My only concern with that was more the conversion from old settings. So say the new setting uses the same name but is now a float, is ats going to throw a warning in the diags logs and then just completely skip the user setting for that variable because it has the wrong type? If thats not the case, where in records.config you could have var float ###, and even though its an int internally and ATS does not care, then yea I would agree that would be the easiest way ________________________________________ From: Leif Hedstrom <zw...@apache.org> Sent: Sunday, August 19, 2018 4:37 PM To: dev@trafficserver.apache.org Subject: [EXTERNAL] Re: [Proposal] Change *attempts_timeout to milliseconds from seconds
> On Aug 19, 2018, at 13:49, John Rushford <jjrushf...@gmail.com> wrote: > > Has using a floating point config variable been completely ruled out? For > example, a timeout of 2.1 can easily be converted to an unsigned int in > milliseconds (2.1 * 1000) for use and no suffix is required. My previous > settings remain unaffected and are just converted to milliseconds. Just > update the docs to indicate the use of a floting point config to specify > fractions of a second. +1 — Leif > > On Fri, Aug 17, 2018 at 3:34 PM Zelkowitz, Evan <evan_zelkow...@comcast.com> > wrote: > >> Ok, added a parallel _ms variable for the 3 timeout values, >> https://github.com/apache/trafficserver/pull/4130 . So if one of the _ms >> versions is set it will override the non-_ms >> >> On 8/17/18, 6:59 AM, "Alan Carroll" <solidwallofc...@oath.com.INVALID> >> wrote: >> >> Not directly, but I do think that when this gets converted to YAML we >> make >> *all* durations seconds by default, with a suffix or metric option to >> specify other time units. For a stop gap, Walt might be right and we >> could >> just add a parallel configuration variable with "_ms" at the end to >> specify >> this value in milliseconds on the presumption that will get cleaned up >> during the conversion. >> >> On Thu, Aug 16, 2018 at 7:35 PM Zelkowitz, Evan < >> evan_zelkow...@comcast.com> >> wrote: >> >>> Anyone have any other thoughts? As of now Im contemplating doing the >>> change to _ms internally, since at least that might help to mitigate >>> records settings confusion across a CDN in the short term if we need >> this >>> for our own purposes. I do like the idea of having types when it's >> time to >>> yamilfy everything but didnt know if people would mind a short term >> change >>> in names before then (like for 8.0.x) or if we hold off doing this >> in the >>> community until yaml time >>> ________________________________________ >>> From: Walt Karas <wka...@oath.com.INVALID> >>> Sent: Wednesday, August 15, 2018 12:45 PM >>> To: dev@trafficserver.apache.org >>> Subject: [EXTERNAL] Re: [Proposal] Change *attempts_timeout to >>> milliseconds from seconds >>> >>> Or jump off. >>> >>> >>> On Wed, Aug 15, 2018 at 1:27 PM, Alan Carroll >>> <solidwallofc...@oath.com.invalid> wrote: >>>> I thought about that, but I think that's not the best approach >> because we >>>> want to make it more general, so it can have seconds or >> milliseconds. >>> OTOH >>>> maybe that will have to wait until we YAMLize it. If we could do >>> anything, >>>> I'd add the type "TIME" vs. "INT" and "STRING" and have that be >> seconds >>>> unless modified with a suffix like "ms" or "minutes", etc. I may >> take a >>>> look at how hard that would be. Unless Walt wants to jump in? >>>> >>>> On Wed, Aug 15, 2018 at 1:22 PM Walt Karas <wka...@oath.com.invalid >>> >>> wrote: >>>> >>>>> Would it perhaps be a good idea to add a _ms suffix to the config >>>>> variable names of timeouts that are not seconds? >>>>> >>>>> On Wed, Aug 15, 2018 at 12:15 PM, Zelkowitz, Evan >>>>> <evan_zelkow...@comcast.com> wrote: >>>>>> ?Currently all the attempts_timeout values are in seconds. We >> have >>> seen >>>>> some issues where we believe being able to have a finer >> granularity over >>>>> this may help alleviate some problems. Also for live streaming >> video >>> in 2 >>>>> second fragments a 1 second timeout does not provide as much >> control. So >>>>> proposing that we change these to milliseconds. >>>>>> >>>>>> >>>>>> We briefly discussed on IRC some alternatives such as using >> floats, >>> but >>>>> that could introduce precision issues, as well as adding values >> such as >>>>> 'ms' 's' etc. The latter might be nice but that is a large change >> since >>>>> there currently is no units parsing for values in records.config >> so it >>>>> might be nice to have in the future but thats a much larger scope >> of >>> work. >>>>> >>>> >>>> >>>> -- >>>> *Beware the fisherman who's casting out his line in to a dried up >>> riverbed.* >>>> *Oh don't try to tell him 'cause he won't believe. Throw some >> bread to >>> the >>>> ducks instead.* >>>> *It's easier that way. *- Genesis : Duke : VI 25-28 >>> >> >> >> -- >> *Beware the fisherman who's casting out his line in to a dried up >> riverbed.* >> *Oh don't try to tell him 'cause he won't believe. Throw some bread to >> the >> ducks instead.* >> *It's easier that way. *- Genesis : Duke : VI 25-28 >> >> >> > > -- > John Rushford > jjrushf...@gmail.com
