I just figured that you would never want an explicit zero _ms timeout, if you did well you could already do that with the existing one based in seconds. So the only time you want _ms would be non-zero so then it should override the non-_ms ________________________________________ From: Leif Hedstrom <zw...@apache.org> Sent: Friday, August 17, 2018 5:59 PM To: dev@trafficserver.apache.org Subject: [EXTERNAL] Re: [PROPOSAL] Change *attempts_timeout to be in milliseconds
Meh, I see the PR with treating non zero values taking precedence. That works, with the somewhat odd behavior that an explicit zero cannot override the old value . I think that’s fine though. I agree with amc that for 9.0.0 (current master) we should cleanup this consistently, floats seems fine. — Leif > On Aug 17, 2018, at 15:46, Leif Hedstrom <zw...@apache.org> wrote: > > As suggested elsewhere, having some new configs might be the safest > transitions. We can then deprecate, and later remove, the 1s ones. > > They would have to take precedence, but that can be tricky to deal with as a > undefined value :-/. Maybe the smallest non-zero value should be used ? > > — Leif > >> On Aug 15, 2018, at 08:00, "evan.zelkow...@gmail.com" >> <evan.zelkow...@gmail.com> wrote: >> >> We started seeing some problems with some of our origins/parents and we >> believe that if we had finer control over the attempts timeout it might help >> alleviate some problems. Currently all of these timeouts are in seconds, and >> when it comes to live streaming with 2sec fragments that granularity may not >> be enough. >> >> So proposing that we move these to millisecond based. I had discussed on IRC >> some other options such as switching to float, but that can bring in >> precision issues, and possibly adding units values but that would be a first >> for a record value and would seem to be more of a wide change to bring in >> units parsing where it has not been used before. So just changing to ms >> seemed the most straight forward. >
