+1 to making more things built-in. I think defaulting to use the timezone on the machine on which the grunt/Pig is invoked, followed by an option to specify what timezone to use will be the best way to go.
Because that way, at least all compute nodes will be computing uniformly using the same timezone. And if the timezone on the machine on which the frontend ran is off, then that's easier to track down. Finally, of course allow for it to be configured inside PigLatin as well. On Fri, Jul 2, 2010 at 8:16 PM, Corbin Hoenes <[email protected]> wrote: > Nice! > > Sent from my iPhone > > > On Jul 2, 2010, at 6:42 PM, Russell Jurney <[email protected]> > wrote: > > I'm planning to get rid of piggybank datetime stuff and make jodatime a >> core >> dependency to these builtin functions. No REGISTER required. >> >> Russ >> >> On Fri, Jul 2, 2010 at 2:27 PM, Corbin Hoenes <[email protected]> wrote: >> >> So is your plan to package the jodatime dependency as part of piggybank >>> or >>> will it require 2 REGISTER jar commands one for piggybank and one for >>> joda? >>> >>> On Jul 2, 2010, at 3:02 PM, Russell Jurney wrote: >>> >>> I was planning to rely completely on Jodatime there, and I believe its >>>> default behavior is to use the local timezone of the machine it is on. >>>> >>> ISO >>> >>>> time encodes the time zone, and jodatime does the conversions back and >>>> forth, taking it into account. >>>> >>>> Should this be a pig config field? You can specify a timezone in >>>> >>> jodatime, >>> >>>> so that should work out ok, it would limit hadoop node shenanigans. >>>> >>>> Russ >>>> >>>> On Fri, Jul 2, 2010 at 1:05 PM, hc busy <[email protected]> wrote: >>>> >>>> I wonder if we can pull timezone support into the system... So some day >>>>> >>>> I >>> >>>> can say things like >>>>> >>>>> >>>>> 'jiffies between 11am EST and 13:11 IST' >>>>> >>>>> and get back the correct answer in PigLatin. >>>>> >>>>> also, how does ISO time know the current time zone? >>>>> >>>>> I've seen people run into this situation where the compute node and >>>>> name >>>>> node are in different time zones (by accident of course, they're >>>>> >>>> physically >>> >>>> in the same room), and it produces some problems that are difficult to >>>>> track >>>>> down. >>>>> >>>>> so, at least your UnixToISO will need to have some kind of convention >>>>> >>>> for >>> >>>> knowing what timezone to send it into, right? What's the plan there? >>>>> >>>>> >>>>> >>>>> On Thu, Jul 1, 2010 at 11:28 PM, Russell Jurney < >>>>> >>>> [email protected] >>> >>>> wrote: >>>>>> >>>>> >>>>> I though I would ping pig-user about this to get comments. I put a >>>>>> proposal >>>>>> for builtin date functions I can have ready for Pig 0.8 at >>>>>> http://issues.apache.org/jira/browse/PIG-1430 >>>>>> >>>>>> There are so many classes to add a type, given my time constraints I >>>>>> >>>>> don't >>>>> >>>>>> see the benefit of a full datetime type over this proposal right now. >>>>>> These >>>>>> functions should be backwards compatible with a full-blown datetime >>>>>> >>>>> type. >>> >>>> >>>>>> Russ >>>>>> >>>>>> >>>>> >>> >>>
