Re: [RFC] Policy Change on half-closed connection
Amos, your squid-users announcement says that it applies to 2.7 as well; will the same changes (switch default, remove later) be applied there? FWIW, I've never seen this happen on 2.x. Has anyone else? Note that turning half_closed_clients off has an effect on quick_abort (at least in 2.x); without it, if the client aborts before headers are received, quick_abort doesn't have time to kick in; with it, it will operate even if the client aborts that early. This behaviour is pretty desirable in an accelerator setup... Cheers, On 13/09/2008, at 6:52 PM, Amos Jeffries wrote: Basically bug 2431 http://www.squid-cache.org/bugs/show_bug.cgi?id=2431. Alex mentions seeing bad side effects and it being maybe unsuitable for a stable release. I posit that it is suitable to change the cause of a major bug in the stable release and a config default is safer than a core code change. As for further testing, we have a few users who have encountered the bug already and tested the config change. With good results. Henrik reports that many of his customers have done so as well. People who have explicitly set their configuration and are happy are not affected. For those who need to enable it, I think we have 3- HEAD and 2.7. Amos -- Please use Squid 2.7.STABLE4 or 3.0.STABLE9 -- Mark Nottingham [EMAIL PROTECTED]
Re: [RFC] Policy Change on half-closed connection
Mark Nottingham wrote: Amos, your squid-users announcement says that it applies to 2.7 as well; will the same changes (switch default, remove later) be applied there? I'm not sure, Henrik will make that decision. Though there is no bug reason for it to happen in 2.7. FWIW, I've never seen this happen on 2.x. Has anyone else? Not that Henrik is aware of. Thus the RFC. We just need it to be tested on all installs so we don't change defaults behind people backs, and rip the code out of 3.2 then find its needed by someone later who was on 2.x today. Note that turning half_closed_clients off has an effect on quick_abort (at least in 2.x); without it, if the client aborts before headers are received, quick_abort doesn't have time to kick in; with it, it will operate even if the client aborts that early. This behaviour is pretty desirable in an accelerator setup... Noted. Thank you. That may just be the very reason for keeping it coded, but in off state by default. Cheers, On 13/09/2008, at 6:52 PM, Amos Jeffries wrote: Basically bug 2431 http://www.squid-cache.org/bugs/show_bug.cgi?id=2431. Alex mentions seeing bad side effects and it being maybe unsuitable for a stable release. I posit that it is suitable to change the cause of a major bug in the stable release and a config default is safer than a core code change. As for further testing, we have a few users who have encountered the bug already and tested the config change. With good results. Henrik reports that many of his customers have done so as well. People who have explicitly set their configuration and are happy are not affected. For those who need to enable it, I think we have 3-HEAD and 2.7. Amos -- Please use Squid 2.7.STABLE4 or 3.0.STABLE9
Re: [RFC] Policy Change on half-closed connection
On Fri, 2008-09-19 at 09:16 +1000, Mark Nottingham wrote: Amos, your squid-users announcement says that it applies to 2.7 as well; will the same changes (switch default, remove later) be applied there? FWIW, I've never seen this happen on 2.x. Has anyone else? Note that turning half_closed_clients off has an effect on quick_abort (at least in 2.x); without it, if the client aborts before headers are received, quick_abort doesn't have time to kick in; with it, it will operate even if the client aborts that early. This behaviour is pretty desirable in an accelerator setup... Sounds like a big enough problem to leave Squid2 settings alone! This scenario should also be tested with Squid3. Thank you, Alex. On 13/09/2008, at 6:52 PM, Amos Jeffries wrote: Basically bug 2431 http://www.squid-cache.org/bugs/show_bug.cgi?id=2431. Alex mentions seeing bad side effects and it being maybe unsuitable for a stable release. I posit that it is suitable to change the cause of a major bug in the stable release and a config default is safer than a core code change. As for further testing, we have a few users who have encountered the bug already and tested the config change. With good results. Henrik reports that many of his customers have done so as well. People who have explicitly set their configuration and are happy are not affected. For those who need to enable it, I think we have 3- HEAD and 2.7. Amos -- Please use Squid 2.7.STABLE4 or 3.0.STABLE9 -- Mark Nottingham [EMAIL PROTECTED]
