Search in sources :

Example 1 with FactScan

use of io.cdap.cdap.data2.dataset2.lib.timeseries.FactScan in project cdap by caskdata.

the class FactTable method getScanner.

private Scanner getScanner(FactScan scan) {
    // sort the measures based on their entity ids and based on that get the start and end row key metric names
    List<String> measureNames = getSortedMeasures(scan.getMeasureNames());
    byte[] startRow = codec.createStartRowKey(scan.getDimensionValues(), measureNames.isEmpty() ? null : measureNames.get(0), scan.getStartTs(), false);
    byte[] endRow = codec.createEndRowKey(scan.getDimensionValues(), measureNames.isEmpty() ? null : measureNames.get(measureNames.size() - 1), scan.getEndTs(), false);
    byte[][] columns;
    if (Arrays.equals(startRow, endRow)) {
        // If on the same timebase, we only need subset of columns
        long timeBase = scan.getStartTs() / rollTime * rollTime;
        int startCol = (int) (scan.getStartTs() - timeBase) / resolution;
        int endCol = (int) (scan.getEndTs() - timeBase) / resolution;
        columns = new byte[endCol - startCol + 1][];
        for (int i = 0; i < columns.length; i++) {
            columns[i] = Bytes.toBytes((short) (startCol + i));
        }
    }
    endRow = Bytes.stopKeyForPrefix(endRow);
    FuzzyRowFilter fuzzyRowFilter = measureNames.isEmpty() ? createFuzzyRowFilter(scan, startRow) : createFuzzyRowFilter(scan, measureNames);
    if (LOG.isTraceEnabled()) {
        LOG.trace("Scanning fact table {} with scan: {}; constructed startRow: {}, endRow: {}, fuzzyRowFilter: {}", timeSeriesTable, scan, toPrettyLog(startRow), toPrettyLog(endRow), fuzzyRowFilter);
    }
    return timeSeriesTable.scan(startRow, endRow, fuzzyRowFilter);
}
Also used : FuzzyRowFilter(co.cask.cdap.data2.dataset2.lib.table.FuzzyRowFilter)

Example 2 with FactScan

use of io.cdap.cdap.data2.dataset2.lib.timeseries.FactScan in project cdap by caskdata.

the class FactTable method findMeasureNames.

/**
 * Finds all measure names of the facts that match given {@link DimensionValue}s and time range.
 * @param allDimensionNames list of all dimension names to be present in the fact record
 * @param dimensionSlice dimension values to filter by, {@code null} means any non-null value.
 * @param startTs start timestamp, in sec
 * @param endTs end timestamp, in sec
 * @return {@link Set} of measure names
 */
// todo: pass a limit on number of measures returned
public Set<String> findMeasureNames(List<String> allDimensionNames, Map<String, String> dimensionSlice, long startTs, long endTs) {
    List<DimensionValue> allDimensions = Lists.newArrayList();
    for (String dimensionName : allDimensionNames) {
        allDimensions.add(new DimensionValue(dimensionName, dimensionSlice.get(dimensionName)));
    }
    byte[] startRow = codec.createStartRowKey(allDimensions, null, startTs, false);
    byte[] endRow = codec.createEndRowKey(allDimensions, null, endTs, false);
    endRow = Bytes.stopKeyForPrefix(endRow);
    FuzzyRowFilter fuzzyRowFilter = createFuzzyRowFilter(new FactScan(startTs, endTs, ImmutableList.<String>of(), allDimensions), startRow);
    Set<String> measureNames = Sets.newHashSet();
    int scannedRecords = 0;
    try (Scanner scanner = timeSeriesTable.scan(startRow, endRow, fuzzyRowFilter)) {
        Row rowResult;
        while ((rowResult = scanner.next()) != null) {
            scannedRecords++;
            if (scannedRecords > MAX_RECORDS_TO_SCAN_DURING_SEARCH) {
                break;
            }
            byte[] rowKey = rowResult.getRow();
            // filter out columns by time range (scan configuration only filters whole rows)
            if (codec.getTimestamp(rowKey, codec.createColumn(startTs)) < startTs) {
                continue;
            }
            if (codec.getTimestamp(rowKey, codec.createColumn(endTs)) > endTs) {
                // we're done with scanner
                break;
            }
            measureNames.add(codec.getMeasureName(rowResult.getRow()));
        }
    }
    LOG.trace("search for measures completed, scanned records: {}", scannedRecords);
    return measureNames;
}
Also used : Scanner(co.cask.cdap.api.dataset.table.Scanner) DimensionValue(co.cask.cdap.api.dataset.lib.cube.DimensionValue) Row(co.cask.cdap.api.dataset.table.Row) FuzzyRowFilter(co.cask.cdap.data2.dataset2.lib.table.FuzzyRowFilter)

Example 3 with FactScan

use of io.cdap.cdap.data2.dataset2.lib.timeseries.FactScan in project cdap by caskdata.

the class FactTable method findSingleDimensionValue.

/**
 * Searches for first non-null valued dimensions in records that contain given list of dimensions and match given
 * dimension values in given time range. Returned dimension values are those that are not defined in given
 * dimension values.
 * @param allDimensionNames list of all dimension names to be present in the record
 * @param dimensionSlice dimension values to filter by, {@code null} means any non-null value.
 * @param startTs start of the time range, in seconds
 * @param endTs end of the time range, in seconds
 * @return {@link Set} of {@link DimensionValue}s
 */
// todo: pass a limit on number of dimensionValues returned
// todo: kinda not cool API when we expect null values in a map...
public Set<DimensionValue> findSingleDimensionValue(List<String> allDimensionNames, Map<String, String> dimensionSlice, long startTs, long endTs) {
    // Algorithm, briefly:
    // We scan in the records which have given allDimensionNames. We use dimensionSlice as a criteria for scan.
    // If record from the scan has non-null values in the dimensions which are not specified in dimensionSlice,
    // we use first of such dimension as a value to return.
    // When we find value to return, since we only fill a single dimension, we are not interested in drilling down
    // further and instead attempt to fast-forward (jump) to a record that has different value in that dimension.
    // Thus we find all results.
    List<DimensionValue> allDimensions = Lists.newArrayList();
    List<DimensionValue> filledDimension = Lists.newArrayList();
    List<Integer> dimToFillIndexes = Lists.newArrayList();
    for (int i = 0; i < allDimensionNames.size(); i++) {
        String dimensionName = allDimensionNames.get(i);
        if (!dimensionSlice.containsKey(dimensionName)) {
            dimToFillIndexes.add(i);
            allDimensions.add(new DimensionValue(dimensionName, null));
        } else {
            DimensionValue dimensionValue = new DimensionValue(dimensionName, dimensionSlice.get(dimensionName));
            filledDimension.add(dimensionValue);
            allDimensions.add(dimensionValue);
        }
    }
    // If provided dimensions contain all values filled in, there's nothing to look for
    if (dimToFillIndexes.isEmpty()) {
        return Collections.emptySet();
    }
    Set<DimensionValue> result = Sets.newHashSet();
    int scans = 0;
    int scannedRecords = 0;
    // build a scan
    byte[] startRow = codec.createStartRowKey(allDimensions, null, startTs, false);
    byte[] endRow = codec.createEndRowKey(allDimensions, null, endTs, false);
    endRow = Bytes.stopKeyForPrefix(endRow);
    FuzzyRowFilter fuzzyRowFilter = createFuzzyRowFilter(new FactScan(startTs, endTs, ImmutableList.<String>of(), allDimensions), startRow);
    Scanner scanner = timeSeriesTable.scan(startRow, endRow, fuzzyRowFilter);
    scans++;
    try {
        Row rowResult;
        while ((rowResult = scanner.next()) != null) {
            scannedRecords++;
            // todo: make configurable
            if (scannedRecords > MAX_RECORDS_TO_SCAN_DURING_SEARCH) {
                break;
            }
            byte[] rowKey = rowResult.getRow();
            // filter out columns by time range (scan configuration only filters whole rows)
            if (codec.getTimestamp(rowKey, codec.createColumn(startTs)) < startTs) {
                continue;
            }
            if (codec.getTimestamp(rowKey, codec.createColumn(endTs)) > endTs) {
                // we're done with scanner
                break;
            }
            List<DimensionValue> dimensionValues = codec.getDimensionValues(rowResult.getRow());
            // At this point, we know that the record is in right time range and its dimensions matches given.
            // We try find first non-null valued dimension in the record that was not in given dimensions: we use it to form
            // next drill down suggestion
            int filledIndex = -1;
            for (int index : dimToFillIndexes) {
                // todo: it may be not efficient, if dimensionValues is not array-backed list: i.e. if access by index is
                // not fast
                DimensionValue dimensionValue = dimensionValues.get(index);
                if (dimensionValue.getValue() != null) {
                    result.add(dimensionValue);
                    filledIndex = index;
                    break;
                }
            }
            // todo: fast-forwarding (jumping) should be done on server-side (CDAP-1421)
            if (filledIndex >= 0) {
                scanner.close();
                scanner = null;
                scans++;
                if (scans > MAX_SCANS_DURING_SEARCH) {
                    break;
                }
                startRow = codec.getNextRowKey(rowResult.getRow(), filledIndex);
                scanner = timeSeriesTable.scan(startRow, endRow, fuzzyRowFilter);
            }
        }
    } finally {
        if (scanner != null) {
            scanner.close();
        }
    }
    LOG.trace("search for dimensions completed, scans performed: {}, scanned records: {}", scans, scannedRecords);
    return result;
}
Also used : Scanner(co.cask.cdap.api.dataset.table.Scanner) DimensionValue(co.cask.cdap.api.dataset.lib.cube.DimensionValue) Row(co.cask.cdap.api.dataset.table.Row) FuzzyRowFilter(co.cask.cdap.data2.dataset2.lib.table.FuzzyRowFilter)

Example 4 with FactScan

use of io.cdap.cdap.data2.dataset2.lib.timeseries.FactScan in project cdap by caskdata.

the class FactTableTest method testMaxResolution.

@Test
public void testMaxResolution() throws Exception {
    // we use Integer.MAX_VALUE as resolution to compute all-time total values
    InMemoryTableService.create("TotalsEntityTable");
    InMemoryTableService.create("TotalsDataTable");
    int resolution = Integer.MAX_VALUE;
    // should not matter when resolution is max
    int rollTimebaseInterval = 3600;
    FactTable table = new FactTable(new InMemoryMetricsTable("TotalsDataTable"), new EntityTable(new InMemoryMetricsTable("TotalsEntityTable")), resolution, rollTimebaseInterval);
    // ts is expected in seconds
    long ts = System.currentTimeMillis() / 1000;
    int count = 1000;
    for (int i = 0; i < count; i++) {
        for (int k = 0; k < 10; k++) {
            // shift one day
            writeInc(table, "metric" + k, ts + i * 60 * 60 * 24, i * k, "dim" + k, "value" + k);
        }
    }
    for (int k = 0; k < 10; k++) {
        // 0, 0 should match timestamp of all data points
        FactScan scan = new FactScan(0, 0, "metric" + k, dimValues("dim" + k, "value" + k));
        Table<String, List<DimensionValue>, List<TimeValue>> expected = HashBasedTable.create();
        expected.put("metric" + k, dimValues("dim" + k, "value" + k), ImmutableList.of(new TimeValue(0, k * count * (count - 1) / 2)));
        assertScan(table, expected, scan);
    }
}
Also used : InMemoryMetricsTable(co.cask.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable) List(java.util.List) ImmutableList(com.google.common.collect.ImmutableList) TimeValue(co.cask.cdap.api.dataset.lib.cube.TimeValue) Test(org.junit.Test)

Example 5 with FactScan

use of io.cdap.cdap.data2.dataset2.lib.timeseries.FactScan in project cdap by caskdata.

the class FactTableTest method testBasics.

@Test
public void testBasics() throws Exception {
    InMemoryTableService.create("EntityTable");
    InMemoryTableService.create("DataTable");
    int resolution = 10;
    int rollTimebaseInterval = 2;
    FactTable table = new FactTable(new InMemoryMetricsTable("DataTable"), new EntityTable(new InMemoryMetricsTable("EntityTable")), resolution, rollTimebaseInterval);
    // aligned to start of resolution bucket
    // "/1000" because time is expected to be in seconds
    long ts = ((System.currentTimeMillis() / 1000) / resolution) * resolution;
    // testing encoding with multiple dims
    List<DimensionValue> dimensionValues = ImmutableList.of(new DimensionValue("dim1", "value1"), new DimensionValue("dim2", "value2"), new DimensionValue("dim3", "value3"));
    // trying adding one by one, in same (first) time resolution bucket
    for (int i = 0; i < 5; i++) {
        for (int k = 1; k < 4; k++) {
            // note: "+i" here and below doesn't affect results, just to confirm
            // that data points are rounded to the resolution
            table.add(ImmutableList.of(new Fact(ts + i, dimensionValues, new Measurement("metric" + k, MeasureType.COUNTER, k))));
        }
    }
    // trying adding one by one, in different time resolution buckets
    for (int i = 0; i < 3; i++) {
        for (int k = 1; k < 4; k++) {
            table.add(ImmutableList.of(new Fact(ts + resolution * i + i, dimensionValues, new Measurement("metric" + k, MeasureType.COUNTER, 2 * k))));
        }
    }
    // trying adding as list
    // first incs in same (second) time resolution bucket
    List<Fact> aggs = Lists.newArrayList();
    for (int i = 0; i < 7; i++) {
        for (int k = 1; k < 4; k++) {
            aggs.add(new Fact(ts + resolution, dimensionValues, new Measurement("metric" + k, MeasureType.COUNTER, 3 * k)));
        }
    }
    // then incs in different time resolution buckets
    for (int i = 0; i < 3; i++) {
        for (int k = 1; k < 4; k++) {
            aggs.add(new Fact(ts + resolution * i, dimensionValues, new Measurement("metric" + k, MeasureType.COUNTER, 4 * k)));
        }
    }
    table.add(aggs);
    // verify each metric
    for (int k = 1; k < 4; k++) {
        FactScan scan = new FactScan(ts - 2 * resolution, ts + 3 * resolution, "metric" + k, dimensionValues);
        Table<String, List<DimensionValue>, List<TimeValue>> expected = HashBasedTable.create();
        expected.put("metric" + k, dimensionValues, ImmutableList.of(new TimeValue(ts, 11 * k), new TimeValue(ts + resolution, 27 * k), new TimeValue(ts + 2 * resolution, 6 * k)));
        assertScan(table, expected, scan);
    }
    // verify each metric within a single timeBase
    for (int k = 1; k < 4; k++) {
        FactScan scan = new FactScan(ts, ts + resolution - 1, "metric" + k, dimensionValues);
        Table<String, List<DimensionValue>, List<TimeValue>> expected = HashBasedTable.create();
        expected.put("metric" + k, dimensionValues, ImmutableList.of(new TimeValue(ts, 11 * k)));
        assertScan(table, expected, scan);
    }
    // verify all metrics with fuzzy metric in scan
    Table<String, List<DimensionValue>, List<TimeValue>> expected = HashBasedTable.create();
    for (int k = 1; k < 4; k++) {
        expected.put("metric" + k, dimensionValues, ImmutableList.of(new TimeValue(ts, 11 * k), new TimeValue(ts + resolution, 27 * k), new TimeValue(ts + 2 * resolution, 6 * k)));
    }
    // metric = null means "all"
    FactScan scan = new FactScan(ts - 2 * resolution, ts + 3 * resolution, dimensionValues);
    assertScan(table, expected, scan);
    // delete metric test
    expected.clear();
    // delete the metrics data at (timestamp + 20) resolution
    scan = new FactScan(ts + resolution * 2, ts + resolution * 3, dimensionValues);
    table.delete(scan);
    for (int k = 1; k < 4; k++) {
        expected.put("metric" + k, dimensionValues, ImmutableList.of(new TimeValue(ts, 11 * k), new TimeValue(ts + resolution, 27 * k)));
    }
    // verify deletion
    scan = new FactScan(ts - 2 * resolution, ts + 3 * resolution, dimensionValues);
    assertScan(table, expected, scan);
    // delete metrics for "metric1" at ts0 and verify deletion
    scan = new FactScan(ts, ts + 1, "metric1", dimensionValues);
    table.delete(scan);
    expected.clear();
    expected.put("metric1", dimensionValues, ImmutableList.of(new TimeValue(ts + resolution, 27)));
    scan = new FactScan(ts - 2 * resolution, ts + 3 * resolution, "metric1", dimensionValues);
    assertScan(table, expected, scan);
    // verify the next dims search
    Collection<DimensionValue> nextTags = table.findSingleDimensionValue(ImmutableList.of("dim1", "dim2", "dim3"), ImmutableMap.of("dim1", "value1"), ts, ts + 1);
    Assert.assertEquals(ImmutableSet.of(new DimensionValue("dim2", "value2")), nextTags);
    Map<String, String> slice = Maps.newHashMap();
    slice.put("dim1", null);
    nextTags = table.findSingleDimensionValue(ImmutableList.of("dim1", "dim2", "dim3"), slice, ts, ts + 1);
    Assert.assertEquals(ImmutableSet.of(new DimensionValue("dim2", "value2")), nextTags);
    nextTags = table.findSingleDimensionValue(ImmutableList.of("dim1", "dim2", "dim3"), ImmutableMap.of("dim1", "value1", "dim2", "value2"), ts, ts + 3);
    Assert.assertEquals(ImmutableSet.of(new DimensionValue("dim3", "value3")), nextTags);
    // add new dim values
    dimensionValues = ImmutableList.of(new DimensionValue("dim1", "value1"), new DimensionValue("dim2", "value5"), new DimensionValue("dim3", null));
    table.add(ImmutableList.of(new Fact(ts, dimensionValues, new Measurement("metric", MeasureType.COUNTER, 10))));
    dimensionValues = ImmutableList.of(new DimensionValue("dim1", "value1"), new DimensionValue("dim2", null), new DimensionValue("dim3", "value3"));
    table.add(ImmutableList.of(new Fact(ts, dimensionValues, new Measurement("metric", MeasureType.COUNTER, 10))));
    nextTags = table.findSingleDimensionValue(ImmutableList.of("dim1", "dim2", "dim3"), ImmutableMap.of("dim1", "value1"), ts, ts + 1);
    Assert.assertEquals(ImmutableSet.of(new DimensionValue("dim2", "value2"), new DimensionValue("dim2", "value5"), new DimensionValue("dim3", "value3")), nextTags);
    // search for metric names given dims list and verify
    Collection<String> metricNames = table.findMeasureNames(ImmutableList.of("dim1", "dim2", "dim3"), ImmutableMap.of("dim1", "value1", "dim2", "value2", "dim3", "value3"), ts, ts + 1);
    Assert.assertEquals(ImmutableSet.of("metric2", "metric3"), metricNames);
    metricNames = table.findMeasureNames(ImmutableList.of("dim1", "dim2", "dim3"), ImmutableMap.of("dim1", "value1"), ts, ts + 1);
    Assert.assertEquals(ImmutableSet.of("metric", "metric2", "metric3"), metricNames);
    metricNames = table.findMeasureNames(ImmutableList.of("dim1", "dim2", "dim3"), ImmutableMap.of("dim2", "value2"), ts, ts + 1);
    Assert.assertEquals(ImmutableSet.of("metric2", "metric3"), metricNames);
    metricNames = table.findMeasureNames(ImmutableList.of("dim1", "dim2", "dim3"), slice, ts, ts + 1);
    Assert.assertEquals(ImmutableSet.of("metric", "metric2", "metric3"), metricNames);
}
Also used : Measurement(co.cask.cdap.api.dataset.lib.cube.Measurement) DimensionValue(co.cask.cdap.api.dataset.lib.cube.DimensionValue) InMemoryMetricsTable(co.cask.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable) List(java.util.List) ImmutableList(com.google.common.collect.ImmutableList) TimeValue(co.cask.cdap.api.dataset.lib.cube.TimeValue) Test(org.junit.Test)

Aggregations

ImmutableList (com.google.common.collect.ImmutableList)6 List (java.util.List)6 Test (org.junit.Test)6 DimensionValue (co.cask.cdap.api.dataset.lib.cube.DimensionValue)5 DimensionValue (io.cdap.cdap.api.dataset.lib.cube.DimensionValue)5 FuzzyRowFilter (co.cask.cdap.data2.dataset2.lib.table.FuzzyRowFilter)3 InMemoryMetricsTable (co.cask.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable)3 FuzzyRowFilter (io.cdap.cdap.data2.dataset2.lib.table.FuzzyRowFilter)3 InMemoryMetricsTable (io.cdap.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable)3 ArrayList (java.util.ArrayList)3 TimeValue (co.cask.cdap.api.dataset.lib.cube.TimeValue)2 Row (co.cask.cdap.api.dataset.table.Row)2 Scanner (co.cask.cdap.api.dataset.table.Scanner)2 FactScan (co.cask.cdap.data2.dataset2.lib.timeseries.FactScan)2 FactTable (co.cask.cdap.data2.dataset2.lib.timeseries.FactTable)2 TimeValue (io.cdap.cdap.api.dataset.lib.cube.TimeValue)2 Row (io.cdap.cdap.api.dataset.table.Row)2 Scanner (io.cdap.cdap.api.dataset.table.Scanner)2 FactScan (io.cdap.cdap.data2.dataset2.lib.timeseries.FactScan)2 FactTable (io.cdap.cdap.data2.dataset2.lib.timeseries.FactTable)2