| Quo-fan | i found a devuan build for odroid-c2 | 15:06 |
|---|---|---|
| Quo-fan | didnt even know it existed | 15:07 |
| nemo | have some bizarre behaviour on the VMs at work with devuan | 19:16 |
| nemo | /tmp is getting wiped clean, killing dbus/tmux sockets in /tmp which screws everything up | 19:17 |
| nemo | (have to send USR1 to tmux to recreate - for dbus haven't found anything better than restarting the vnc session) | 19:17 |
| nemo | besudes inotify monitoring of /tmp - anyone have any suspects for what might be causing this erasure? | 19:17 |
| nemo | *besides | 19:18 |
| nemo | oh... /run/user/1000/at-spi/bus_1 was also missing - but I'm not sure if that was due to same thing, or just accessibility losing communication with dbus in /tmp | 19:19 |
| nemo | also... why is dbus even creating these in /tmp instead of in /run ? that itself feels like a devuan specific issue perhaps | 19:21 |
| nemo | hm although it probably wouldn't help here because /run is a tmpfs which could be suffering from the same erasure | 19:22 |
| Quo-fan | nemo: its normal for dbus daemon | 19:22 |
| Quo-fan | dont worry | 19:22 |
| nemo | ok. do you have any idea what might be causing wipes of what appears to be *ALL* tmpfses on these VMs? | 19:23 |
| rwp | nemo, This feels to me like a buggy script or program which is deciding to remove everything in /tmp/*. | 19:25 |
| rwp | But what is the mode of the directory? ls -ld /tmp should say drwxrwxrwt root root there. | 19:25 |
| rwp | That 't' is the critical bit for /tmp. Has that been cleared? | 19:26 |
| nemo | it's still 't' | 19:26 |
| rwp | In that case it is a root process that is doing it. | 19:26 |
| nemo | odd thing is, I've spotted this behaviour across all of our devuan installs | 19:27 |
| nemo | I just never paid much attention to it before | 19:27 |
| nemo | hmmm | 19:27 |
| rwp | In a default installation /tmp is cleared at boot time. But that's boot time. | 19:27 |
| nemo | let me see if I had to enable that stupid system scripty thing they insisted on, on all these machines | 19:27 |
| nemo | one moment :) | 19:27 |
| nemo | hm. nope. it's not on the dev instances. | 19:28 |
| rwp | "tmpreaper" installed and misconfigured perhaps? | 19:29 |
| nemo | rwp: nope | 19:38 |
| rwp | Something is doing it. Clearing /tmp after boot time is definitely not normal and definitely bad. NOT a Devuan/Debian thing. | 19:47 |
| nemo | yeah. dunno. the odd thing is it's only happening to all the devuan servers. the ubuntu ones seem fine | 20:09 |
| nemo | regardless of whether vmtools are installed | 20:09 |
| rwp | How often is it happening? I might install an inotify something to log activity on a system. But I expect that to be a LOT of activity for /tmp. | 20:27 |
| FatPhil | /etc/default/rcS: TMPTIME | 20:33 |
| FatPhil | ah, is that only at boot? Dang, I only reboot every few years. | 20:34 |
| rwp | FatPhil, /etc/default/rcS TMPTIME also configures and controls "tmpreaper" too. | 20:38 |
| rwp | But at boot it is checkroot-bootclean.sh (uses TMPTIME) at boot time that cleans /tmp since boot time is the only completely safe time to do so. | 20:42 |
| FatPhil | I think tmpwatch was in the default install when I first installed a linux, and I just presumed debianalikes did the same by default too. | 21:54 |
| FatPhil | you learn something new every day. I'll add tmpreaper to my list of packages to install on all systems, as my reboot cycles are glacial. | 21:55 |
| rwp | "tmpwatch"? I recall that being available on the rpm side of the family but not the deb side. tmpreaper is the deb family side attempt. | 22:34 |
| rwp | Note that hardliners will object to any attempt to manipulate /tmp and point to user private tmp directories as the solution. | 22:35 |
| rwp | Whereas on my system I claim that /tmp is my own user private directory. :-) | 22:36 |
| FatPhil | yup, it would have been redhat in those days. Had at least 6 months on each of slack, rh, and suse before settling on debian. | 22:58 |
| rwp | Traditionally Debian had always been everyone's last system. Because then we realized we were home for a long stay. :-) | 22:59 |
| FatPhil | you can fudge an unreapabel user tmpdir by making a directory, and a directory inside that, and then making the outer one 600 | 23:00 |
| FatPhil | I stayed on suse on my DEC Alpha for a bit longer, as the graphics card support was better on minority architectures. | 23:01 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!