grpc-web memory leak(?) on long-running streams

grpc-web stores the entire contents of a stream in memory, which bogs down and causes performance issues over time as well as using increasing amounts of memory. I have been able to work around by cancelling and restarting requests after they exceed a certain size, but was wondering if this is a memory leak or working as intended.

If this is how it's supposed to work, I was wondering what the best practices around managing long-running streams are supposed to be. I haven't found any documentation on the subject.

Thank you!

No files, tests, or entry points are named in the issue. Start by reproducing a long-running grpc-web stream and measuring whether received contents remain in memory; compare the behavior with the documented streaming expectations. Done means either the retention issue is fixed and verified, or the supported approach for managing long-running streams is clearly documented.

Written by the indexing model from the issue text.