Open Stoatwblr opened 3 years ago
whatever the issue is, it's not there in v2.3.2 clients, but manifests in ones reporting 2.3.3 and later to the emule client
verified this is manifesting with ubuntu-distro (21.04) amule 2.3.3 clients too.
We're going to start seeing more complaints shortly as people upgrade :/
I know you're a pro (usually). As the code didn't change (as far as I can see). Can you see anything in the surroundings that changed? Maybe something new in Boost or wx code. I've seen boost warnings about global indexing or something like that while compiling.
I'm at a loss on this one, at the moment, honestly. I suspect that whatever's causing this was introduced several years ago and is subtle as heck
It's buffering. I had it set large (1.5MB) and every time it flushed, things would hang.
quite WHY this is happening on a 24 core ZFS system with SLOG(ssd write caching - kind of) is a mystery and the OS itself isn't missing a beat
The answer (on linux) is set it as small as possible and let the OS take care of write buffering
I've filed this as an emule bug. It shouldn't fall over its shoelaces if the receiver pauses to flush. Let's see if Fox88 does anything with that
On the amule side I suggest defaulting to minimum buffering on linux and adding a stronger warning against changing it
it's still occurring even with amule buffers set as small as I can make them. The Temp area is on a dedicated f2fs ssd which is more than fast enough for the task (samsung 840pro, and iostat not showing much activity when this happens)
Various emule versions definitely have trouble talking to amule 2.4 even when I'm not sharing insane numbers of files and I've eliminated NAT as a potential contributor by obtaining dedicated real-world IPs for testing purposes
On Manjaro (Arch Based) this happens even with version 2.3.2.
Any news on this? Is this project still alive? Stopping and continuing makes the download restart at good speed for some seconds, after which everything slows down again...
Can you check out my build from https://github.com/amule-project/amule/issues/290#issuecomment-2171901412 and test if that fixes this for you?
I've been testing this for the last few months to verify and I'm pretty sure that it's an amule issue of some kind
Whatever the problem is, manifests on the sending (emule) end as "stalled" file sends after 1.59MB, and then the transfer times out
Amule sending to emule works fine
I don't know enough to dig deeper into what's going on but if you run a couple of instances of amule/emule side by side and look for the same files it's clear they have different views of the ed2k/kad universe