Repository: cassandra Updated Branches: refs/heads/trunk 0f58f6c65 -> b2d20d4bb
Count deleted rows scanned during reads for tracing and warning tombstone thresholds. If a row is read but is not live anymore (which happens with row level tombstones) it is not counted anywhere in the metrics nor reported in tracing. Row tombstones themselves are not reported anywhere. The consequence is that some delete heavy workloads will show no tombstone read but endure severe performance issues. This commit counts deleted rows as standard tombstone cells. Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/419552a9 Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/419552a9 Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/419552a9 Branch: refs/heads/trunk Commit: 419552a93f4e164dfce3cb6fea15328abade5d3f Parents: 0ac5a17 Author: Alexander Dejanovski <a...@thelastpickle.com> Authored: Mon Dec 18 18:27:44 2017 +0100 Committer: Jon Haddad <j...@jonhaddad.com> Committed: Thu Jan 11 11:24:59 2018 -0800 ---------------------------------------------------------------------- .../org/apache/cassandra/db/ReadCommand.java | 26 ++- .../apache/cassandra/metrics/TableMetrics.java | 2 +- .../apache/cassandra/db/ReadCommandTest.java | 171 +++++++++++++++++++ 3 files changed, 193 insertions(+), 6 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/cassandra/blob/419552a9/src/java/org/apache/cassandra/db/ReadCommand.java ---------------------------------------------------------------------- diff --git a/src/java/org/apache/cassandra/db/ReadCommand.java b/src/java/org/apache/cassandra/db/ReadCommand.java index 1d74c16..128f8f3 100644 --- a/src/java/org/apache/cassandra/db/ReadCommand.java +++ b/src/java/org/apache/cassandra/db/ReadCommand.java @@ -433,14 +433,26 @@ public abstract class ReadCommand extends MonitorableImpl implements ReadQuery @Override public Row applyToRow(Row row) { - if (row.hasLiveData(ReadCommand.this.nowInSec(), enforceStrictLiveness)) - ++liveRows; - + boolean hasTombstones = false; for (Cell cell : row.cells()) { if (!cell.isLive(ReadCommand.this.nowInSec())) + { countTombstone(row.clustering()); + hasTombstones = true; // allows to avoid counting an extra tombstone if the whole row expired + } } + + if (row.hasLiveData(ReadCommand.this.nowInSec(), enforceStrictLiveness)) + ++liveRows; + else if (!row.primaryKeyLivenessInfo().isLive(ReadCommand.this.nowInSec()) + && row.hasDeletion(ReadCommand.this.nowInSec()) + && !hasTombstones) + { + // We're counting primary key deletions only here. + countTombstone(row.clustering()); + } + return row; } @@ -474,7 +486,9 @@ public abstract class ReadCommand extends MonitorableImpl implements ReadQuery boolean warnTombstones = tombstones > warningThreshold && respectTombstoneThresholds; if (warnTombstones) { - String msg = String.format("Read %d live rows and %d tombstone cells for query %1.512s (see tombstone_warn_threshold)", liveRows, tombstones, ReadCommand.this.toCQLString()); + String msg = String.format( + "Read %d live rows and %d tombstone cells for query %1.512s (see tombstone_warn_threshold)", + liveRows, tombstones, ReadCommand.this.toCQLString()); ClientWarn.instance.warn(msg); if (tombstones < failureThreshold) { @@ -484,7 +498,9 @@ public abstract class ReadCommand extends MonitorableImpl implements ReadQuery logger.warn(msg); } - Tracing.trace("Read {} live and {} tombstone cells{}", liveRows, tombstones, (warnTombstones ? " (see tombstone_warn_threshold)" : "")); + Tracing.trace("Read {} live rows and {} tombstone cells{}", + liveRows, tombstones, + (warnTombstones ? " (see tombstone_warn_threshold)" : "")); } }; http://git-wip-us.apache.org/repos/asf/cassandra/blob/419552a9/src/java/org/apache/cassandra/metrics/TableMetrics.java ---------------------------------------------------------------------- diff --git a/src/java/org/apache/cassandra/metrics/TableMetrics.java b/src/java/org/apache/cassandra/metrics/TableMetrics.java index 5c4a849..7ce2f16 100644 --- a/src/java/org/apache/cassandra/metrics/TableMetrics.java +++ b/src/java/org/apache/cassandra/metrics/TableMetrics.java @@ -125,7 +125,7 @@ public class TableMetrics public final Gauge<Double> keyCacheHitRate; /** Tombstones scanned in queries on this CF */ public final TableHistogram tombstoneScannedHistogram; - /** Live cells scanned in queries on this CF */ + /** Live rows scanned in queries on this CF */ public final TableHistogram liveScannedHistogram; /** Column update time delta on this CF */ public final TableHistogram colUpdateTimeDeltaHistogram; http://git-wip-us.apache.org/repos/asf/cassandra/blob/419552a9/test/unit/org/apache/cassandra/db/ReadCommandTest.java ---------------------------------------------------------------------- diff --git a/test/unit/org/apache/cassandra/db/ReadCommandTest.java b/test/unit/org/apache/cassandra/db/ReadCommandTest.java index 64ac627..61ce4bd 100644 --- a/test/unit/org/apache/cassandra/db/ReadCommandTest.java +++ b/test/unit/org/apache/cassandra/db/ReadCommandTest.java @@ -50,6 +50,8 @@ import org.apache.cassandra.exceptions.ConfigurationException; import org.apache.cassandra.io.util.DataInputBuffer; import org.apache.cassandra.io.util.DataOutputBuffer; import org.apache.cassandra.io.util.WrappedDataOutputStreamPlus; +import org.apache.cassandra.metrics.TableMetrics; +import org.apache.cassandra.metrics.TableMetrics.TableHistogram; import org.apache.cassandra.net.MessagingService; import org.apache.cassandra.schema.KeyspaceParams; import org.apache.cassandra.schema.TableMetadata; @@ -340,4 +342,173 @@ public class ReadCommandTest count++; } } + + @Test + public void testCountDeletedRows() throws Exception + { + ColumnFamilyStore cfs = Keyspace.open(KEYSPACE).getColumnFamilyStore(CF3); + // cfs = Mockito.spy(cfs); + // TableMetrics metricsMock = Mockito.mock(TableMetrics.class); + // Mockito.doReturn(metricsMock).when(cfs).metric; + + String[][][] groups = new String[][][] { + new String[][] { + new String[] { "1", "key1", "aa", "a" }, // "1" indicates to create the data, "-1" to delete the + // row + new String[] { "1", "key2", "bb", "b" }, + new String[] { "1", "key3", "cc", "c" } + }, + new String[][] { + new String[] { "1", "key3", "dd", "d" }, + new String[] { "1", "key2", "ee", "e" }, + new String[] { "1", "key1", "ff", "f" } + }, + new String[][] { + new String[] { "1", "key6", "aa", "a" }, + new String[] { "1", "key5", "bb", "b" }, + new String[] { "1", "key4", "cc", "c" } + }, + new String[][] { + new String[] { "1", "key2", "aa", "a" }, + new String[] { "1", "key2", "cc", "c" }, + new String[] { "1", "key2", "dd", "d" } + }, + new String[][] { + new String[] { "-1", "key6", "aa", "a" }, + new String[] { "-1", "key2", "bb", "b" }, + new String[] { "-1", "key2", "ee", "e" }, + new String[] { "-1", "key2", "aa", "a" }, + new String[] { "-1", "key2", "cc", "c" }, + new String[] { "-1", "key2", "dd", "d" } + } + }; + + List<ByteBuffer> buffers = new ArrayList<>(groups.length); + int nowInSeconds = FBUtilities.nowInSeconds(); + ColumnFilter columnFilter = ColumnFilter.allRegularColumnsBuilder(cfs.metadata()).build(); + RowFilter rowFilter = RowFilter.create(); + Slice slice = Slice.make(ClusteringBound.BOTTOM, ClusteringBound.TOP); + ClusteringIndexSliceFilter sliceFilter = new ClusteringIndexSliceFilter( + Slices.with(cfs.metadata().comparator, slice), false); + + for (String[][] group : groups) + { + cfs.truncateBlocking(); + + List<SinglePartitionReadCommand> commands = new ArrayList<>(group.length); + + for (String[] data : group) + { + if (data[0].equals("1")) + { + new RowUpdateBuilder(cfs.metadata(), 0, ByteBufferUtil.bytes(data[1])) + .clustering(data[2]) + .add(data[3], ByteBufferUtil.bytes("blah")) + .build() + .apply(); + } + else + { + RowUpdateBuilder.deleteRow(cfs.metadata(), FBUtilities.timestampMicros(), + ByteBufferUtil.bytes(data[1]), data[2]).apply(); + } + commands.add(SinglePartitionReadCommand.create(cfs.metadata(), nowInSeconds, columnFilter, rowFilter, + DataLimits.NONE, Util.dk(data[1]), sliceFilter)); + } + + cfs.forceBlockingFlush(); + + ReadQuery query = new SinglePartitionReadCommand.Group(commands, DataLimits.NONE); + + try (ReadExecutionController executionController = query.executionController(); + UnfilteredPartitionIterator iter = query.executeLocally(executionController); + DataOutputBuffer buffer = new DataOutputBuffer()) + { + UnfilteredPartitionIterators.serializerForIntraNode().serialize(iter, + columnFilter, + buffer, + MessagingService.current_version); + buffers.add(buffer.buffer()); + } + } + + assertEquals(5, cfs.metric.tombstoneScannedHistogram.cf.getSnapshot().getMax()); + } + + @Test + public void testCountWithNoDeletedRow() throws Exception + { + ColumnFamilyStore cfs = Keyspace.open(KEYSPACE).getColumnFamilyStore(CF3); + + String[][][] groups = new String[][][] { + new String[][] { + new String[] { "1", "key1", "aa", "a" }, // "1" indicates to create the data, "-1" to delete the + // row + new String[] { "1", "key2", "bb", "b" }, + new String[] { "1", "key3", "cc", "c" } + }, + new String[][] { + new String[] { "1", "key3", "dd", "d" }, + new String[] { "1", "key2", "ee", "e" }, + new String[] { "1", "key1", "ff", "f" } + }, + new String[][] { + new String[] { "1", "key6", "aa", "a" }, + new String[] { "1", "key5", "bb", "b" }, + new String[] { "1", "key4", "cc", "c" } + } + }; + + List<ByteBuffer> buffers = new ArrayList<>(groups.length); + int nowInSeconds = FBUtilities.nowInSeconds(); + ColumnFilter columnFilter = ColumnFilter.allRegularColumnsBuilder(cfs.metadata()).build(); + RowFilter rowFilter = RowFilter.create(); + Slice slice = Slice.make(ClusteringBound.BOTTOM, ClusteringBound.TOP); + ClusteringIndexSliceFilter sliceFilter = new ClusteringIndexSliceFilter( + Slices.with(cfs.metadata().comparator, slice), false); + + for (String[][] group : groups) + { + cfs.truncateBlocking(); + + List<SinglePartitionReadCommand> commands = new ArrayList<>(group.length); + + for (String[] data : group) + { + if (data[0].equals("1")) + { + new RowUpdateBuilder(cfs.metadata(), 0, ByteBufferUtil.bytes(data[1])) + .clustering(data[2]) + .add(data[3], ByteBufferUtil.bytes("blah")) + .build() + .apply(); + } + else + { + RowUpdateBuilder.deleteRow(cfs.metadata(), FBUtilities.timestampMicros(), + ByteBufferUtil.bytes(data[1]), data[2]).apply(); + } + commands.add(SinglePartitionReadCommand.create(cfs.metadata(), nowInSeconds, columnFilter, rowFilter, + DataLimits.NONE, Util.dk(data[1]), sliceFilter)); + } + + cfs.forceBlockingFlush(); + + ReadQuery query = new SinglePartitionReadCommand.Group(commands, DataLimits.NONE); + + try (ReadExecutionController executionController = query.executionController(); + UnfilteredPartitionIterator iter = query.executeLocally(executionController); + DataOutputBuffer buffer = new DataOutputBuffer()) + { + UnfilteredPartitionIterators.serializerForIntraNode().serialize(iter, + columnFilter, + buffer, + MessagingService.current_version); + buffers.add(buffer.buffer()); + } + } + + assertEquals(1, cfs.metric.tombstoneScannedHistogram.cf.getSnapshot().getMax()); + } + } --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org