Re: java.io.IOException: Failed to close some writers

2018-05-17 Thread Eugene Kirpichov
Hmm I don't see anything obviously wrong with the code, assuming TableReference and TableSchema both implement a hashCode() consistent with equals() [I'm not 100% sure - could you check that?]... I'm hoping saving heap dumps will shed some light. On Thu, May 17, 2018 at 12:25 AM Jose Ignacio

Re: java.io.IOException: Failed to close some writers

2018-05-17 Thread Jose Ignacio Honrado Benítez
Thanks Eugene, I will try "saveHeapDumpsToGcsPath" to see if there are any head dumps that I am missing. I am using DynamicDestinations, indeed. You can find the code here: https://pastebin.com/yrmXFQLH Note: DocumentWithSchema is a case class that contains the document JSON along with its

Re: java.io.IOException: Failed to close some writers

2018-05-14 Thread Eugene Kirpichov
Thanks. This definitely looks like a memory usage issue to me, I recommend you try running with --saveHeapDumpsToGcsPath=gs://... and see if workers leave any heap dumps over there - perhaps there were some OOMs you missed? If there are heap dumps, you can use any regular Java memory profiler on

Re: java.io.IOException: Failed to close some writers

2018-05-13 Thread Jose Ignacio Honrado Benítez
Thanks for your response Eugene. I don't see any more error traces than the shown in the previous mail in the"worker" category, but checking the "jvm-gc" logs category I can see a lot of these traces: I 5498.913: [GC (Allocation Failure) 2018-05-09T17:42:59.661+: 5498.913: [DefNew:

Re: java.io.IOException: Failed to close some writers

2018-05-11 Thread Eugene Kirpichov
Do you see any other exceptions in Stackdriver, e.g. (guess) OOM errors? "worker lost contact with the service" is commonly caused by OOM crashes, and GC thrashing could also cause network issues of the sort that you're seeing. On Fri, May 11, 2018 at 2:54 AM Jose Ignacio Honrado Benítez <