Quoth gnufan42 via 9fans <[email protected]>: > I'm having a weird networking issue on 9front. Like the subject says, my > Drawterm session will randomly freeze after some use, and after which the > entire machine will lose network access - ip/ping will stop working, and hget > will fail. Every time this happens I have to restart the machine. And here > are things I've tried to try to find the cause: > > First, I left the machine on for a whole day. Nothing happens, I can still > connect to it. > Next, I think this might happen if the CPU is under heavy load, so I ran Doom > on both the Drawterm and the Plan 9 console. Still nothing happens. After > several hours Doom's demo is still playing inside Drawterm. > Then, I thought this might have to do with cwfs64x being unstable, so I tried > to hammer the file server with this script on my Linux machine: > > #!/bin/bash > tfile="$(mktemp -d /tmp/test.XXXXXXXXX)" || exit 1 > # 9pfs from https://github.com/ftrvxmtrx/9pfs > 9pfs -p "$2" -u "$3" "$1" "$tfile" > if [ "$(stat -f -c %T $tfile)" != fuseblk ]; then > rmdir "$tfile" > exit 1 > fi > max=20 > for ((i = 0 ; i < max ; i++ )); do > echo "$i"/"$max"\: > tar -c "$tfile" | dd status=progress of=/dev/null > done > umount -l "$tfile" > rmdir "$tfile" > > And it cannot even finish one round without freezing. But when I tried to > restart the machine and run the script again, out of 4 times, every single > time tar freezes (and the network stops working) after reading exactly > 97013760 bytes like this: > > 0/20: > tar: Removing leading `/' from member names > 2232832 bytes (2.2 MB, 2.1 MiB) copied, 2 s, 1.1 MB/star: > /tmp/test.KhkGpOEiH/adm/keys: Cannot open: Permission denied > tar: /tmp/test.KhkGpOEiH/adm/keys.who: Cannot open: Permission denied > 95805952 bytes (96 MB, 91 MiB) copied, 65 s, 1.5 MB/s^C > 189480+0 records in > 189480+0 records out > 97013760 bytes (97 MB, 93 MiB) copied, 80.3674 s, 1.2 MB/s > > However, just as me about to blame cwfs64x, I decided to try something else. > I restarted the machine, and this time I run cat /dev/zero > > /mnt/term/dev/null in Drawterm. Okay, nothing happens, the connection is > still good after a few hours. But by chance I scrolled in a rio window and > the whole thing immediately freezes. I also tried to replicate that, and yes, > if I run cat /dev/random and rapidly scroll the window, it also triggers > freezing. > > I have no idea, really. I cannot pin down this being caused by cwfs64x. I > also don't understand if it's truly cwfs64x's problem, how can it affect > networking at all - it doesn't crash after me losing the Drawterm session. I > can still run commands inside the Plan 9 Console, and all files are intact. > Can somebody suggest me what to look at next? >
And there are no messages on the console? no stall error, or anything else? when you walk up and type on the console, is the machine hung? ------------------------------------------ 9fans: 9fans Permalink: https://9fans.topicbox.com/groups/9fans/T006c8f1c3d70cd80-M27ecf0db0d361c8e1ee8ac35 Delivery options: https://9fans.topicbox.com/groups/9fans/subscription
