feat(pubsub): pause the publisher for an ordering key when there is an error #4286

PhongChuong · 2026-01-15T15:06:01Z

Pause the Publisher when we encounter an error when there is a send error.

When an error is encountered for a pending batch, we:

In BatchWorker, pause publishing and send out errors for pending_msgs.
In the pending batch, send out error for its messages.
New messages in the rx receiver are handled as they are received by the BatchWorker.

A resume operation will be added in a later PR.

This PR also introduce PublishError. Further work is needed to handle error propagation more fully

codecov · 2026-01-15T15:18:41Z

Codecov Report

❌ Patch coverage is 88.23529% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 94.82%. Comparing base (13bef44) to head (7030904).
⚠️ Report is 8 commits behind head on main.

Files with missing lines	Patch %	Lines
src/pubsub/src/publisher/worker.rs	66.66%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4286      +/-   ##
==========================================
- Coverage   94.85%   94.82%   -0.04%     
==========================================
  Files         187      187              
  Lines        7194     7207      +13     
==========================================
+ Hits         6824     6834      +10     
- Misses        370      373       +3

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

dbolduc

I didn't get to the changes in worker.rs

Consider doing the error type refactor first, then the unit tests look more like what we want.

src/pubsub/src/error.rs

dbolduc · 2026-01-15T17:09:24Z

src/pubsub/src/error.rs

+
+    /// Publish is paused for the ordering key.
+    #[error("the ordering key was paused")]
+    OrderingKeyPaused(()),


Usability question, which is not trivial, so feel free to think about it later:

Publish consumes the application's PubsubMessage. If the operation fails, what should the application do to resend the message? Would they need to hold onto a clone of the message until the operation is complete?

It would be super convenient if we could give them their message back. The plumbing on our end could be brutal, but the application would appreciate it. 🤷

our GAPICs all have this problem too.... idk if publishing is special. Something seems wrong about just dropping their message without trying to send it. But maybe I am worrying too much about the unhappy case. 🤷

Agreed.
Ideally, in the error case we should give the user the message back. IMO, we can do this in 2 ways:

Keep a clone of it internally and pass it back to the user if there is a failure.

Augment generated code/error to pass the message back if there is an error during Send.
I think it's a bigger discussion that should be left out of this PR.

src/pubsub/src/lib.rs

dbolduc · 2026-01-15T17:16:11Z

src/pubsub/src/publisher/model_ext.rs

    }
 }
+
+fn convert_error(e: crate::error::PublishError) -> gax::error::Error {


comment: ah, I think you are hesitating to change Output to a Result<String, PublishError> because then we have to update all the code downstream of this.

Ok.

This is really yucky though. It might have been nicer to change the error type first.

I think we should discuss as a group with @suzmue if returning gax::error::Error or Publish error directly is ideal. We decided to with gax::error::Error for now as it is consistent with the other clients.

It feels so wrong to me, but we do have a precedence for just throwing something in an Error::io

google-cloud-rust/src/storage/src/storage/bidi/worker.rs

Lines 148 to 149 in d51e1ca

// TODO(#3626) - reconsider the error kind.

result.map_err(crate::Error::io)

Although in GCS, I think there was a more compelling reason. (We wanted to reuse the ReadObjectResponse type which was in terms of gax::Error.)

src/pubsub/src/publisher/model_ext.rs

src/pubsub/src/publisher/publisher.rs

PhongChuong

Thanks for the review.

PhongChuong · 2026-01-15T18:38:01Z

src/pubsub/src/error.rs

+
+    /// Publish is paused for the ordering key.
+    #[error("the ordering key was paused")]
+    OrderingKeyPaused(()),


Agreed.
Ideally, in the error case we should give the user the message back. IMO, we can do this in 2 ways:

Keep a clone of it internally and pass it back to the user if there is a failure.

Augment generated code/error to pass the message back if there is an error during Send.
I think it's a bigger discussion that should be left out of this PR.

src/pubsub/src/lib.rs

PhongChuong · 2026-01-15T18:44:01Z

src/pubsub/src/publisher/model_ext.rs

    }
 }
+
+fn convert_error(e: crate::error::PublishError) -> gax::error::Error {


I think we should discuss as a group with @suzmue if returning gax::error::Error or Publish error directly is ideal. We decided to with gax::error::Error for now as it is consistent with the other clients.

PhongChuong added 2 commits January 14, 2026 20:27

publish error

19fdeb1

add tests

7725fd7

product-auto-label bot added the api: pubsub Issues related to the Pub/Sub API. label Jan 15, 2026

fix format

830766c

PhongChuong marked this pull request as ready for review January 15, 2026 16:35

PhongChuong requested a review from a team as a code owner January 15, 2026 16:35

PhongChuong requested a review from suzmue January 15, 2026 16:35

dbolduc reviewed Jan 15, 2026

View reviewed changes

PhongChuong added 2 commits January 15, 2026 18:27

better test validation

5f334c1

address comments

7030904

PhongChuong commented Jan 15, 2026

View reviewed changes

	// TODO(#3626) - reconsider the error kind.
	result.map_err(crate::Error::io)

feat(pubsub): pause the publisher for an ordering key when there is an error #4286

Are you sure you want to change the base?

feat(pubsub): pause the publisher for an ordering key when there is an error #4286

Uh oh!

Conversation

PhongChuong commented Jan 15, 2026

Uh oh!

codecov bot commented Jan 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

dbolduc left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dbolduc Jan 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

PhongChuong left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

codecov bot commented Jan 15, 2026 •

edited

Loading

dbolduc Jan 15, 2026 •

edited

Loading