Simon Frei

Ok, then probably don't delay for my stuff, because merging till next Tuesday is hardly realistic. I probably shouldn't start describing it too much, that would just cause confusion. The...

I updated the links (and wrote to github support, something is broken here). Anyway, the takeaway for me is that a disk-overflow type is probably good to have mid to...

I indeed forgot to `git add` a crucial file - symptom of hardly writing any original code at all :) I cleaned it up and filed a (wip) PR, referencing...

Can anyone that experiences this on a 64bit system please grab heap profile during hihg memory usage (https://docs.syncthing.net/users/profiling.html). And ideally also after it went back to normal for a baseline.

@scienmind Most memory usage is due to requests (other devices pulling the data from that device). By default every device can request up to 64MB of data. The largest profile...

I completely forgot about the comments in this issue. I am fairly sure it is completely unrelated to the original issue in this thread, but a problem with our pooling...

It's a buffer pool for packets. So might be valid usage due to a lot of churn (seems overly excessive though), or quic-go failing to hand buffers back to the...

Accompanied by a 30MB increase in that particalur pool in the profile. Can you als grap goroutine traces please, to check if there's an unexpected amount of open streams or...

There's a ton of message writing routines. And a quic conn to be closed. Maybe the are contending or even dead-locking on something. @leipert Could you grap goroutines again with...

They all try to write to the same `*netFD` protected by a lock which is held by one routine in iowait/netpoll since 1289min. No idea why it's blocked, but regardless...