Hi folks, I'm proposing a change to Pig's PigProgressNotificationListener. Because PPNL is an interface, this would be backwards incompatible, so Oozie would need a minor change to match the new interface. Please read the motivation and proposal below, and comment on the ticket if you have any concerns. I'd be happy to provide the modified PPNL for oozie that simply ignored the new method.
Dmitriy On Mon, Feb 13, 2012 at 10:43 AM, Alan Gates <[email protected]> wrote: > AFAIK the only user of this interface is Oozie. You might want to shoot a > message to [email protected] and let them know about the > change. > > Alan. > > On Feb 12, 2012, at 5:12 PM, Dmitriy Ryaboy wrote: > > > I created https://issues.apache.org/jira/browse/PIG-2528 to track this > > issue. > > Please let me know early if you have grave concerns, as I am about to > sink > > some hours into it.. > > > > Dmitriy > > > > On Sun, Feb 12, 2012 at 5:01 PM, Dmitriy Ryaboy <[email protected]> > wrote: > > > >> Hi folks, > >> For the purposes of tracing usage and statistics of Pig scripts, I'd > like > >> to be able to correlate the information PPNL can now collect with the > >> script that was actually running. Currently, a script id gets passed > into > >> every PPNL call, but it's just a UUID, so one can't compare it to other > >> executions of the same script. > >> > >> It seems to me like most of what I need would be taken care of by > adding a > >> setScriptPlan(String lp) method to the PPNL interface, and calling it > from > >> ScriptState's setScriptFeatures(LogicalPlan lp). My PPNL could then walk > >> the plan and come up with some representation that would be reusable > >> between different runs of the same script. > >> > >> This would be a backwards incompatible change, as it requires adding a > >> method to an interface. PPNL is labeled as evolving, and I don't believe > >> that it's widely used at the moment. > >> > >> Thoughts / Suggestions / Objections? > >> > >> -Dmitriy > >> > >
