http server: fix regression set backlog to 1024

[niklasad1]

Closing #716

[jsdw]

LGTM!

Any idea what the negative effects of a higher backlog is? As a followup I wonder whether it's worth making this configurable?

[niklasad1]

Any idea what the negative effects of a higher backlog is?

I don't think there any downsides with a bigger backlog rather than that the kernel might use more memory and potential DOS for big values (also the OS might truncate this value).

For instance the max backlog size for Linux is 4096 and it is highly platform dependent (BSD uses one queue for both "ESTABLISHED" and "SYN waiting to be ACK:ed"; Linux uses two different queues for "ESTABLISHED" and "SYN waiting to be ACK:ed") so yes we should make this configurable for use but 1024 seems like more reasonable default value for instance tokio/mio and paritytech/jsonrpc uses 1024 already for the backlog. Perhaps even 4096 is reasonable here considering the BSD implementation.....

As a followup I wonder whether it's worth making this configurable?

Yes, let's take care of that in another PR.

[lexnv]

I wonder if this is capped by kern.ipc.somaxconn, currently the MacOS default is kern.ipc.somaxconn=128.