[ https://issues.apache.org/jira/browse/CASSANDRA-3203?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13104124#comment-13104124 ]
Brandon Williams commented on CASSANDRA-3203: --------------------------------------------- Here is what I've noticed: Things proceed normally for a while: {noformat} INFO 00:50:16,801 Enqueuing flush of Memtable-Standard1@304913075(36443580/45554475 serialized/live bytes, 714580 ops) {noformat} Eventually, something bad happens: {noformat} INFO 00:51:00,001 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='Counter1') (estimated 0 bytes) INFO 00:51:00,002 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='Super1') (estimated 0 bytes) INFO 00:51:00,002 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='SuperCounter1') (estimated 0 bytes) INFO 00:51:00,002 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='Standard1') (estimated 88291581 bytes) INFO 00:51:00,004 Enqueuing flush of Memtable-Standard1@741697653(2646900/88291581 serialized/live bytes, 51900 ops) INFO 00:51:00,004 flushing high-traffic column family ColumnFamilyStore(table='system', columnFamily='NodeIdInfo') (estimated 0 bytes) INFO 00:51:00,004 flushing high-traffic column family ColumnFamilyStore(table='system', columnFamily='IndexInfo') (estimated 0 bytes) INFO 00:51:00,004 flushing high-traffic column family ColumnFamilyStore(table='system', columnFamily='LocationInfo') (estimated 87 bytes) INFO 00:51:00,005 Enqueuing flush of Memtable-LocationInfo@1706933590(70/87 serialized/live bytes, 2 ops) INFO 00:51:00,005 flushing high-traffic column family ColumnFamilyStore(table='system', columnFamily='Migrations') (estimated 0 bytes) INFO 00:51:00,006 flushing high-traffic column family ColumnFamilyStore(table='system', columnFamily='HintsColumnFamily') (estimated 0 bytes) INFO 00:51:00,006 flushing high-traffic column family ColumnFamilyStore(table='system', columnFamily='Schema') (estimated 0 bytes) INFO 00:51:00,006 estimated 0 bytes used by all memtables pre-flush INFO 00:51:00,008 flushing ColumnFamilyStore(table='Keyspace1', columnFamily='Standard1') to free up 459320 bytes INFO 00:51:00,009 Enqueuing flush of Memtable-Standard1@456936648(14280/476332 serialized/live bytes, 280 ops) INFO 00:51:00,010 flushing ColumnFamilyStore(table='system', columnFamily='Schema') to free up 0 bytes INFO 00:51:00,010 flushing ColumnFamilyStore(table='system', columnFamily='HintsColumnFamily') to free up 0 bytes INFO 00:51:00,011 flushing ColumnFamilyStore(table='system', columnFamily='Migrations') to free up 0 bytes INFO 00:51:00,011 flushing ColumnFamilyStore(table='system', columnFamily='LocationInfo') to free up 0 bytes INFO 00:51:00,012 flushing ColumnFamilyStore(table='system', columnFamily='IndexInfo') to free up 0 bytes INFO 00:51:00,013 flushing ColumnFamilyStore(table='system', columnFamily='NodeIdInfo') to free up 0 bytes INFO 00:51:00,013 flushing ColumnFamilyStore(table='Keyspace1', columnFamily='SuperCounter1') to free up 0 bytes INFO 00:51:00,014 flushing ColumnFamilyStore(table='Keyspace1', columnFamily='Super1') to free up 0 bytes INFO 00:51:00,014 flushing ColumnFamilyStore(table='Keyspace1', columnFamily='Counter1') to free up 0 bytes INFO 00:51:01,016 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='Counter1') (estimated 0 bytes) INFO 00:51:01,016 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='Super1') (estimated 0 bytes) INFO 00:51:01,016 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='SuperCounter1') (estimated 0 bytes) INFO 00:51:01,016 flushing high-traffic column family ColumnFamilyStore(table='Keyspace1', columnFamily='Standard1') (estimated 83808954 bytes) INFO 00:51:01,017 Enqueuing flush of Memtable-Standard1@911886300(2512770/83817460 serialized/live bytes, 49270 ops) {noformat} After this point, it stays bad: {noformat} INFO 00:51:01,879 Enqueuing flush of Memtable-Standard1@937412989(2670870/96531021 serialized/live bytes, 52370 ops) INFO 00:51:02,029 Enqueuing flush of Memtable-Standard1@282282499(2668065/88997573 serialized/live bytes, 52315 ops) {noformat} > Odd flush behavior > ------------------ > > Key: CASSANDRA-3203 > URL: https://issues.apache.org/jira/browse/CASSANDRA-3203 > Project: Cassandra > Issue Type: Bug > Components: Core > Affects Versions: 1.0.0 > Reporter: Brandon Williams > Priority: Critical > Fix For: 1.0.0 > > > Given the same workload against 0.8, trunk is creating more than twice the > amount of sstables. Even though a uniform stress workload is being > generated, flush size degrades quickly: > {noformat} > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:22,878 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@2058235391(7741 > 035/110172631 serialized/live bytes, 151785 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:24,888 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@1520390052(3887 > 220/72403158 serialized/live bytes, 76220 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:26,890 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@1868496516(4097 > 085/76255481 serialized/live bytes, 80335 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:28,893 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@498232521(43513 > 20/80922269 serialized/live bytes, 85320 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:29,895 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@1592308290(2310 > 810/44514839 serialized/live bytes, 45310 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:30,897 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@775439677(22684 > 80/64984390 serialized/live bytes, 44480 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:31,899 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@928217914(26741 > 85/76231422 serialized/live bytes, 52435 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:32,901 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@158103119(27511 > 95/77317732 serialized/live bytes, 53945 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:33,903 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@2035169258(3132 > 420/88934701 serialized/live bytes, 61420 ops) > INFO [NonPeriodicTasks:1] 2011-09-09 18:24:34,905 ColumnFamilyStore.java > (line 658) Enqueuing flush of Memtable-Standard1@1097314626(2979 > 675/83651699 serialized/live bytes, 58425 ops) > {noformat} > The serialized to live size ratio appears completely out of whack. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira