Use TDigest fast mode in ESQL (ESQL-1361)

Ideally, we should allow users to select the T-digest mode: fast or
accuracy; but the default should be the fast mode.

Author: dnhatn

x-pack/plugin/esql/compute/src/main/java/org/elasticsearch/compute/aggregation/QuantileStates.java

@@ -49,7 +49,7 @@ static int serializeDigest(TDigestState digest, byte[] ba, int offset) {
 
     static TDigestState deserializeDigest(byte[] ba, int offset) {
         final double compression = (double) doubleHandle.get(ba, offset);
-        final TDigestState digest = TDigestState.createOptimizedForAccuracy(compression);
+        final TDigestState digest = TDigestState.create(compression);
         final int positions = (int) intHandle.get(ba, offset + 8);
         offset += 12;
         for (int i = 0; i < positions; i++) {
@@ -72,7 +72,7 @@ static class SingleState implements AggregatorState<SingleState> {
         private final Double percentile;
 
         SingleState(double percentile) {
-            this.digest = TDigestState.createOptimizedForAccuracy(DEFAULT_COMPRESSION);
+            this.digest = TDigestState.create(DEFAULT_COMPRESSION);
             this.percentile = percentileParam(percentile);
         }
 
@@ -160,7 +160,7 @@ private TDigestState getOrAddGroup(int groupId) {
             }
             TDigestState qs = digests.get(groupId);
             if (qs == null) {
-                qs = TDigestState.createOptimizedForAccuracy(DEFAULT_COMPRESSION);
+                qs = TDigestState.create(DEFAULT_COMPRESSION);
                 digests.set(groupId, qs);
             }
             return qs;

x-pack/plugin/esql/compute/src/test/java/org/elasticsearch/compute/aggregation/PercentileDoubleAggregatorFunctionTests.java

@@ -47,7 +47,7 @@ protected SourceOperator simpleInput(int size) {
 
     @Override
     protected void assertSimpleOutput(List<Block> input, Block result) {
-        TDigestState td = TDigestState.createOptimizedForAccuracy(QuantileStates.DEFAULT_COMPRESSION);
+        TDigestState td = TDigestState.create(QuantileStates.DEFAULT_COMPRESSION);
         input.stream().flatMapToDouble(b -> allDoubles(b)).forEach(td::add);
         double expected = td.quantile(percentile / 100);
         double value = ((DoubleBlock) result).getDouble(0);

x-pack/plugin/esql/compute/src/test/java/org/elasticsearch/compute/aggregation/PercentileDoubleGroupingAggregatorFunctionTests.java

@@ -50,7 +50,7 @@ protected SourceOperator simpleInput(int end) {
 
     @Override
     protected void assertSimpleGroup(List<Page> input, Block result, int position, long group) {
-        TDigestState td = TDigestState.createOptimizedForAccuracy(QuantileStates.DEFAULT_COMPRESSION);
+        TDigestState td = TDigestState.create(QuantileStates.DEFAULT_COMPRESSION);
         input.stream().flatMapToDouble(p -> allDoubles(p, group)).forEach(td::add);
         if (td.size() > 0) {
             double expected = td.quantile(percentile / 100);

x-pack/plugin/esql/compute/src/test/java/org/elasticsearch/compute/aggregation/PercentileIntAggregatorFunctionTests.java

@@ -47,7 +47,7 @@ protected SourceOperator simpleInput(int size) {
 
     @Override
     protected void assertSimpleOutput(List<Block> input, Block result) {
-        TDigestState td = TDigestState.createOptimizedForAccuracy(QuantileStates.DEFAULT_COMPRESSION);
+        TDigestState td = TDigestState.create(QuantileStates.DEFAULT_COMPRESSION);
         input.stream().flatMapToInt(b -> allInts(b)).forEach(td::add);
         double expected = td.quantile(percentile / 100);
         double value = ((DoubleBlock) result).getDouble(0);

x-pack/plugin/esql/compute/src/test/java/org/elasticsearch/compute/aggregation/PercentileIntGroupingAggregatorFunctionTests.java

@@ -51,7 +51,7 @@ protected SourceOperator simpleInput(int size) {
 
     @Override
     protected void assertSimpleGroup(List<Page> input, Block result, int position, long group) {
-        TDigestState td = TDigestState.createOptimizedForAccuracy(QuantileStates.DEFAULT_COMPRESSION);
+        TDigestState td = TDigestState.create(QuantileStates.DEFAULT_COMPRESSION);
         input.stream().flatMapToInt(p -> allInts(p, group)).forEach(td::add);
         if (td.size() > 0) {
             double expected = td.quantile(percentile / 100);

x-pack/plugin/esql/compute/src/test/java/org/elasticsearch/compute/aggregation/PercentileLongAggregatorFunctionTests.java

@@ -47,7 +47,7 @@ protected SourceOperator simpleInput(int size) {
 
     @Override
     protected void assertSimpleOutput(List<Block> input, Block result) {
-        TDigestState td = TDigestState.createOptimizedForAccuracy(QuantileStates.DEFAULT_COMPRESSION);
+        TDigestState td = TDigestState.create(QuantileStates.DEFAULT_COMPRESSION);
         input.stream().flatMapToLong(p -> allLongs(p)).forEach(td::add);
         double expected = td.quantile(percentile / 100);
         double value = ((DoubleBlock) result).getDouble(0);

x-pack/plugin/esql/compute/src/test/java/org/elasticsearch/compute/aggregation/PercentileLongGroupingAggregatorFunctionTests.java

@@ -51,7 +51,7 @@ protected SourceOperator simpleInput(int size) {
 
     @Override
     protected void assertSimpleGroup(List<Page> input, Block result, int position, long group) {
-        TDigestState td = TDigestState.createOptimizedForAccuracy(QuantileStates.DEFAULT_COMPRESSION);
+        TDigestState td = TDigestState.create(QuantileStates.DEFAULT_COMPRESSION);
         input.stream().flatMapToLong(p -> allLongs(p, group)).forEach(td::add);
         if (td.size() > 0) {
             double expected = td.quantile(percentile / 100);

x-pack/plugin/esql/qa/testFixtures/src/main/resources/stats_percentile.csv-spec

@@ -17,7 +17,7 @@ FROM employees
 
 // tag::percentile-result[]
 p0:double   | p50:double  | p99:double
-25324       | 47003       | 74984.5
+25324       | 47003       | 74970.29
 // end::percentile-result[]
 ;
 
@@ -26,7 +26,7 @@ percentileOfDouble
 from employees | stats p0 = percentile(salary_change, 0), p50 = percentile(salary_change, 50), p99 = percentile(salary_change, 99);
 
 p0:double   | p50:double  | p99:double
--9.81       | 0.75        | 14.663499999999999
+-9.81       | 0.75        | 14.639000000000001
 ;
 
 
@@ -35,30 +35,30 @@ from employees | stats p90 = percentile(salary_change.long, 90) by job_positions
 
 p90:double         | job_positions:keyword
  7                 | "Python Developer"
-10.399999999999999 | "Business Analyst"
-11                 | "Accountant"
-11                 | "Tech Lead"
+9.600000000000001  | "Business Analyst"
+10.200000000000006 | "Data Scientist"
+10.399999999999999 | "Senior Python Developer"
 ;
 
 percentileOfIntegerByKeyword
 from employees | stats p90 = percentile(salary, 90) by job_positions | sort p90 | limit 4;
 
-p90:double      | job_positions:keyword
-53397.8         | "Business Analyst"
-56840.4         | "Support Engineer"
-57565           | "Head Human Resources"
-61358           | "Reporting Analyst"
+p90:double           | job_positions:keyword
+50249.0              | "Business Analyst"
+54462.0              | "Support Engineer"
+56308.799999999996   | "Reporting Analyst"
+56645.0              | "Head Human Resources"
 ;
 
 
 percentileOfDoubleByKeyword
 from employees | stats p90 = percentile(salary_change, 90) by job_positions | sort p90 | limit 4;
 
 p90:double            | job_positions:keyword
-7.652                 | "Python Developer"
-10.994999999999997    | "Business Analyst"
-11.301000000000002    | "Senior Team Lead"
-11.514000000000001    | "Data Scientist"
+7.5760000000000005    | "Python Developer"
+10.095000000000002    | "Business Analyst"
+10.362000000000007    | "Data Scientist"
+10.964999999999998    | "Senior Python Developer"
 ;