If you do not have JNA truncate has to fork an 'ln -s'' command for the snapshots. I think that makes it un-predicatable. Truncate has its own timeout value now (separate from the other timeouts). If possible I think it is better to make each test use it's own CF and avoid truncate entirely.
On Thu, Apr 11, 2013 at 9:48 AM, Ondřej Černoš <cern...@gmail.com> wrote: > Hi, > > I use C* 1.2.3 and CQL3. > > I integrated cassandra into our testing environment. In order to make the > tests repeatable I truncate all the tables that need to be empty before the > test run via ssh session to the host cassandra runs on and by running cqlsh > where I issue the truncate. > > It works, only sometimes it silently fails (1 in 400 runs of the truncate, > actually). > > At the same time the truncate fails I see system ks compaction. > Additionally, it seems there is quite a lot of these system ks compactions > (the numbers in the filenames go up pretty fast to thousands). > > I googled truncate and found out there were some issues with race > conditions and with slowing down if truncate is used frequently (as is my > case, where truncate is run before each test in quite a big test suite). > > Any hints? > > Regards, > Ondřej Černoš >