On 10/13/2015 01:58 PM, Modha, Digant wrote:
I am running Cassandra 2.1.10 and noticed intermittent
DirectoryNotEmptyExceptions during repair. My cassandra data drive
is reiserfs.
Why? I'm genuinely interested in this filesystem selection, since it is
unmaintained, has been dropped from some mainstream linux distributions,
and some may call it "dead". ;)
I noticed that on reiserfs wiki site
https://en.m.wikipedia.org/wiki/ReiserFS#Criticism, it states that
unlink operation is not synchronous. Is that the reason for the
exception below:
ERROR [ValidationExecutor:137] 2015-10-13 00:46:30,759
CassandraDaemon.java:227 - Exception in thread
Thread[ValidationExecutor:137,1,main]
org.apache.cassandra.io.FSWriteError:
java.nio.file.DirectoryNotEmptyException:
at
org.apache.cassandra.io.util.FileUtils.deleteWithConfirm(FileUtils.java:135)
~[apache-cassandra-2.1.10.jar:2.1.10]
<...>
This seems like a reasonable explanation. Using a modern filesystem like
ext4 or xfs would certainly be helpful in getting you within the realm
of a "common" hardware setup.
https://wiki.apache.org/cassandra/CassandraHardware
https://www.safaribooksonline.com/library/view/cassandra-high-performance/9781849515122/ch04s06.html
I think Al Tobey had a slide deck on filesystem tuning for C*, but I
didn't find it quickly.
--
Kind regards,
Michael
