Makes sense. It would not be difficult to add this to Prism to enable local iteration and non-Dataflow smoke test validation for the open source side.
In an ideal case we'd have a benchmark that shows an improvement with the feature enabled due to avoiding the head of line blocking. Robert On Wed, Jun 17, 2026, 6:53 AM Danny McCormick via dev <[email protected]> wrote: > +1 - I left a question, but this LGTM. > > Thanks, > Danny > > On Wed, Jun 17, 2026 at 9:46 AM Kenneth Knowles <[email protected]> wrote: > >> Thanks for writing this up. This makes a ton of sense. I'm honestly >> surprised we've gotten this far without running into the issue. I guess it >> has to do with the particulars of how Java and Python use (or don't use) >> the multiplexing of the data plane connection. >> >> Kenn >> >> On Wed, Jun 17, 2026 at 7:47 AM Sam Whittle <[email protected]> wrote: >> >>> Hi everyone, >>> I put together a design and a PR to address the lack of bundle isolation >>> within an SDK given the single shared data stream. This lack of isolation >>> can lead to starvation of input data for bundles if there are other bundles >>> within the SDK that are not consuming data quickly. >>> >>> Please let me know what you think or any concerns >>> >>> https://docs.google.com/document/d/1nahf5UcWMEpuN7g8ezqgE-FApirrPOmhHuZzMp9Peb0 >>> >>> Thanks, >>> Sam >>> >>
