Search in sources :

Example 1 with IndexUpdateManager

use of org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager in project phoenix by apache.

the class NonTxIndexBuilder method getIndexUpdate.

@Override
public Collection<Pair<Mutation, byte[]>> getIndexUpdate(Mutation mutation, IndexMetaData indexMetaData) throws IOException {
    // create a state manager, so we can manage each batch
    LocalTableState state = new LocalTableState(env, localTable, mutation);
    // build the index updates for each group
    IndexUpdateManager manager = new IndexUpdateManager(indexMetaData);
    batchMutationAndAddUpdates(manager, state, mutation, indexMetaData);
    if (LOG.isTraceEnabled()) {
        LOG.trace("Found index updates for Mutation: " + mutation + "\n" + manager);
    }
    return manager.toMap();
}
Also used : IndexUpdateManager(org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager)

Example 2 with IndexUpdateManager

use of org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager in project phoenix by apache.

the class TestIndexUpdateManager method testCancelingUpdates.

/**
   * When making updates we need to cancel out {@link Delete} and {@link Put}s for the same row.
   * @throws Exception on failure
   */
@Test
public void testCancelingUpdates() throws Exception {
    IndexUpdateManager manager = new IndexUpdateManager(mockIndexMetaData);
    long ts1 = 10, ts2 = 11;
    // at different timestamps, so both should be retained
    Delete d = new Delete(row, ts1);
    Put p = new Put(row, ts2);
    manager.addIndexUpdate(table, d);
    manager.addIndexUpdate(table, p);
    List<Mutation> pending = new ArrayList<Mutation>();
    pending.add(p);
    pending.add(d);
    validate(manager, pending);
    // add a delete that should cancel out the put, leading to only one delete remaining
    Delete d2 = new Delete(row, ts2);
    manager.addIndexUpdate(table, d2);
    pending.add(d);
    validate(manager, pending);
    // double-deletes of the same row only retain the existing one, which was already canceled out
    // above
    Delete d3 = new Delete(row, ts2);
    manager.addIndexUpdate(table, d3);
    pending.add(d);
    validate(manager, pending);
    // if there is just a put and a delete at the same ts, no pending updates should be returned
    manager = new IndexUpdateManager(mockIndexMetaData);
    manager.addIndexUpdate(table, d2);
    manager.addIndexUpdate(table, p);
    validate(manager, Collections.<Mutation>emptyList());
    // different row insertions can be tricky too, if you don't get the base cases right
    manager = new IndexUpdateManager(mockIndexMetaData);
    manager.addIndexUpdate(table, p);
    // this row definitely sorts after the current row
    byte[] row1 = Bytes.toBytes("row1");
    Put p1 = new Put(row1, ts1);
    manager.addIndexUpdate(table, p1);
    // this delete should completely cover the given put and both should be removed
    Delete d4 = new Delete(row1, ts1);
    manager.addIndexUpdate(table, d4);
    pending.clear();
    pending.add(p);
    validate(manager, pending);
}
Also used : Delete(org.apache.hadoop.hbase.client.Delete) IndexUpdateManager(org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager) ArrayList(java.util.ArrayList) Mutation(org.apache.hadoop.hbase.client.Mutation) Put(org.apache.hadoop.hbase.client.Put) Test(org.junit.Test)

Example 3 with IndexUpdateManager

use of org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager in project phoenix by apache.

the class TestIndexUpdateManager method testMutationComparator.

@Test
public void testMutationComparator() throws Exception {
    IndexUpdateManager manager = new IndexUpdateManager(mockIndexMetaData);
    Comparator<Mutation> comparator = manager.COMPARATOR;
    Put p = new Put(row, 10);
    // lexigraphically earlier should sort earlier
    Put p1 = new Put(Bytes.toBytes("ro"), 10);
    assertTrue("lexigraphically later sorting first, should be earlier first.", comparator.compare(p, p1) > 0);
    p1 = new Put(Bytes.toBytes("row1"), 10);
    assertTrue("lexigraphically later sorting first, should be earlier first.", comparator.compare(p1, p) > 0);
    // larger ts sorts before smaller, for the same row
    p1 = new Put(row, 11);
    assertTrue("Smaller timestamp sorting first, should be larger first.", comparator.compare(p, p1) > 0);
    // still true, even for deletes
    Delete d = new Delete(row, 11);
    assertTrue("Smaller timestamp sorting first, should be larger first.", comparator.compare(p, d) > 0);
    // for the same row, t1, the delete should sort earlier
    d = new Delete(row, 10);
    assertTrue("Delete doesn't sort before put, for the same row and ts", comparator.compare(p, d) > 0);
    // but for different rows, we still respect the row sorting.
    d = new Delete(Bytes.toBytes("row1"), 10);
    assertTrue("Delete doesn't sort before put, for the same row and ts", comparator.compare(p, d) < 0);
}
Also used : Delete(org.apache.hadoop.hbase.client.Delete) IndexUpdateManager(org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager) Mutation(org.apache.hadoop.hbase.client.Mutation) Put(org.apache.hadoop.hbase.client.Put) Test(org.junit.Test)

Example 4 with IndexUpdateManager

use of org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager in project phoenix by apache.

the class CoveredColumnIndexer method getIndexUpdateForFilteredRows.

@Override
public Collection<Pair<Mutation, byte[]>> getIndexUpdateForFilteredRows(Collection<KeyValue> filtered, IndexMetaData indexMetaData) throws IOException {
    // stores all the return values
    IndexUpdateManager updateMap = new IndexUpdateManager(indexMetaData);
    // batch the updates by row to make life easier and ordered
    Collection<Batch> batches = batchByRow(filtered);
    for (Batch batch : batches) {
        Cell curKV = batch.getKvs().iterator().next();
        Put p = new Put(curKV.getRowArray(), curKV.getRowOffset(), curKV.getRowLength());
        for (Cell kv : batch.getKvs()) {
            // we only need to cleanup Put entries
            byte type = kv.getTypeByte();
            Type t = KeyValue.Type.codeToType(type);
            if (!t.equals(Type.Put)) {
                continue;
            }
            // add the kv independently
            p.add(kv);
        }
        // do the usual thing as for deletes
        Collection<Batch> timeBatch = createTimestampBatchesFromMutation(p);
        LocalTableState state = new LocalTableState(env, localTable, p);
        for (Batch entry : timeBatch) {
            //just set the timestamp on the table - it already has all the future state
            state.setCurrentTimestamp(entry.getTimestamp());
            this.addDeleteUpdatesToMap(updateMap, state, entry.getTimestamp(), indexMetaData);
        }
    }
    return updateMap.toMap();
}
Also used : Type(org.apache.hadoop.hbase.KeyValue.Type) IndexUpdateManager(org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager) Batch(org.apache.phoenix.hbase.index.covered.Batch) Cell(org.apache.hadoop.hbase.Cell) Put(org.apache.hadoop.hbase.client.Put) LocalTableState(org.apache.phoenix.hbase.index.covered.LocalTableState)

Aggregations

IndexUpdateManager (org.apache.phoenix.hbase.index.covered.update.IndexUpdateManager)4 Put (org.apache.hadoop.hbase.client.Put)3 Delete (org.apache.hadoop.hbase.client.Delete)2 Mutation (org.apache.hadoop.hbase.client.Mutation)2 Test (org.junit.Test)2 ArrayList (java.util.ArrayList)1 Cell (org.apache.hadoop.hbase.Cell)1 Type (org.apache.hadoop.hbase.KeyValue.Type)1 Batch (org.apache.phoenix.hbase.index.covered.Batch)1 LocalTableState (org.apache.phoenix.hbase.index.covered.LocalTableState)1