You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, many modules don't have backpressure implemented, because the relationship between a message coming from the sink, and the n messages produced by that message, going out in the stream, is not well defined.
This also conflicts with the granularity of messages we expect from a client, as too broad, and it could produce a large n, too granular, and we can get it down to the ideal 1:1 ration of sink message to stream message, but that makes it difficult for clients and causes a lot of needless communication.
Instead, we should standardize that perhaps all of these modules keeps a list of messages to go out in the stream, and if that list is above some threshold k, then any messages coming on from the sink, which could produce messages in the stream (or maybe just all messages coming in regardless), will respond with a NotReady if n is currently larger than k to allow clients to drain the stream until n is less than k.
This way, its not really well defined how large n could be, but it makes it easy for clients and implementors, while also injecting some soft backpressure into the system.
UtMetadataModule
HonestRevealModule
StandardSelectModule
The text was updated successfully, but these errors were encountered:
Currently, many modules don't have backpressure implemented, because the relationship between a message coming from the sink, and the
n
messages produced by that message, going out in the stream, is not well defined.This also conflicts with the granularity of messages we expect from a client, as too broad, and it could produce a large
n
, too granular, and we can get it down to the ideal 1:1 ration of sink message to stream message, but that makes it difficult for clients and causes a lot of needless communication.Instead, we should standardize that perhaps all of these modules keeps a list of messages to go out in the stream, and if that list is above some threshold
k
, then any messages coming on from the sink, which could produce messages in the stream (or maybe just all messages coming in regardless), will respond with aNotReady
ifn
is currently larger thank
to allow clients to drain the stream untiln
is less thank
.This way, its not really well defined how large
n
could be, but it makes it easy for clients and implementors, while also injecting some soft backpressure into the system.The text was updated successfully, but these errors were encountered: