Interesting. Does it simplify further to RF=1 and 2 nodes? On Wed, Nov 10, 2010 at 8:58 AM, Eric van Orsouw <eric.van.ors...@eventis.nl> wrote: > No, all nodes were up and running while the single key was inserted. > The insert however was with consistency One. I assume however that the > replicas are still written in this case. > It is btw also very reproducible. > > -----Original Message----- > From: Jonathan Ellis [mailto:jbel...@gmail.com] > Sent: woensdag 10 november 2010 15:44 > To: user > Subject: Re: iteration does not yield all data with consistency ONE > > Was the node that should have the other replica of this row down when > it was inserted? > > On Wed, Nov 10, 2010 at 6:08 AM, Eric van Orsouw > <eric.van.ors...@eventis.nl> wrote: >> >> Hello, >> >> >> >> We have a cluster of 4 nodes (0.6.6) and use the random partitioner and a >> replication of 2. >> >> When I insert a number of rows I can always retrieve them by their explicit >> id (get_range_slices("<key>","", 1). >> >> Playing with consistency levels and temporarily shutting down a Cassandra >> node all yields the expected result. >> >> >> >> However when I use get_range_slices("","", n) to iterate over all rows, I >> sometimes don't get anything (depending on the node). >> >> >> >> I then reduced the problem to inserting just a single row. >> >> Specifically, the 'iteration' only seems to succeed when I issue the request >> to the node that contains the first copy. >> >> I Discovered that when I iterate using a consistency level of Quorum/All the >> iteration always succeeds and I properly get the one row. >> >> >> >> So a solution would be to always use consistency level One/All but that has >> a performance penalty. >> >> >> >> Can anyone explain why iterating using get_range_slices("","",n) does not >> always function with consistency level One on all nodes? >> >> >> >> Thanks, >> >> Eric >> >> >> >> P.S. To rule out any discussion on whether or not to use iteration in the >> first place, we only plan to use it for backup and periodic cleanup cycles. > > > -- > Jonathan Ellis > Project Chair, Apache Cassandra > co-founder of Riptano, the source for professional Cassandra support > http://riptano.com >
-- Jonathan Ellis Project Chair, Apache Cassandra co-founder of Riptano, the source for professional Cassandra support http://riptano.com
