Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
Just to follow up on this, I was looking at the wrong log file and I was certainly running out of file descriptors. The default config from squid-3.3.9 also didn't have a cache_dir statement which I was used to having. Not sure if this affected anything but it was another issue. On Tue, 17 Sep 2013, Brett Lymn wrote: (sorry if this messes up the threading I just joined the list to respone to this thread) Firstly, I have been using squid on NetBSD for many years and it works fine for me. I am running squid 3.3.8 on a recent-ish NetBSD-current without issue. I wouldn't totally rule out a problem with squid and NetBSD but there are many many things that can impact the performance of squid which need to be looked at first. As a general note, if you are going to post your squid.conf to a mailing list, please do everyone a favour and do something like: grep -v '^\(#\|[ ]*$\)' /usr/pkg/etc/squid/squid.conf and post that output - this will strip the comments and blank lines leaving a more readable summary of what you actually have in the configuration file (BTW there is a tab and a space between the square brackets above). One thing that can help a lot in working out what is going on with squid is to look in the cache.log file, sometimes slowness and cpu utilisation can be caused by squid restarting which can happen if its child processes (redirector, authenticators, external acl and so on) are exiting, if these processes exit too often/quickly then squid will restart. Similarly, if too many requests waiting to be serviced by the helpers then squid will restart. You should see indications of these restarts and the reasons why in the cache.log. Another thing it may be is squid running out of file descriptors, again, this will be mentioned in the cache.log, try adding: ulimit -n 3000 near the top of /etc/rc.d/squid and restarting squid. This probably should be added to the sample start up script as squid has recently stopped managing the limit itself and now expects a suitable file handle limit. Also, it may be unrealistic expectations - one common problem is that people configure a large disk cache on a memory limited machine. Squid requires around 10 - 14MB or memory per GB of cache you specify simply for managing the cache contents.. it will use more on top of that. So, for example, trying to configure squid with a 100GB cache will result in squid needing more than 1 - 1.4GB of RAM. If you try that on a machine with only 2GB of memory then performance will be poor because the machine will be paging to try and keep up with squid and everything else as well. HTH - if not, please look in the cache.log and post any errors you find, they may provide some clues... also a stripped squid.conf can help. -- Hisashi T Fujinaka - ht...@twofifty.com BSEE(6/86) + BSChem(3/95) + BAEnglish(8/95) + MSCS(8/03) + $2.50 = latte
Re: [Fwd: Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time]
Thus wrote Adrian Immanuel Kieß (adr...@kiess.at): No comment. I've considered this a yes and have acted accordingly. regards, spz
Re: [Fwd: Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time]
On Wed, 09 Oct 2013 19:01:22 +0200, Adrian Immanuel Kieß wrote: No comment. Publishing private mail... so, yes, you are a troll. hauke -- The ASCII Ribbon CampaignHauke Fath () No HTML/RTF in emailInstitut für Nachrichtentechnik /\ No Word docs in email TU Darmstadt Respect for open standards Ruf +49-6151-16-3281
Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
I guess there are many trolls around here these days. Gosh! Adrian Immanuel KIESS On Tue, 2013-09-17 at 21:10 +0930, Brett Lymn wrote: (sorry if this messes up the threading I just joined the list to respone to this thread) Firstly, I have been using squid on NetBSD for many years and it works fine for me. I am running squid 3.3.8 on a recent-ish NetBSD-current without issue. I wouldn't totally rule out a problem with squid and NetBSD but there are many many things that can impact the performance of squid which need to be looked at first. As a general note, if you are going to post your squid.conf to a mailing list, please do everyone a favour and do something like: grep -v '^\(#\|[ ]*$\)' /usr/pkg/etc/squid/squid.conf and post that output - this will strip the comments and blank lines leaving a more readable summary of what you actually have in the configuration file (BTW there is a tab and a space between the square brackets above). One thing that can help a lot in working out what is going on with squid is to look in the cache.log file, sometimes slowness and cpu utilisation can be caused by squid restarting which can happen if its child processes (redirector, authenticators, external acl and so on) are exiting, if these processes exit too often/quickly then squid will restart. Similarly, if too many requests waiting to be serviced by the helpers then squid will restart. You should see indications of these restarts and the reasons why in the cache.log. Another thing it may be is squid running out of file descriptors, again, this will be mentioned in the cache.log, try adding: ulimit -n 3000 near the top of /etc/rc.d/squid and restarting squid. This probably should be added to the sample start up script as squid has recently stopped managing the limit itself and now expects a suitable file handle limit. Also, it may be unrealistic expectations - one common problem is that people configure a large disk cache on a memory limited machine. Squid requires around 10 - 14MB or memory per GB of cache you specify simply for managing the cache contents.. it will use more on top of that. So, for example, trying to configure squid with a 100GB cache will result in squid needing more than 1 - 1.4GB of RAM. If you try that on a machine with only 2GB of memory then performance will be poor because the machine will be paging to try and keep up with squid and everything else as well. HTH - if not, please look in the cache.log and post any errors you find, they may provide some clues... also a stripped squid.conf can help. -- With greetings from Leipzig, Germany. Adrian Immanuel Kieß Administrator programmer Unix / Perl / LaTeX mail: adrian (at) kiess.at www: http://www.kiess.at signature.asc Description: This is a digitally signed message part
[Fwd: Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time]
Have to dispense here. :-D -- With greetings from Leipzig, Germany. Adrian Immanuel Kieß Administrator programmer Unix / Perl / LaTeX mail: adrian (at) kiess.at www: http://www.kiess.at ---BeginMessage--- On Tue, Sep 17, 2013 at 02:43:43PM +0200, Adrian Immanuel Kie? wrote: I guess there are many trolls around here these days. oh, ok - sorry for feeding you then. -- Brett Lymn Staple Guns: because duct tape doesn't make that KerCHUNK sound - xkcd.com ---End Message--- signature.asc Description: This is a digitally signed message part
Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
ktrace is a command, look it up. also testing dns time was a very good suggestion, as well. On Sun, Sep 15, 2013 at 7:12 AM, Adrian Immanuel Kieß adr...@kiess.atwrote: Dear List, Metthew and Matthias, I don't know how to trace a application within NetBSD -- and yes it must be cache related. After migrating, if I remeber right, I even tried to rebuild the cache by hand to check if everything is OK. I guess it loops around somewhere with wrong OPENDEV command for the network interface. My second guess it is busy scanning the HTML file or something. I just can guess. :-) I even added max_filedescriptors 4096 in squid.conf because I've stumbled across something with the open file limit in NetBSD already. ^_^ As I already noted it worked under Debian. But under NetBSD 6.1 I now have the high CPU usage. I attached my squid.conf file. Adrian Immanuel KIESS On Sat, 2013-09-14 at 14:24 -0400, matthew sporleder wrote: On Thu, Sep 12, 2013 at 11:22 PM, Adrian Immanuel Kieß adr...@kiess.at wrote: Dear list, since several months and upgrades I encountered the problem that squid uses too much CPU time under NetBSD 6.1 and my proxy server is nearly to unusable now. I installed the default squid from /usr/pkgsrc/www/squid using pkgsrc-2013-Q2. Every HTTP webpage request lets squid meditate for several seconds until the page is served. The network download itself seems comparatively fast. Maybe someone seen this behavior of squid too and give me a hint? I moved my squid installation from GNU/Debian to this NetBSD box and I remember -- there I did not have this problem with nearly the same configuration. Thank you very much! Sincerely, Adrian Immanuel KIESS -- With greetings from Leipzig, Germany. Adrian Immanuel Kieß Administrator programmer Unix / Perl / LaTeX mail: adrian (at) kiess.at www: http://www.kiess.at Have you tried ktrace-ing squid to see what it's doing during these pauses? It sounds like it's probably scanning/writing the cache? -- With greetings from Leipzig, Germany. Adrian Immanuel Kieß Administrator programmer Unix / Perl / LaTeX mail: adrian (at) kiess.at www: http://www.kiess.at
Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
On Sun, Sep 15, 2013 at 01:39:33PM +0200, Adrian Immanuel Kieß wrote: I thank you for your suggestions but testing the DNS time is nonsense because my DNS server runs on another boxen and the IPNAT configuration on this NetBSD boxen works very fine because without the squid proxy I can connect very much faster to the outer world. You are making the assumption here that the behaviour you observe in a browser allows conclusion to Squid's behaviour. So please humour me and run those two commands. It will only take a few seconds. Kind regards -- Matthias Scheler http://zhadum.org.uk/
Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
On Sun, Sep 15, 2013 at 08:44:42AM -0700, Hisashi T Fujinaka wrote: On Sun, 15 Sep 2013, Matthias Scheler wrote: On Sun, Sep 15, 2013 at 01:39:33PM +0200, Adrian Immanuel Kie? wrote: I thank you for your suggestions but testing the DNS time is nonsense because my DNS server runs on another boxen and the IPNAT configuration on this NetBSD boxen works very fine because without the squid proxy I can connect very much faster to the outer world. You are making the assumption here that the behaviour you observe in a browser allows conclusion to Squid's behaviour. So please humour me and run those two commands. It will only take a few seconds. There does appear to be something that uses a lot of CPU in squid. I've just been restarting it several times a day. Okay. ktrace and what was the other command you wanted him to run? I'll look into this as well. My commands were related to the big delays in page loads. If you don't experience those it is not related. Kind regards -- Matthias Scheler http://zhadum.org.uk/
Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
On Sun, 15 Sep 2013, Matthias Scheler wrote: On Sun, Sep 15, 2013 at 08:44:42AM -0700, Hisashi T Fujinaka wrote: On Sun, 15 Sep 2013, Matthias Scheler wrote: On Sun, Sep 15, 2013 at 01:39:33PM +0200, Adrian Immanuel Kie? wrote: I thank you for your suggestions but testing the DNS time is nonsense because my DNS server runs on another boxen and the IPNAT configuration on this NetBSD boxen works very fine because without the squid proxy I can connect very much faster to the outer world. You are making the assumption here that the behaviour you observe in a browser allows conclusion to Squid's behaviour. So please humour me and run those two commands. It will only take a few seconds. There does appear to be something that uses a lot of CPU in squid. I've just been restarting it several times a day. Okay. ktrace and what was the other command you wanted him to run? I'll look into this as well. My commands were related to the big delays in page loads. If you don't experience those it is not related. I usually see the delays, and then check to see if I'm using 100% CPU (which I am). I'm guessing it's some weird spin lock, but I haven't looked into it yet. I probably have time today. -- Hisashi T Fujinaka - ht...@twofifty.com BSEE(6/86) + BSChem(3/95) + BAEnglish(8/95) + MSCS(8/03) + $2.50 = latte
Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
On Thu, Sep 12, 2013 at 11:22 PM, Adrian Immanuel Kieß adr...@kiess.atwrote: Dear list, since several months and upgrades I encountered the problem that squid uses too much CPU time under NetBSD 6.1 and my proxy server is nearly to unusable now. I installed the default squid from /usr/pkgsrc/www/squid using pkgsrc-2013-Q2. Every HTTP webpage request lets squid meditate for several seconds until the page is served. The network download itself seems comparatively fast. Maybe someone seen this behavior of squid too and give me a hint? I moved my squid installation from GNU/Debian to this NetBSD box and I remember -- there I did not have this problem with nearly the same configuration. Thank you very much! Sincerely, Adrian Immanuel KIESS -- With greetings from Leipzig, Germany. Adrian Immanuel Kieß Administrator programmer Unix / Perl / LaTeX mail: adrian (at) kiess.at www: http://www.kiess.at Have you tried ktrace-ing squid to see what it's doing during these pauses? It sounds like it's probably scanning/writing the cache?
Re: netbsd-6.1: squid from pkgsrc-2013-Q2 uses too much CPU time
On Fri, Sep 13, 2013 at 05:22:47AM +0200, Adrian Immanuel Kieß wrote: I installed the default squid from /usr/pkgsrc/www/squid using pkgsrc-2013-Q2. Every HTTP webpage request lets squid meditate for several seconds until the page is served. The network download itself seems comparatively fast. Can you please run the following two commands and tell us the output? time host -t www.google.com. time host -t a www.google.com. Thanks in advance -- Matthias Scheler http://zhadum.org.uk/
