andygrove opened a new pull request, #2798: URL: https://github.com/apache/datafusion-ballista/pull/2798
# Which issue does this PR close? No single issue. Split out of #2792. Related: #2520 (performance review tracking). # Rationale for this change The scheduler serves gRPC through axum, which, unlike tonic's own server, leaves Nagle's algorithm on for accepted connections. A poll response is several small HTTP/2 frames, so the later ones wait for the executor's delayed ACK, about 40 ms per poll on Linux. Under pull scheduling that is added to every task launch, so a stage of N waves pays N x 40 ms. TPC-H q2 at 128 partitions spent about 2 s of its 5.6 s waiting. At 128 partitions on one machine (TPC-H SF100), all of #2792 takes 148 s against 204 s on `main`, mostly from this change. # What changes are included in this PR? `TCP_NODELAY` is set on every connection the scheduler accepts, via `ListenerExt::tap_io`. A failure to set it logs a warning and the connection proceeds. No test: the effect is a socket option on accepted connections. # Are there any user-facing changes? No config or API changes. Lower task-launch latency under pull scheduling. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
