A place for developers to talk about building on Farcaster.

/dev

surprising 10x speed up in throughput with this change:

Before:
- await promise 1
- await p2
- await p3
- await p4

After:
- await [p1, p2]
- await [p3, p4]

Running node.js. Expected minor speed up. Underestimated cost of context switching.

building • prev manifold.xyz, twilio.com

Was that simple; diff below.

context: this is a standard http request handler, does about 1rps. The subroutines here are not expensive. I was expecting a minor speed up in proportion to request duration but median request went from ~1500ms to ~70ms.

Building ponder.sh. EVM, TypeScript, open-source enjoyer.

Was that simple; diff below.

context: this is a standard http request handler, does about 1rps. The subroutines here are not expensive. I was expecting a minor speed up in proportion to request duration but median request went from ~1500ms to ~70ms.

https://gist.github.com/jtgi/a9d9694186360df255481466fd19fac8

I'm not too knowledgeable with v8, but I suspect saving the event loop 2 interrupts nets more concurrency.

In bursts of traffic, requests used to stack during processing: 1000ms -> 1200ms -> 1400ms... increasing in latency, now they complete ~50ms, p99 @ ~100ms

ah, nah, you read it right. I fixed that in a follow up commit.

I might be misreading the diff, but looks like you removed the query for `alreadyProcessed`, such that it will now always be undefined? (And perhaps that query was expensive?)

Regardless, super curious to get to the bottom of it