What is row level isolation on cassandra
Download as PPTX, PDF1 like2,444 views
Cassandra provides row-level isolation where a transaction is atomic for a single query executed on one node. A transaction in Cassandra updates either all columns in a row or none of them. Complex multi-row or multi-query transactions are not supported. The implementation uses a copy-on-write approach utilizing a concurrent data structure called SnapTree to clone columns being updated, ensuring atomic and isolated updates at the row level for a single query.
What is row level isolation on cassandra
- 1. What is Row Level Isolation on Cassandra INTHEFOERST Co., Ltd. tomitakazutaka
- 2. What is meaning of Cassandra’s transaction? One query is one transaction ! In Cassandra, a transaction is provided atomicity guarantees in executed one query. Complex transactions isn’t supported. And it is only by one node.
- 3. Ex) CQL) UPDATE CBench SET Tag1 = ‘data1’, Tag2=‘data2’ where KEY = ‘john smith’; Atomic Hector API) cluster = HFactory.getOrCreateCluster("Intheforest Cluster",chc); keyspace = HFactory.createKeyspace("CassandraBenchmark", cluster); mutator = HFactory.createMutator(keyspace, StringSerializer.get()); mutator.addInsertion("venezia-20120613004300-000", "CBench",HFactory.createStringColumn("Tag1", "data1")) .addInsertion("venezia-20120613004300-000", "CBench", HFactory.createStringColumn("Tag2", "data2")) Atomic .addInsertion("venezia-20120613004300-000", "CBench",HFactory.createStringColumn("Tag3", "data3")); mutator.execute();
- 4. Implementation Read is not changed. As long as I investigated .... Perhaps….
- 5. Implementation Write to Commitlog Commitlog is not changed for Isolation. As long as I investigated .... Perhaps….
- 6. Implementation Writing process to MemTable org.apache.cassandra.db.RowMutation.apply() ↓ org.apache.cassandra.db.Table.apply() ↓ org.apache.cassandra.db.ColumnFamilyStore.apply() ↓ org.apache.cassandra.db. Memtable.put() ↓ org.apache.cassandra.db. Memtable. resolve()
- 7. Implementation 1.0.10 org.apache.cassandra.db. Memtable. resolve() private void resolve(DecoratedKey key, ColumnFamily cf) { ~~~~ Omission ~~~~ clonedCf.delete(cf); for (IColumn column : cf.getSortedColumns()) clonedCf.addColumn(column.localCopy(cfs, allocator), allocator); } increment simply for every column.
- 8. Implementation 1.1.1 org.apache.cassandra.db.Memtable.resolve() private void resolve(DecoratedKey key, ColumnFamily cf) { ~~~~ Omission ~~~~ previous.addAll(cf, allocator, localCopyFunction); }
- 9. Implementation org.apache.cassandra.db.AbstractColumnContainer.addAll() org.apache.cassandra.db.AtomicSortedColumns.addAll() public void addAll(ISortedColumns cm, Allocator allocator, Function<IColumn, IColumn> transformation) { Holder current, modified; main_loop: do { current = ref.get(); DeletionInfo newDelInfo = current.deletionInfo; if (newDelInfo.markedForDeleteAt < cm.getDeletionInfo().markedForDeleteAt) newDelInfo = cm.getDeletionInfo(); modified = new Holder(current.map.clone(), newDelInfo); SnapTree for (IColumn column : cm.getSortedColumns()) { (copy-on-write clone facilities) modified.addColumn(transformation.apply(column), allocator); // bail early if we know we've been beaten if (ref.get() != current) continue main_loop; CAS } } (See:http://en.wikipedia.org/wiki/Compare-and-swap) while (!ref.compareAndSet(current, modified)); }
- 10. Implementation What’s SnapTree ? SnapTree is a concurrent AVL tree with fast cloning. See) http://ppl.stanford.edu/papers/ppopp207-bronson.pdf
- 11. Conclusion Notice the Cassandra’s transaction about the range. Cassandra's row level isolation is atomic by one query and by one node
- 12. Thank you !