Killing the process is the recommended way. If you want to be 100% sure no data that has been acked is lost, use CommitLogSync=batch. Otherwise, as much as CommitLogSyncPeriodInMS of the most recent writes may be lost (but in a clustered situation, will be on other replicas).
More about this kind of system design (called "crash only," meaning "no graceful shutdown"), if you're curious: http://lwn.net/Articles/191059/ http://dslab.epfl.ch/pubs/crashonly/crashonly.pdf On Wed, Oct 21, 2009 at 9:16 AM, Johannes Schaback <[email protected]> wrote: > Hi, > > I am currently stopping my (single!) Cassandra node by ctrl+c in my > bash console. From what I believe to know about the internals of > Cassandra is the risk of data loss/inconsistency relatively small. Is > that still correct for stopping Cassadra this way? > > Is there a recommended way to shut down cassandra? > > Thanks > > Johannes >
