Yup, thank you, Amos, for details. I do not remember all changelogs exactly. 29.05.2017 3:09, Amos Jeffries пишет: > On 29/05/17 05:21, Yuri wrote: >> >> 28.05.2017 21:09, Eduardo Carneiro пишет: >>> Ok. Tell me what technical details you need and I post here. >> Configs. Logs. Topology. Usecases. Examples for single transactions. >>> But if this were an ignoring cache-control issue, wouldn't that >>> happen on >>> squid 3.5.19 and previous versions as well? >>> >>> With the same conf works on 3.5.19 but not in newest squid3 versions. >> Once more. Ignoring cache-control in squid is deprecated due to RFC >> requirements. So, it leads to drops down with byte HIT in general. I do >> not remember when it exactly desupported in squid's code, but seems >> approx. at some 3.5.x. > > From 3.5.19 onwards .. > > 3.5.21 added support for Squid to differentiate between revalidation > checks that were always vs sometimes necessary. Fixing the situation > where responses had both a CC:private or CC:no-cache header and > information stating cacheability "freshness". > > 3.5.22 fixed issues that were then seen with missing Last-Modified and > old Date values, and the Collapsed Forwarding feature. > > 3.5.23 added support for caching objects with "Vary: *" which depended > on the change in .21, and caching of 3034 responses. Also fixed the > main outstanding issue behind Host verify MISS happening, and a > regression in the .22 collapsed forwarding change. > > 3.5.24 fixed a regression in the "cache deny" access control handling > which was causing some thing to be cached and revalidated when they > should not have. This may make some things now un-cacheable if there > is a misconfiguration in that directive. > > 3.5.25 fixed a stall that happens when SSL-Bump interacts with Host > verification through the SNI value. While not strictly related, having > it fixed allows a lot more Google/YouTube/Akamai transactions to reach > the caching stages so their operations start to be visible instead of > splice/tunnel'ed. > > > The expected effect of these changes was to fix several major > unnecessary bandwidth uses; > - cached YouTube videos starting mid-video when someone else using > the proxy had only partially watched the same video - causing the user > to force-refresh and purge the cached video. > - Chrome and similar Google originated downloads being a MISS for > identical objects. > - the often cited terrible HIT reduction on some major hosting > services (again Google, though also Akamai) traffic since 3.2 Host > verify was added. > - collapsed forwarding resulting in large objects being stored to the > cache then immediately deleted by a parallel fetch. > > > Amos > _______________________________________________ > squid-users mailing list > squid-users@xxxxxxxxxxxxxxxxxxxxx > http://lists.squid-cache.org/listinfo/squid-users
Attachment:
signature.asc
Description: OpenPGP digital signature
_______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
