The data probably came from either hints or commitlog replay. If you use `truncate` from CQL, it solves both of those concerns.
On Tue, Aug 11, 2020 at 8:42 AM Kunal <kunal.v...@gmail.com> wrote: > HI, > > We have a 3 nodes cassandra cluster and one of the table grew big, around > 2 gb while it was supposed to be few MBs. During nodetool repair, one of > the cassandra went down. Even after multiple restart, one of the node was > going down after coming up for few mins. We decided to truncate the table > by removing the corresponding sstable from the disk since truncating a > table from cqlsh needs all the nodes to be up which was not the case in our > env. After deleting sstable from disk on all the 3 nodes, we brought up > cassandra and all the nodes came up fine and dont see any issue , but we > observed the size of the sstable is~100MB which was bit strange and the > table has old rows (around 20K) from previous date, before removing the > rows were 500K. Not sure how the table has old records and sstable is of > ~100M even after removing the sstable. > Any ideas ? Any help to understand this would be appreciated. > > Regards, > Kunal >