https://issues.apache.org/jira/browse/CASSANDRA-2158, fixed in 0.7.3
you could have saved a lot of time just by upgrading first. :) On Fri, Mar 11, 2011 at 2:02 PM, Erik Forkalsrud <eforkals...@cj.com> wrote: > On 03/11/2011 04:56 AM, Zhu Han wrote: >> >> When I run it on my laptop (Fedora 14, 64-bit, 4 cores, 8GB RAM) it >> flushes one Memtable with 5000 operations >> When I run it on a server (RHEL5, 64-bit, 16 cores, 96GB RAM) it flushes >> 100 Memtables with anywhere between 1 operation and 359 operations (35 bytes >> and 12499 bytes) > > What's the settings of commit log flush, periodic or in batch? > > > It's whatever the default setting is, (in the cassandra.yaml that is > packaged in the apache-cassandra-0.7.3-bin.tar.gz download) specifically: > > commitlog_rotation_threshold_in_mb: 128 > commitlog_sync: periodic > commitlog_sync_period_in_ms: 10000 > flush_largest_memtables_at: 0.75 > > If I describe keyspace I get: > > [default@unknown] describe keyspace Events; > Keyspace: Events: > Replication Strategy: org.apache.cassandra.locator.SimpleStrategy > Replication Factor: 1 > Column Families: > ColumnFamily: Event > Columns sorted by: org.apache.cassandra.db.marshal.TimeUUIDType > Row cache size / save period: 0.0/0 > Key cache size / save period: 200000.0/14400 > Memtable thresholds: 14.109375/3010/1440 > GC grace seconds: 864000 > Compaction min/max thresholds: 4/32 > Read repair chance: 1.0 > Built indexes: [] > > > It turns out my suspicion was right. When I tried overriding the jvm memory > parameters calculated in conf/cassandra-env.sh to use the values calculated > on my 8GB laptop like this: > > MAX_HEAP_SIZE=3932m HEAP_NEWSIZE=400m ./mutate.sh > > That made the server behave much nicer. This time it kept all 5000 > operations in a single Memtable. Also, when running with these memory > settings the Memtable thresholds changed to "1.1390625/243/1440" (from > "14.109375/3010/1440") (all the other output from "describe keyspace" > remains the same) > > So it looks like something goes wrong when cassandra gets too much memory. > > > -- > Erik Forkalsrud > Commission Junstion > > -- Jonathan Ellis Project Chair, Apache Cassandra co-founder of DataStax, the source for professional Cassandra support http://www.datastax.com
