ok, just find it a bit hard to be forced to shutdown the node in case of
an IOException, but I understand why. The exception occurred because of
a missing native snappy library on the server, but the error only occur
because we initialized a column family incorrectly (we are using
cassandra embedded and are working directly against the internal APIs).
Regards
Mikael Wikblom
On 05/02/2012 03:03 PM, Sylvain Lebresne wrote:
On Wed, May 2, 2012 at 2:42 PM, Mikael Wikblom
<mikael.wikb...@sitevision.se> wrote:
Given an IOException in writeSortedContents the latch.countDown() will not
be called. Wouldn't it be better to place the latch.countDown() in the
finally statement?
No because having the latch being countDown means 'the sstable has
been flushed successfully and the data can be safely deleted in the
commit log', which is not the case if you get an IOException.
We've had issues with IOExceptions in writeSortedContents when doing a snapshot
which hung a thread (and still hangs) for 4 days.
It would be interesting to know what triggered the IOException. If
that's due to a bug, then that's the one we should fix in priority. If
that's you running out of disk-space or something like that, you
should probably fix that and restart C*.
--
Sylvain
--
Mikael Wikblom
Software Architect
SiteVision AB
019-217058
mikael.wikb...@sitevision.se
http://www.sitevision.se
