Mathew Heard

@keithchew could you share an actual patch? Thats far beyond just adding a line or two and requires an actual understanding of the control flow of the relatively undocumented KeyDB...

1. Heres the commit that introduces the issue https://github.com/Snapchat/KeyDB/commit/ffac55273a322d53c32372536cd3043d0b413989 2. I'm willing to be proven wrong (and I havent yet tested it) but in reviewing the code I think the...

BTW while I did have a go at a fixing this I felt the current release was too much of a bad idea and ended up doing a downgrade. We...

While I'm getting oppinions it would be extremely benificial to vhost-net and packet buffering if the lifetime of the `vpn_packet_t` (better name would be `net_packet_t` soon) could be taken out...

> If we can do it without reference counting, that would be even better. If we just make sure there's a single owner of each packet? That might be possible,...

Also perhaps instead of an offset we simply include a protocol field, the offset can be inferred from this (and there is value in knowing the contents of the packet).

@gsliepen Maskes sense. I think the tasks for this should involve. Agree? 1. Merge epoll & AES (and any other big commits that won't merge easily after) 2. Small PR...

The way nginx does it, potentially ignoring Content-Length seems a bit strange. However most request / responses do follow the non infinite upload property... I think a timeout would be...

It will block the same amount as http/1.1.That should be acceptable and consistent in behaviour.

BTW I've reverted this for two different client applications and continue to experience no real problems. We don't have the concern of spiked requests though. The problems really are limited...