Long-lived TCP connections die. Networking can fail in a million ways, even within data centers, and 'keep-alive' is a desperate plea at best. Even HTTP itself will force this: there's cases like a response body failing half-way through that are unrecoverable in HTTP/1.1 without killing the connection entirely.

串文

2025-03-16 01:54

讚

回覆

轉發

作者

Ballin Los
cpvrvpl

粉絲

253

串文

91+

讚

回覆

轉發

24小時粉絲增長

無資料

互動率

(讚 + 回覆 + 轉發) / 粉絲數

0.40%

回覆 (BETA)

最先回覆的內容
發文後	用戶	內容
幾秒內	Ballin Los cpvrvpl	Any spikes in traffic mean you'll end up with the wrong number of TCP connections one way or the other: either you need an enormous unused pool available at all times, or you'll need to open a flood of new connections as traffic spikes come in, and so deal with TCP slow start, RTT & extra TLS handshakes as you do so.