On poll_flush timing of RPC client and server

[0x5459]

Hi,

Big fan of this repo.

I found that each time a write request to transport in RPC client, the poll_flush method will be called by the pump_write of RequestDispatch. This approach may not be cache friendly. I wrote a quick demo using start_send in a loop to mimic multiple requests and then pull_flush all these requests in one try.

In my demo, pump_write no longer writes one request at a time but it's writing requests all the time. Same with pump_read. If you think this is the right approach, I'm more than happy to submit a PR for this.

[tikue]

Hi! Thanks for the kind words 😊

Thank you for pointing this out! I think it's a great suggestion to flush less frequently. Please send a PR!

[tikue]

Looking more closely, I think this is the problem line: https://github.com/0x5459/ext-proc-channel/blob/85a8f275d0e0ff3e517d1a96bef5ab3ac27a89e4/src/client.rs#L269 because it doesn't check if the transport is ready for a request to be buffered.

start_send returns an Err when the write buffer is full, which means we need to flush, right? Otherwise the channel could spuriously fail.

The ensure_writeable fn doesn't always flush; it tries to only flush when the transport's buffer is full.

[0x5459]

Looking more closely, I think this is the problem line: https://github.com/0x5459/ext-proc-channel/blob/85a8f275d0e0ff3e517d1a96bef5ab3ac27a89e4/src/client.rs#L269 because it doesn't check if the transport is ready for a request to be buffered.

start_send returns an Err when the write buffer is full, which means we need to flush, right? Otherwise the channel could spuriously fail.

The ensure_writeable fn doesn't always flush; it tries to only flush when the transport's buffer is full.

I'm sorry. You're right. Great code!