Search in sources :

Example 1 with FactTable

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

the class FactTableTest method testPreSplits.

@Test
public void testPreSplits() throws Exception {
    InMemoryTableService.create("presplitEntityTable");
    InMemoryTableService.create("presplitDataTable");
    int resolution = 10;
    int rollTimebaseInterval = 2;
    InMemoryMetricsTable metricsTable = new InMemoryMetricsTable("presplitDataTable");
    FactTable table = new FactTable(metricsTable, new EntityTable(new InMemoryMetricsTable("presplitEntityTable")), resolution, rollTimebaseInterval);
    byte[][] splits = FactTable.getSplits(3);
    long ts = System.currentTimeMillis() / 1000;
    DimensionValue dimVal1 = new DimensionValue("dim1", "value1");
    DimensionValue dimVal2 = new DimensionValue("dim2", "value2");
    DimensionValue dimVal3 = new DimensionValue("dim3", "value3");
    // first agg view: dim1
    table.add(ImmutableList.of(new Fact(ts, ImmutableList.of(dimVal1), new Measurement("metric1", MeasureType.COUNTER, 1))));
    // second agg view: dim1 & dim2
    table.add(ImmutableList.of(new Fact(ts, ImmutableList.of(dimVal1, dimVal2), new Measurement("metric1", MeasureType.COUNTER, 1))));
    // third agg view: dim3
    table.add(ImmutableList.of(new Fact(ts, ImmutableList.of(dimVal3), new Measurement("metric1", MeasureType.COUNTER, 1))));
    // Verify all written records are spread across splits
    Scanner scanner = metricsTable.scan(null, null, null);
    Row row;
    Set<Integer> splitsWithRows = Sets.newHashSet();
    while ((row = scanner.next()) != null) {
        boolean added = false;
        for (int i = 0; i < splits.length; i++) {
            if (Bytes.compareTo(row.getRow(), splits[i]) < 0) {
                splitsWithRows.add(i);
                added = true;
                break;
            }
        }
        if (!added) {
            // falls into last split
            splitsWithRows.add(splits.length);
        }
    }
    Assert.assertEquals(3, splitsWithRows.size());
}
Also used : Measurement(co.cask.cdap.api.dataset.lib.cube.Measurement) Scanner(co.cask.cdap.api.dataset.table.Scanner) DimensionValue(co.cask.cdap.api.dataset.lib.cube.DimensionValue) InMemoryMetricsTable(co.cask.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable) Row(co.cask.cdap.api.dataset.table.Row) Test(org.junit.Test)

Example 2 with FactTable

use of io.cdap.cdap.data2.dataset2.lib.timeseries.FactTable 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 3 with FactTable

use of io.cdap.cdap.data2.dataset2.lib.timeseries.FactTable 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)

Example 4 with FactTable

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

the class DefaultCube method query.

@Override
public Collection<TimeSeries> query(CubeQuery query) {
    /*
      CubeQuery example: "dataset read ops for app per dataset". Or:

      SELECT count('read.ops')                                           << measure name and type
      FROM aggregation1.1min_resolution                                  << aggregation and resolution
      GROUP BY dataset,                                                  << groupByDimensions
      WHERE namespace='ns1' AND app='myApp' AND program='myFlow' AND     << dimensionValues
            ts>=1423370200 AND ts{@literal<}1423398198                   << startTs and endTs
      LIMIT 100                                                          << limit

      Execution:

      1) (optional, if aggregation to query in is not provided) find aggregation to supply results

      Here, we need aggregation that has following dimensions: 'namespace', 'app', 'program', 'dataset'.

      Ideally (to reduce the scan range), 'dataset' should be in the end, other dimensions as close to the beginning
      as possible, and minimal number of other "unspecified" dimensions.

      Let's say we found aggregation: 'namespace', 'app', 'program', 'instance', 'dataset'

      2) build a scan in the aggregation

      For scan we set "any" into the dimension values that aggregation has but query doesn't define value for:

      'namespace'='ns1', 'app'='myApp', 'program'='myFlow', 'instance'=*, 'dataset'=*

      Plus specified measure & aggregation?:

      'measureName'='read.ops'
      'measureType'='COUNTER'

      3) While scanning build a table: dimension values -> time -> value. Use measureType as values aggregate
         function if needed.
    */
    incrementMetric("cube.query.request.count", 1);
    if (!resolutionToFactTable.containsKey(query.getResolution())) {
        incrementMetric("cube.query.request.failure.count", 1);
        throw new IllegalArgumentException("There's no data aggregated for specified resolution to satisfy the query: " + query.toString());
    }
    // 1) find aggregation to query
    Aggregation agg;
    String aggName;
    if (query.getAggregation() != null) {
        aggName = query.getAggregation();
        agg = aggregations.get(query.getAggregation());
        if (agg == null) {
            incrementMetric("cube.query.request.failure.count", 1);
            throw new IllegalArgumentException(String.format("Specified aggregation %s is not found in cube aggregations: %s", query.getAggregation(), aggregations.keySet().toString()));
        }
    } else {
        ImmutablePair<String, Aggregation> aggregation = findAggregation(query);
        if (aggregation == null) {
            incrementMetric("cube.query.request.failure.count", 1);
            throw new IllegalArgumentException("There's no data aggregated for specified dimensions " + "to satisfy the query: " + query.toString());
        }
        agg = aggregation.getSecond();
        aggName = aggregation.getFirst();
    }
    // tell how many queries end up querying specific pre-aggregated views and resolutions
    incrementMetric("cube.query.agg." + aggName + ".count", 1);
    incrementMetric("cube.query.res." + query.getResolution() + ".count", 1);
    // 2) build a scan for a query
    List<DimensionValue> dimensionValues = Lists.newArrayList();
    for (String dimensionName : agg.getDimensionNames()) {
        // if not defined in query, will be set as null, which means "any"
        dimensionValues.add(new DimensionValue(dimensionName, query.getDimensionValues().get(dimensionName)));
    }
    FactScan scan = new FactScan(query.getStartTs(), query.getEndTs(), query.getMeasurements().keySet(), dimensionValues);
    // 3) execute scan query
    FactTable table = resolutionToFactTable.get(query.getResolution());
    FactScanner scanner = table.scan(scan);
    Table<Map<String, String>, String, Map<Long, Long>> resultMap = getTimeSeries(query, scanner);
    incrementMetric("cube.query.request.success.count", 1);
    incrementMetric("cube.query.result.size", resultMap.size());
    Collection<TimeSeries> timeSeries = convertToQueryResult(query, resultMap);
    incrementMetric("cube.query.result.timeseries.count", timeSeries.size());
    return timeSeries;
}
Also used : FactScan(co.cask.cdap.data2.dataset2.lib.timeseries.FactScan) TimeSeries(co.cask.cdap.api.dataset.lib.cube.TimeSeries) FactScanner(co.cask.cdap.data2.dataset2.lib.timeseries.FactScanner) FactTable(co.cask.cdap.data2.dataset2.lib.timeseries.FactTable) DimensionValue(co.cask.cdap.api.dataset.lib.cube.DimensionValue) LinkedHashMap(java.util.LinkedHashMap) Map(java.util.Map)

Example 5 with FactTable

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

the class DefaultCubeTest method getCube.

@Override
protected Cube getCube(final String name, int[] resolutions, Map<String, ? extends Aggregation> aggregations) {
    FactTableSupplier supplier = new FactTableSupplier() {

        @Override
        public FactTable get(int resolution, int rollTime) {
            String entityTableName = "EntityTable-" + name;
            InMemoryTableService.create(entityTableName);
            String dataTableName = "DataTable-" + name + "-" + resolution;
            InMemoryTableService.create(dataTableName);
            return new FactTable(new InMemoryMetricsTable(dataTableName), new EntityTable(new InMemoryMetricsTable(entityTableName)), resolution, rollTime);
        }
    };
    return new DefaultCube(resolutions, supplier, aggregations, ImmutableMap.<String, AggregationAlias>of());
}
Also used : EntityTable(co.cask.cdap.data2.dataset2.lib.timeseries.EntityTable) FactTable(co.cask.cdap.data2.dataset2.lib.timeseries.FactTable) InMemoryMetricsTable(co.cask.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable)

Aggregations

Test (org.junit.Test)11 DimensionValue (io.cdap.cdap.api.dataset.lib.cube.DimensionValue)9 DimensionValue (co.cask.cdap.api.dataset.lib.cube.DimensionValue)8 FactTable (co.cask.cdap.data2.dataset2.lib.timeseries.FactTable)7 InMemoryMetricsTable (co.cask.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable)6 ImmutableList (com.google.common.collect.ImmutableList)6 InMemoryMetricsTable (io.cdap.cdap.data2.dataset2.lib.table.inmemory.InMemoryMetricsTable)6 FactTable (io.cdap.cdap.data2.dataset2.lib.timeseries.FactTable)6 List (java.util.List)6 ArrayList (java.util.ArrayList)4 LinkedHashMap (java.util.LinkedHashMap)4 Map (java.util.Map)4 Measurement (io.cdap.cdap.api.dataset.lib.cube.Measurement)3 Measurement (co.cask.cdap.api.dataset.lib.cube.Measurement)2 TimeValue (co.cask.cdap.api.dataset.lib.cube.TimeValue)2 FactScan (co.cask.cdap.data2.dataset2.lib.timeseries.FactScan)2 TimeValue (io.cdap.cdap.api.dataset.lib.cube.TimeValue)2 FactScan (io.cdap.cdap.data2.dataset2.lib.timeseries.FactScan)2 HashMap (java.util.HashMap)2 CubeFact (co.cask.cdap.api.dataset.lib.cube.CubeFact)1