Question/Observation on MapReduce 'global' variables. I used to think you couldn't define a script-global variable and have it survive multiple invocations of

map()

? Years ago I recall it seemed like

map()

was invoked with a new execution context, so that any global variable would be reset. It doesn't seem to be doing that any longer?

I ran into this recently actually, it seems "global" vars persist during the same Stages e.g. you could push data to same shared array between multiple map executions, but could not pass array from Input to Map etc.

wow, that's a big change, even if only within the 'same stage'

typically I would suggest gathering what you can in Input and then passing to each Map etc. to avoid any funky-ness since this doesnt seem to be really documented

looks like each processor has its own global state

Why not use N/cache?

finished, testing, yielding does indeed reset the global state

N/cache is the typical 'global var simulator' but more complex than just assigning a variabl3e

Yeah I don't use globals in MR but they definitely did not work before.