View previous topic :: View next topic |
Author |
Message |
DawgG l33t
Joined: 17 Sep 2003 Posts: 866
|
Posted: Fri Jun 22, 2018 12:29 pm Post subject: squid does not respect cache_dir size but uses all diskspace |
|
|
i run squid-3.5.27 on an up-to-date amd64-server, its cache-dirs (and logs) are placed in their own ext2-partition with a size of 100GB. i want the cache to use 75GB of this so the directive in squid.conf is
Code: | cache_dir aufs /mnt/cache/squid 75000 16 256 |
recently the webcache became inaccessible because the whole partition was 100% full (space, not inodes) and when i checked i noticed that squid frequently occupies more than the allocated 75GB; eg just now
Code: | squid2 ~ # df -h
/dev/sdb2 99G 82G 13G 87% /mnt/cache |
this webcache is used mainly for windoze-updates and has some unconventional settings because of this (see https://wiki.squid-cache.org/ConfigExamples/Caching/WindowsUpdates), especially
Code: | range_offset_limit 6000 MB
maximum_object_size 6000 MB
quick_abort_min -1
quick_abort_max -1
quick_abort_pct -1 |
but i think this should not let the whole cache become inaccessible or override the cache_dir-setting above.
what do you think? _________________ DUMM KLICKT GUT. |
|
Back to top |
|
|
Irom Tux's lil' helper
Joined: 07 Oct 2003 Posts: 95 Location: am arsch..
|
Posted: Mon Jun 25, 2018 8:10 pm Post subject: |
|
|
It's very unlikely because the size difference is immense, but here's at least something to try: Maybe the space is wasted by the minimum block size of your fs?
How much GB does
Code: | find /mnt/cache/squid -printf '%k\n' | awk '{ sum += $1 } END { printf("%.2f GB\n", sum / 1024 / 1024 ) }' |
show? _________________ http://ftp.fukt.bsnet.se/pub/movies/stallman/ (Please watch this before you form an opinion about GNU)
https://apfelboymchen.net/gnu/ |
|
Back to top |
|
|
DawgG l33t
Joined: 17 Sep 2003 Posts: 866
|
Posted: Fri Jun 29, 2018 9:47 am Post subject: |
|
|
THX for your reply!
the difference is minimal:
Code: | find /mnt/cache/squid -printf '%k\n' | awk '{ sum += $1 } END { printf("%.2f GB\n", sum / 1024 / 1024 ) }'
93.04 GB | vs.
Code: | df /mnt/cache/
/dev/sdb2 103212320 97898508 70932 100% /mnt/cache
df -h /mnt/cache/
/dev/sdb2 99G 94G 0 100% /mnt/cache |
so the service went down again (from cache.log):
Code: | 2018/06/29 11:34:03 kid6| diskHandleWrite: FD 18: disk write error: (28) No space left on device
FATAL: Write failure -- check your disk space and cache.log
Squid Cache (Version 3.5.27): Terminated abnormally.
CPU Usage: 0.018 seconds = 0.014 user + 0.005 sys
Maximum Resident Size: 70608 KB
Page faults with physical i/o: 9 |
i guess the cache_swap-settings somehow conflict with the large object_size and the non-terminating transfers of the f***ing windoze-updates. i'll look deeper into that. _________________ DUMM KLICKT GUT. |
|
Back to top |
|
|
Irom Tux's lil' helper
Joined: 07 Oct 2003 Posts: 95 Location: am arsch..
|
Posted: Fri Jun 29, 2018 10:08 am Post subject: |
|
|
I just saw the following as a comment in my own squid.conf and think it could explain your problem.
http://www.squid-cache.org/Doc/config/cache_dir/ wrote: | Do NOT put the size of your disk drive here. Instead, if you want Squid to use the entire disk drive, subtract 20% and use that value.
|
75 GB + 20 GB fits your 95 GB. _________________ http://ftp.fukt.bsnet.se/pub/movies/stallman/ (Please watch this before you form an opinion about GNU)
https://apfelboymchen.net/gnu/ |
|
Back to top |
|
|
DawgG l33t
Joined: 17 Sep 2003 Posts: 866
|
Posted: Fri Jun 29, 2018 11:30 am Post subject: |
|
|
Quote: | Do NOT put the size of your disk drive here. Instead, if you want Squid to use the entire disk drive, subtract 20% and use that value. |
yeah, seen&done that, too :wink:
i set
Code: | cache_swap_low 70
cache_swap_high 80 |
(default was 90/95) in order to evict objects earlier and managed to restart squid w/out having to erase the whole cache. fs-usage went down to 60-70GB in between but now the service went down again b/c "disk full."
i reduced the above settings to 60/75 but if this does not help it's time for the squid-mailing-list, i guess. _________________ DUMM KLICKT GUT. |
|
Back to top |
|
|
bunder Bodhisattva
Joined: 10 Apr 2004 Posts: 5934
|
Posted: Fri Jun 29, 2018 11:41 am Post subject: |
|
|
have you tried with diskd instead of aufs? mine seems to be working fine...
Code: | cache_dir diskd /var/cache/squid 8000 16 256
|
_________________
Neddyseagoon wrote: | The problem with leaving is that you can only do it once and it reduces your influence. |
banned from #gentoo since sept 2017 |
|
Back to top |
|
|
DawgG l33t
Joined: 17 Sep 2003 Posts: 866
|
Posted: Sat Jun 30, 2018 4:09 pm Post subject: |
|
|
@bunder
thx for your reply, but i am quite certain the store-type is not the problem since i run an (almost) identical squid-server (cache_dir gets 200 of 250GB) w/out all those modifications for windoze-updates and it's been running flawlessly for years
Code: | cache_dir aufs /mnt/cache/squid 200000 32 51
squid ~ # df -h
/dev/sdb1 247G 179G 56G 77% /mnt/cache/squid
|
likely a case for the squid mailing list (or squid's bugzilla). _________________ DUMM KLICKT GUT. |
|
Back to top |
|
|
DawgG l33t
Joined: 17 Sep 2003 Posts: 866
|
Posted: Sat Jul 14, 2018 10:02 am Post subject: |
|
|
i asked the squid mailinglist and (most likely, not completely resolved yet) the reason is that (an SMP-setting for squid) cannot be used with aufs. it is supposed to work with ufs, but in a quick test my impression was that it dows not work with that, either.
i turned of the SMP-setting, but will look into it further (as i still think this is related to those windoze-update-settings.) _________________ DUMM KLICKT GUT. |
|
Back to top |
|
|
|