@Xeha
@orignal
Arch
Danny
Irc2PGuest12735
Irc2PGuest28644
Irc2PGuest4450
Irc2PGuest80483
Irc2PGuest87589
Meow
R4SAS
RN
RN_
SigSegv
acetone
aeiou
ardu
b3t4f4c3__
eyedeekay
hagen
hk
mareki2p
onon_
poriori_
qend-irc2p
r00tobo
radakayot
rapidash
semantica
shiver_
thetia
u5657
weko_
woodwose
x74a6
det
R4SAS: ok here's a simple patch: y6d4fs3rpqrctuv77ltfajf5m4tl4kzcu7rtwhxgiohylfxxow4q.b32.i2p/i2pd-exhausting-descriptors.patch
det
I still say the bundle works better though y6d4fs3rpqrctuv77ltfajf5m4tl4kzcu7rtwhxgiohylfxxow4q.b32.i2p/i2pd-exhausting-descriptors.bundle
R4SAS
orignal: ping
R4SAS
> number of descriptor also depends on number of timer. That means SSU and Streaming
det
NTCP2 was what was taking the most descriptors. I don't really know about SSU I thought it was connectionless?
R4SAS
SSU has timers
R4SAS
and streams
R4SAS
so every timer is file descriptor
det
Ooh okay.
det
Wait, a file descriptor for every timer? Really?
det1
I'd just... keep a linked list of timing events, in the... event loop.
det1
I don't know how boost does it.
orignal
what?
orignal
det Linux uses timerfd for each timer
orignal
unfortunately SSU uses timer for each session
orignal
going to be changed in SSU2
orignal
your proposal is on the right tracks
orignal
but if you decline NTCP2 connection it will use more SSU session that wouldn't solve a problem
orignal
we know for sure that boost::deadline_time uses timerfd on linux