fixed algos to make tolerance 1%

Author: pabloem

x-pack/plugin/esql/src/internalClusterTest/java/org/elasticsearch/xpack/esql/action/RandomizedTimeSeriesIT.java

@@ -59,7 +59,7 @@ public class RandomizedTimeSeriesIT extends AbstractEsqlIntegTestCase {
     private static final Long TIME_RANGE_SECONDS = 3600L;
     private static final String DATASTREAM_NAME = "tsit_ds";
     private static final Integer SECONDS_IN_WINDOW = 60;
-    private List<XContentBuilder> documents = null;
+    private List<XContentBuilder> documents;
     private TSDataGenerationHelper dataGenerationHelper;
 
     List<List<Object>> consumeRows(EsqlQueryResponse resp) {
@@ -208,7 +208,21 @@ protected Collection<Class<? extends Plugin>> nodePlugins() {
     record RateRange(Double lower, Double upper) implements Comparable<RateRange> {
         @Override
         public int compareTo(RateRange o) {
-            return this.lower.compareTo(o.lower);
+            // Compare first by lower bound, then by upper bound
+            int cmp = this.lower.compareTo(o.lower);
+            if (cmp == 0) {
+                return this.upper.compareTo(o.upper);
+            }
+            return cmp;
+        }
+
+        public int compareToFindingMax(RateRange o) {
+            // Compare first by upper bound, then by lower bound
+            int cmp = this.upper.compareTo(o.upper);
+            if (cmp == 0) {
+                return this.lower.compareTo(o.lower);
+            }
+            return cmp;
         }
     }
 
@@ -248,16 +262,16 @@ static RateStats calculateRateAggregation(
             }
             // TODO: Remove tolerances since we are already allowing a min-max range
             return new RateRange(
-                counterGrowth / secondsInWindow * 0.95, // Add 5% tolerance to the lower bound
-                counterGrowth / (lastTs.toEpochMilli() / 1000 - firstTs.toEpochMilli() / 1000) * 1.2 // Add 20% tolerance to the upper bound
+                counterGrowth / secondsInWindow * 0.99, // Add 1% tolerance to the lower bound
+                1000.0 * counterGrowth / (lastTs.toEpochMilli() - firstTs.toEpochMilli()) * 1.01 // Add 1% tolerance to the upper bound
             );
         }).filter(Objects::nonNull).toList();
         if (allRates.isEmpty()) {
             return new RateStats(0L, null, null, null, new RateRange(0.0, 0.0));
         }
         return new RateStats(
             (long) allRates.size(),
-            allRates.stream().max(RateRange::compareTo).orElseThrow(),
+            allRates.stream().max(RateRange::compareToFindingMax).orElseThrow(),
             new RateRange(
                 allRates.stream().mapToDouble(r -> r.lower).average().orElseThrow(),
                 allRates.stream().mapToDouble(r -> r.upper).average().orElseThrow()
@@ -319,7 +333,7 @@ void checkWithin(Double actual, RateRange expected) {
 
     void assertNoFailedWindows(List<String> failedWindows, List<List<Object>> rows) {
         // TODO: WE have a 15% tolerance for failed windows. Must remove.
-        if (failedWindows.size() < 0.15 * rows.size()) {
+        if (failedWindows.size() < 0.01 * rows.size()) {
             logger.warn(
                 "Failed " + failedWindows.size() + " windows out of " + rows.size() + ", failures:\n" + String.join("\n", failedWindows)
             );

x-pack/plugin/esql/src/internalClusterTest/java/org/elasticsearch/xpack/esql/action/TSDataGenerationHelper.java

@@ -54,18 +54,21 @@ private static Object randomDimensionValue(String dimensionName) {
         // Making a list-to-set-to-list to ensure uniqueness.
         this.numDocs = numDocs;
         var maxAttributes = (int) Math.sqrt(numDocs);
-        attributesForMetrics = List.copyOf(
+        List<String> tempAttributeSet = List.copyOf(
             Set.copyOf(ESTestCase.randomList(1, maxAttributes, () -> ESTestCase.randomAlphaOfLengthBetween(2, 30)))
         );
         var maxTimeSeries = (int) Math.sqrt(numDocs);
         var minTimeSeries = Math.max(1, maxTimeSeries / 4);
         numTimeSeries = ESTestCase.randomIntBetween(minTimeSeries, maxTimeSeries);
+        Set<String> usedAttributeNames = new HashSet<>();
         // allTimeSeries contains the list of dimension-values for each time series.
         List<List<Tuple<String, Object>>> allTimeSeries = IntStream.range(0, numTimeSeries).mapToObj(tsIdx -> {
-            List<String> dimensionsInMetric = ESTestCase.randomNonEmptySubsetOf(attributesForMetrics);
+            List<String> dimensionsInMetric = ESTestCase.randomNonEmptySubsetOf(tempAttributeSet);
             // TODO: How do we handle the case when there are no dimensions? (i.e. regular randomSubsetof(...)
+            usedAttributeNames.addAll(dimensionsInMetric);
             return dimensionsInMetric.stream().map(attr -> new Tuple<>(attr, randomDimensionValue(attr))).collect(Collectors.toList());
         }).toList();
+        attributesForMetrics = List.copyOf(usedAttributeNames);
 
         // We want to ensure that all documents have different timestamps.
         var timeRangeMs = timeRangeSeconds * 1000;