[ https://issues.apache.org/jira/browse/CASSANDRA-2279?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Joaquin Casares updated CASSANDRA-2279: --------------------------------------- Attachment: nodeB.txt nodeA.txt Here are all the commands that were run. I was messing around with T2 references and the bug didn't show up the first time, so I tried it again. The second time it worked following the exact steps that I listed. I set the GCSeconds to be 30 minutes to wait for the repair to finish 100%. This time however, the SSTables actually have the old values as well. > Tombstones not collected post-repair > ------------------------------------ > > Key: CASSANDRA-2279 > URL: https://issues.apache.org/jira/browse/CASSANDRA-2279 > Project: Cassandra > Issue Type: Bug > Components: Tools > Affects Versions: 0.6 > Reporter: Joaquin Casares > Assignee: Sylvain Lebresne > Priority: Minor > Fix For: 0.6.13, 0.7.5 > > Attachments: RowIteration-unit-tests.patch, > fix-RowIteratorFactory.patch, nodeA.txt, nodeB.txt > > > The keys would only show up in sstables2json and look like this: > (root@aps4):/opt/cassandra/storage/queue/data/Panama Wed Feb 23 07:24:34am > ===> /opt/cassandra/bin/sstable2json Queue-2527-Data.db -k > waq:publicMessageIndexingWorkArea:PUM8a65ce95-9d35-4941-928c-dd5965e8b29b > 2011-02-23 07:24:43,710 INFO [org.apache.cassandra.config.DatabaseDescriptor] > - DiskAccessMode 'auto' determined to be mmap, indexAccessMode is mmap > 2011-02-23 07:24:43,972 INFO [org.apache.cassandra.io.SSTableReader] - > Opening /opt/cassandra/storage/queue/data/Panama/Queue-2527-Data.db > { > "waq:publicMessageIndexingWorkArea:PUM8a65ce95-9d35-4941-928c-dd5965e8b29b": > [] > } > (root@aps4):/opt/cassandra/storage/queue/data/Panama Wed Feb 23 07:24:44am > ===> > The steps that I took to reproduce it were: > Create a keyspace, column family, and a key > Delete the key on Node 1 using the cli (del cf['key'];) > Flush > Repair on a cluster with more than 1 node > Wait GCSeconds > Compact > And the empty row would appear on Node 2 > However, when I was able to get rid of the empty rows, I was following these > steps on a single machine: > Create a keyspace, column family, and a key > Delete the key > Flush > Sample write (writing to some temporary key) > Deleting the attribute to that temporary key (not the entire key) > Flush > Compact > or these steps: > Create a keyspace, column family, and a key > Delete the key > Flush > Wait GCseconds > Compact -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira