[ https://issues.apache.org/jira/browse/CASSANDRA-1183?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jonathan Ellis reassigned CASSANDRA-1183: ----------------------------------------- Assignee: (was: Brandon Williams) > Column insert via batch_mutate under some conditions sometimes fails to add > the column > -------------------------------------------------------------------------------------- > > Key: CASSANDRA-1183 > URL: https://issues.apache.org/jira/browse/CASSANDRA-1183 > Project: Cassandra > Issue Type: Bug > Components: Core > Affects Versions: 0.6.1, 0.6.2 > Environment: OS (via uname -a): Linux 2.6.27.5-117.fc10.x86_64 #1 > SMP Tue Nov 18 11:58:53 EST 2008 x86_64 x86_64 x86_64 GNU/Linux > Java build 1.6.0_20-b02 > Cassandra 0.6.2 & 0.6.1 > hector-0.6.0-14.jar > Reporter: Scott McCarty > Attachments: BatchMutateBug.java > > > Under heavy load batch_mutate sometimes fails to add a column but doesn't > report an error to the client. Details: > * client is running a tight loop that does: write to randomly named column; > do a get_slice that should return that newly created column; delete the > column if it was returned in get_slice > * the mutation map for batch_insert has exactly one mutation in it: insert > the column > * if the column insert is done with the insert() API it doesn't ever fail > (for as long as I've run it) > * the column really isn't inserted as doing a 'get' on the entire rowkey in > the command line interface doesn't show it, so it's not a timing issue > * read consistency level is ONE, write consistency level is ALL > * it fails on a single node cluster and on a 4 node cluster (replication > factor = 3) > I am attaching a Java cmdline program that demonstrates the problem; it uses > the 'hector' java client (I hope that's okay--I verified that no errors or > warning were being masked by hector). > To show the failure using the program, just run it with the command line: > bash> java BatchMutateBug localhost:9160 true keyspace columnfamily rowkey > substituting in the appropriate Cassandra server connection info, a valid > keyspace name, and a valid column family name. The second parameter ("true" > in the example) tells the program to use batch_mutate(). If you pass in > "false" it will use insert() and the program won't fail. > I've tried the above on two different setups and each one fails within 10-15 > seconds generally. It seems to fail quickest if you use a non-existing > rowkey. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.