Unable to determine which orderingKeys need resumePublishing

[mgabeler-lee-6rs]

Environment details

• OS: Linux (but N/A)
• Node.js version: 14.x (but N/A)
• npm version: N/A
• @google-cloud/pubsub version: v2.19.0

Steps to reproduce

1. Induce or wait for errors on a publisher when using orderingKeys
2. Try to figure out which ordering keys need a call to resumePublishing
3. Discover that you cannot, at least not without reaching fingers into private members & implementation details

Details

The resumePublishing API is difficult and frustrating to use:

• There is no way to find which ordering keys need the resume, if you want to do a "full reset"
• Once a queue has an error, it doesn't emit a drain event, and so it stays in memory forever, but with no way to find out it is there and clear the error

Related Issues

I discovered this when investigating why my applications are affected by issues very similar to #1425 and #1462, and strange behaviors observed on retries.

[feywind]

I could've sworn we had a discussion about this one too... sorry to keep pinging you, but @kamalaboulhosn ?

[kamalaboulhosn]

The promise returned from a publish call would have to be tracked along with its ordering key. For any publish with an ordering key for which the message failed, it would be necessary to call resumePublish.

[mgabeler-lee-6rs]

That's still prone to race conditions. You also need to know that there isn't any code in the process of enqueuing multiple sequential publish calls. If you resumePublish in the middle of that, it will cause out-of-order publishes to happen. I'm not sure there's anything this library can do about that, just calling out that proper implementation of an ordered publisher is ... surprisingly complicated.

[kamalaboulhosn]

Yes, there are definitely some subtleties. We do have a prober written in Java that endeavors to provide an example of how to do it well: https://github.com/GoogleCloudPlatform/pubsub/tree/master/ordering-keys-prober.

[kamalaboulhosn]

Since this is not specific to Node, I have created an issue in the public issue tracker to keep track of it: https://issuetracker.google.com/issues/288118039. However, at this time, there is no timeline for if and when we may provide this functionality.