I just noticed that one that one of my splitfile downloads was stuck (no
data is downloaded and each try to fetch the download status using the
browser causes another thread to be stuck), the cause seems to be a
deadlock. The last thing the wab interface said was 'Decoding FEC block,
please be patient' so it would seem that that was the last thing the node
was doing before the deadlock occurred (one of the supplied stacktraces
indicates that blocks are beeing enqueued for background inserting when the
deadlock occurred)

Attached is two full stacktraces showing where the different threads in the
JVM is stuck, one file contains the stacktraces of all threads and the
second file is an extract of the stacktraces of the threads that I think is
involved in the deadlock (and some small notes made by me about aquired
locks for the different threads).

I seem the be unable to find the exact cause of the deadlock. One of the
monitors involved is a monitor on an instance of SplitfileRequestManager,
the exact other monitor is not known but might possibly be the monitor on
the SFRContext associated with the SplitfileRequestManager instance (or are
those two actually the same object.. doesn't seem so but the relevant code
isn't exactly trivial).


/N

Attachment: deadlock.zip
Description: Zip compressed data

Reply via email to