det
orignal: My proposal doesn't decline NTCP2 connections, it declines all connections when no more connections can be accepted. I did not add any code to account for when timerfds (which are a terrible idea) suddenly start failing en-masse.
det
I don't know how to write code that fixes that issue with timers, sorry. Just trying to stop the log from filling my hard disk with "Accept failed" lines.
Xeha
det: increase the open files ulimit?
orignal
or reduce transit tunnels limit
det
Xeha: I mean I don't know how to write code that fixes the fact that i2pd shouldn't be using timerfds at all, and event loops should pause to execute timers with no file descriptors needed.
orignal
I don't know either
orignal
it's old problem
det
it's boost...
orignal
that doesn't have a good solution
orignal
however SSU2 will use one timer for all session
det
I do know how to deal with timerfds wasting 2 descriptors for no reason for every session. Increasing the open files ulimit, and such.
orignal
and it should solve the problem
orignal
SSU code is just old and bad
orignal
no worth to rewrtie it
orignal
for NTCP2 there is only on decrriptor per session
det
orignal: That sounds like a good solution, to be honest. That's what I don't know how to do, is write SSU2.
orignal
and you can't avoid it because TCP/IP
orignal
we are actively develpp SSU2 now
orignal
will be in prod in few month
det
I only see a branch called "openssl" but it does have some code to talk with routers speaking a different SSU protocol.
orignal
openssl is the main branch
orignal
and SSU2 code is there
orignal
look at SSU2.h and SSU2.cpp
det
I'll take a look sometime, thanks.
orignal
but it's very begiining
orignal
only hanhshake basically
orignal
no polling and resend for multiple sessions, etc.
orignal
will be the next step
orignal
if you are insterested we discuss SSU2 at #ls2
orignal
and have a meeting today