hadoop - Why Spark driver without executors completes an application unexpectedly after 24 hours -


we have spark application submit @ our amazon emr yarn cluster. app uses executors read input , continues job on driver container (executors shutdown within several minutes after app starts). 24 hours , near 5 minutes more after executors shutdown driver unexpectedly starts last job (saveastextfile in our case) while data processed still remain.

(i know app designed incorrectly , doesn't fit spark ideology. , know spark applications should utilize executors calculation on data.)

i can't figure out causes spark or yarn drop data being processed , start last job of app. seems there 24 hours long timeout spark's or yarn's setting somewhere causes behavior described. can't find setting if exists actually.

the final state of application finished , final status succeeded in hadoop cluster web ui. behavior described repeats every submit of app.


Comments

Popular posts from this blog

python Tkinter Capturing keyboard events save as one single string -

android - InAppBilling registering BroadcastReceiver in AndroidManifest -

javascript - Z-index in d3.js -