I haven't tried this, but you may be able to use the socket api with some NTP client code to determine the actual clock skew of each instance and use that to compensate.
Jeff On Sun, Nov 3, 2013 at 4:06 PM, Kaan Soral <[email protected]> wrote: > Thanks for all your replies > > My usecase tolerates time differences, as each item has a unique id which > is checked before applying, that's why I was able to detect it, otherwise > applied cache results would be off sometimes > > One solution would be to delay stuff by 15 seconds etc, so that even if > there is a <15 second skew, it would be tolerated, I'm just going to ignore > <15 second time delays for know and see if there are >15 second time delays > > It might have been a good idea if there was a Clock API :) - I wonder if > ndb entities get their default datetime's from a unified clock > > > On Monday, November 4, 2013 12:05:11 AM UTC+2, Chad Vincent wrote: >> >> It seems to be fixed, but at one point on a previous project we were >> seeing clock skews of over 30 minutes. >> >> On Saturday, November 2, 2013 7:33:32 AM UTC-4, timh wrote: >>> >>> I have at times experienced definite clock skews (I documented some up >>> as much a minute in the early stages.) >>> >>> I personally think it is a mistake to design a solution that needs < 5 >>> sec clock accuracy with appengine. >>> >>> T >>> >>> On Saturday, November 2, 2013 12:32:43 PM UTC+8, Vinny P wrote: >>>> >>>> On Fri, Nov 1, 2013 at 9:44 PM, Kaan Soral <[email protected]> wrote: >>>> >>>>> I've been seeing the errors pop up, but it's always between datetime's >>>>> that differ only by 2-3 seconds >>>>> Than I remembered a *really* old discussion about instance time >>>>> differences and it seems that the cause of these 2-3 second differences is >>>>> instance time differences. Could this be the case? >>>>> >>>> >>>> >>>> Yes. >>>> >>>> Instance clocks are not guaranteed to be accurate - there is always >>>> some skew, especially for large applications which span many instances ( in >>>> the physical sense, span multiple machines which may have some clock drift >>>> ). It's rarely mentioned though, since few applications need that accurate >>>> of a clock. >>>> >>>> >>>> On Fri, Nov 1, 2013 at 9:44 PM, Kaan Soral <[email protected]> wrote: >>>> >>>>> I'm thinking of logging an error only-if the difference is >>>>> >some_seconds, however unsure how many seconds that should be, is there an >>>>> upper limit to this time difference, if it exists? >>>>> >>>>> >>>> >>>> When App Engine was still in beta, I personally noted clock skews >>>> exceeding 30 seconds, sometimes close to a minute. With that said, instance >>>> clocks seem to be synchronized to a much closer standard nowadays. If I had >>>> to suggest an upper limit for you, I would go with around 10-15 seconds. >>>> Note that I'm not claiming this to be a hard and fast number - it's my >>>> rough approximation. Feel free to ignore it or go with a lower number. >>>> >>>> >>>> ----------------- >>>> -Vinny P >>>> Technology & Media Advisor >>>> Chicago, IL >>>> >>>> App Engine Code Samples: http://www.learntogoogleit.com >>>> >>>> >>> -- > You received this message because you are subscribed to the Google Groups > "Google App Engine" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at http://groups.google.com/group/google-appengine. > For more options, visit https://groups.google.com/groups/opt_out. > -- You received this message because you are subscribed to the Google Groups "Google App Engine" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/google-appengine. For more options, visit https://groups.google.com/groups/opt_out.
