The problem it is that it is only noted the chain of functions, not the exact function. Furthermore, the same code run several time on a smaller dataset (1M records), but fails on this one (9.2M records). Curiously, the CPUs of the cluster were saturated right before the crush (see picture attached).
saluti, Stefano [image: Immagine in linea 2] 2014-11-25 11:15 GMT+01:00 Flavio Pompermaier <[email protected]>: > No..they report just a null pointer in a taskmanager during a certain > function > > On Tue, Nov 25, 2014 at 10:59 AM, Stephan Ewen <[email protected]> wrote: > >> The TaskManager log or the JobManager log or the JobManager WebFrontend >> show nothing? >> >> On Tue, Nov 25, 2014 at 10:30 AM, Flavio Pompermaier < >> [email protected]> wrote: >> >>> The problem is that RuntimeExceptions are not catched by user code so >>> (IMO) Flink should log somewhere all the stacktrace of those exceptions.. >>> At the moment I see only that during a step there was a >>> NullPointerException but I don't know nothing more (no class or line or >>> stacktrace) :( >>> >>> On Tue, Nov 25, 2014 at 10:24 AM, Stephan Ewen <[email protected]> wrote: >>> >>>> Your functions can alway print to sysout or syserr (for example on >>>> exceptions). It will be redirected to the *.out files in the Flink log >>>> directory. >>>> >>>> The TaskManagers themselves should log the exceptions that cause tasks >>>> to fail. >>>> >>>> The JobManager should also display exceptions that cause failure in the >>>> web frontend. >>>> >>>> This is not specific to NullPointerExceptions, however. >>>> >>>> Does this help? >>>> Am 25.11.2014 09:24 schrieb "Flavio Pompermaier" <[email protected] >>>> >: >>>> >>>> Hy to all, >>>>> >>>>> is there a way to print/log all the stacktrace of >>>>> NullPointerExceptions during the execution of a function in Flink..? >>>>> >>>>> Best, >>>>> Flavio >>>>> >>>> >>> >> >
