Having used this for debugging in some trying circumstances, the knowledge that the most recent events in the history are in fact the most recent events was critical. The circularity could easily be handled by providing iterators that masked the implementation. One might also note that if the buffer is big enough to not overflow, it's irrelevant if it's circular.
Let's just leave it for now, I was just surprised at this change. I may try to find time to do the iterator approach. On Fri, Nov 10, 2017 at 3:34 AM, Chao Xu <ok...@apache.org> wrote: > Hi AMC, > > It could get confused if circle buffer overwrote. > Therefore, in my opinion, linear buffer is better than circular buffer. > > Oknet > > 2017-11-10 7:05 GMT+08:00 scw00 <616955...@qq.com>: > > > Hi AMC > > > > > > Yes I’ve changed history in the linear buffer because the bug is not > > always related to the last records . And buffer is large enough to hold > all > > records currently! > > > > > > If you want . I can change it back to circle buffer. > > > > > > > > > > scw00 > > > > > > ------------------ Original ------------------ > > From: Alan Carroll <solidwallofc...@oath.com.INVALID> > > Date: 周五,11月 10,2017 4:07 上午 > > To: dev@trafficserver.apache.org <dev@trafficserver.apache.org> > > Subject: Re: Change to HttpSM history > > > > > > > > I was looking at commit for e634659783bcadbedeb1bca411d36640d18b3b59 PR > > 2272. This moves the history mechanism to separate file. It also changes > > history from a circular buffer to a linear buffer. This means the events > > stored in it are the first events for the HttpSM, not necessarily the > most > > recent ones. I consider this a rather significant change but I see no > > discussion of that in the PR and I don't remember it on the mailing list. > > Did I miss it? I much prefer the circular buffer style. > > >
