Skip to content

pekko: avoid backpressure for staggered broadcast#1873

Merged
brharrington merged 3 commits intoNetflix:mainfrom
brharrington:stagger
Feb 24, 2026
Merged

pekko: avoid backpressure for staggered broadcast#1873
brharrington merged 3 commits intoNetflix:mainfrom
brharrington:stagger

Conversation

@brharrington
Copy link
Copy Markdown
Contributor

Allow the set of cluster members to be updated even if the downstream is still working based on the injected delays.

Allow the set of cluster members to be updated even if the
downstream is still working based on the injected delays.
@brharrington brharrington added this to the 1.9.0 milestone Feb 24, 2026
@brharrington brharrington merged commit 18bf58f into Netflix:main Feb 24, 2026
5 checks passed
@brharrington brharrington deleted the stagger branch February 24, 2026 04:42
brharrington added a commit to brharrington/atlas that referenced this pull request Feb 24, 2026
Allow the set of cluster members to be updated even if the
downstream is still working based on the injected delays.
brharrington added a commit that referenced this pull request Feb 24, 2026
* pekko: add warn log if byte limit is hit (#1870)

Ensures it will get logged and be easier to debug. If relying
on the caller it might get burried in the pekko plumbing.

* lwcapi: heartbeat immediately on ws connect (#1871)

When subscription delivery is delayed (e.g., due to staggered
broadcast), the websocket connection could hit the idle timeout
before any data flows. This was because the data source
(including heartbeats) was not pushed downstream until the
client sent its first subscription message.

- Push the data source on the first downstream pull instead of
  waiting for a client message, so heartbeats flow as soon as
  the websocket connects.

- Emit a default heartbeat with a 5s step when no subscriptions
  are established yet. Once subscriptions arrive, the real
  step-based heartbeats take over on the next tick.

- Add debug logging for staggered broadcast delay.

* pekko: remove redundant unit in log message (#1872)

* pekko: avoid backpressure for staggered broadcast (#1873)

Allow the set of cluster members to be updated even if the
downstream is still working based on the injected delays.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant