In a 3 node cassandra 2.1.16 cluster where, one node has old mutation and
two nodes have evict-able (crossed gc grace period) tombstone produced by
TTL. A read range query with local quorum return the old mutation as
result. However expected result should be empty. Next time running the same
query results no data as expected. Why this strange behaviour?
*Steps to Reproduce :*
Create a cassandra-2.1.16 3 node cluster. Disable hinted handoff for each
node.
#ccm node1 nodetool ring
Datacenter: datacenter1
==========
Address Rack Status State Load Owns
Token
3074457345618258602
127.0.0.1 rack1 Up Normal 175.12 KB 100.00%
-9223372036854775808
127.0.0.2 rack1 Up Normal 177.87 KB 100.00%
-3074457345618258603
127.0.0.3 rack1 Up Normal 175.13 KB 100.00%
3074457345618258602
#Connect to cqlsh and set CONISISTENCY LOCAL_QUORUM;
cqlsh> CREATE KEYSPACE IF NOT EXISTS test WITH REPLICATION = { 'class' :
'NetworkTopologyStrategy', 'datacenter1' : 3 };
cqlsh> CREATE TABLE test.table1 (key text, col text, val text,PRIMARY KEY
((key), col));
cqlsh> ALTER TABLE test.table1 with GC_GRACE_SECONDS = 120;
cqlsh> INSERT INTO test.table1 (key, col, val) VALUES ('key2',
'abc','xyz');
#ccm flush
#ccm node3 stop
cqlsh> INSERT INTO test.table1 (key, col, val) VALUES ('key2',
'abc','xyz') USING TTL 60;
#ccm flush
#wait for 3 min so that the tombstone crosses its gc grace period.
#ccm node3 start
cqlsh> select * from test.table1 where token (key) > 3074457345618258602
and token (key) < -9223372036854775808 ;
key | col | val
------+-----+-----
key2 | abc | xyz
(1 rows)
#ccm flush
-> Here read repair triggers and the old mutation moves to the one of the
node where tombstone is present (not both the node)
cqlsh> select * from test.vouchers where token (key) > 3074457345618258602
and token (key) < -9223372036854775808 ;
key | col | val
-----+-----+-----
(0 rows)
--
regards,
Laxmikant Upadhyay
