal code sample to reproduce? Maybe look at the job rows in your database to see if anything looks off? For a discarded job you should see at least some evidence of attempts and having been worked by a worker.

[sxwebdev]

I’m not clear on what’s happening or what actually fixed it. You’re saying that you were using insert time UniqueOpts on a prior version with only ByArgs with no problems? And after upgrading, your jobs are somehow being inserted directly into a discarded stste?

I don’t think the discarded state can ever be reached at insert time, or really ever, without the job executing first and having errors, which is why this confuses me. However I don’t doubt that something broke here and It’s likely pretty important. Are there any other logs you can provide, or a minimal code sample to reproduce? Maybe look at the job rows in your database to see if anything looks off? For a discarded job you should see at least some evidence of attempts and having been worked by a worker.

I remember that there were no attempts, the task was never completed, it immediately crashed with the “discarded” status. At the same time, an error with “discarded” was recorded in the meta. Also, the new column with the unique state was also filled with a bit mask. In other words, the task crashed with an error on uniqueness. I will be able to reproduce this only tomorrow. And I did not check whether the uniqueness works now or not, when I specified the method in the worker.

at the same time, all the arguments are absolutely unique

[bgentry]

@sxwebdev I've found at least one bug here that's new to v0.12.0. While I've fixed that one locally, I'm working on additional test cases and making sure all edge cases are covered. Hope to have something up shortly.

[sxwebdev]

@bgentry thank you, waiting for the release

[bgentry]

@sxwebdev I believe #619 should address the issue you ran into along with another one I found. Added missing coverage to ensure it's fixed and doesn't break again. I'll ship that shortly in a v0.12.1.

Really sorry about the issue here, thank you so much for reporting it.

[bgentry]

@sxwebdev ok v0.12.1 should be live now (barring any of the annoying issues with go module proxy caching), can you give that a try and let me know if you see anything else fishy? Thank you again 🙏

[sxwebdev]

@bgentry Hi. I just rolled back my changes.

Removed this:

func (s *Worker) InsertOpts() river.InsertOpts {
	return river.InsertOpts{
		UniqueOpts: river.UniqueOpts{
			ByArgs: true,
		},
	}
}

Returned this:

riverClient.InsertTx(ctx, dbTx,
	taskmanager.SomeArgs{
		Key: value,
	},
	&river.InsertOpts{
		UniqueOpts: river.UniqueOpts{
			ByArgs: true,
		},
		ScheduledAt: time.Now().Add(time.Second*30),
	},
)

And updated the river to 0.12.1

In other words, I did it as it was before.

Now everything works as before version 0.12. The problem is no longer observed, thanks for the promptness.