> 在 2015年1月28日，下午11:48，Bardwell, William <wbard...@akamai.com> 写道： > > Proposal #1 sounds reasonable to me, although I would say that any CC header > should take effect when enabled == 1, not just max-age but also any sort of > no-store or similar "don't caching" value. (Maybe that is what you already > meant.) I could also see having min and max lifetime values which get > applied no matter what, which could replace having a different mode. (and > maybe the existing setting would be the max value.) Then you would just have > if (enabled != 0) > ttl = min(max(expires or CC calculated value, min value), max value); > (However note that my patch would also support ignore_server_no_cache, but > maybe there should be a separate config just like that just but for negative > caching.) Yes, you get what I mean, and the min() is a good idea. and I think ignore_server_no_cache should be a less affective control when we talking about negative specified response, should we?
> > Should I still push my patch as an intermediate step? If not I can just put > it on the jira ticket and someone can take it as raw material. why not put it on the jira first? any changes we make here will change the behavior of the negative caching, I think many people will care, let us make it in carefully. > > Isn't TS-2633 indicating that 'current' isn't actually working right now? I > do see in the code that some of the code around negative_caching_lifetime is > looking to see if there is an EXPIRES header...but not checking for a CC > header...which is odd, but also means that 'current' isn't really what it > does now. It is always a dark corner when the state is not that commonly looked, bugs everywhere so we should fix it, haha > > -William Bardwell > ________________________________________ > From: Yongming Zhao <ming....@gmail.com> > Sent: Tuesday, January 27, 2015 9:29 PM > To: dev@trafficserver.apache.org > Subject: negative expire time > > well, TS-3331 TS-2633 all talking about negative caching issues, during the > change, I’d proposal a more adaptable change for your comments: > > > The negative cache is causing some confusion when it is cached by accident, > the current state machine working as: > > ===current=== > if expire time set(by Expire or CC-maxage) > then set expire time according to origin server response. > else > if http.negative_caching_enabled = 1 ( which default to 0 ) > then set expire time to http.negative_caching_lifetime ( which default to > 1800s ) > else > no cache > > > > when some origin server in trouble, there will be very common case to get > some negative response with expire time set, and long expire time as 200 > responses maybe. that will give us trouble when we running in CDN service. > so there should make some change on that state machine to make a forced > expire time when ATS get a negatived response from origin: > > ===proposal #1=== > case http.negative_caching_enabled = 0 ( which default to 0 ) > no cache at all ( that is, remove the expiretime even it have ) > case http.negative_caching_enabled = 1 > if expire time set(by Expire or CC-maxage) > then set expire time according to origin server response. > else > set expire time to http.negative_caching_lifetime ( which default to 1800s ) > case http.negative_caching_enabled = 2 > then set expire time to http.negative_caching_lifetime to all ( mostly we > should set a very short expire time here ) > > > well, a modified version that will compatible with current codes, we force > all the expire time to negative_caching_lifetime if > http.negative_caching_enabled = 2 > > === proposal #2=== > if expire time set(by Expire or CC-maxage) > if http.negative_caching_enabled = 2 > then set ALL negative response expire time to > http.negative_caching_lifetime ( mostly we should set a very short expire > time here ) > else > then set expire time according to origin server response. > else > if http.negative_caching_enabled > 0 ( which default to 0 ) > then set expire time to http.negative_caching_lifetime ( which default to > 1800s ) ### this set no expire time negative to 1800s, mostly we should set a > very short expire time here > else > no cache > > > > thanks for your comments > > > - Yongming Zhao 赵永明 - Yongming Zhao 赵永明
