[ML] Fix null pointer exception after all datafeed indices deleted (e…

…lastic#76401)

A datafeed will not start running if no indices exist that match
the pattern it's configured to search. However, it is possible that
after the datafeed has started somebody could delete all the indices
that it is configured to search.

This change handles that situation more gracefully, avoiding
repeated logging of null pointer exceptions to the Elasticsearch log.

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/datafeed/extractor/chunked/ChunkedDataExtractor.java

@@ -51,7 +51,7 @@
  */
 public class ChunkedDataExtractor implements DataExtractor {
 
-    private interface DataSummary {
+    interface DataSummary {
         long estimateChunk();
         boolean hasData();
         long earliestTime();
@@ -209,11 +209,11 @@ private DataSummary newScrolledDataSummary() {
             LOGGER.debug("[{}] Scrolling Data summary response was obtained", context.jobId);
             timingStatsReporter.reportSearchDuration(searchResponse.getTook());
 
-            Aggregations aggregations = searchResponse.getAggregations();
             long earliestTime = 0;
             long latestTime = 0;
             long totalHits = searchResponse.getHits().getTotalHits().value;
             if (totalHits > 0) {
+                Aggregations aggregations = searchResponse.getAggregations();
                 Min min = aggregations.get(EARLIEST_TIME);
                 earliestTime = (long) min.getValue();
                 Max max = aggregations.get(LATEST_TIME);
@@ -231,6 +231,13 @@ private DataSummary newAggregatedDataSummary() {
             timingStatsReporter.reportSearchDuration(searchResponse.getTook());
 
             Aggregations aggregations = searchResponse.getAggregations();
+            // This can happen if all the indices the datafeed is searching are deleted after it started.
+            // Note that unlike the scrolled data summary method above we cannot check for this situation
+            // by checking for zero hits, because aggregations that work on rollups return zero hits even
+            // when they retrieve data.
+            if (aggregations == null) {
+                return AggregatedDataSummary.noDataSummary(context.histogramInterval);
+            }
             Min min = aggregations.get(EARLIEST_TIME);
             Max max = aggregations.get(LATEST_TIME);
             return new AggregatedDataSummary(min.getValue(), max.getValue(), context.histogramInterval);
@@ -309,13 +316,18 @@ public boolean hasData() {
         }
     }
 
-    private static class AggregatedDataSummary implements DataSummary {
+    static class AggregatedDataSummary implements DataSummary {
 
         private final double earliestTime;
         private final double latestTime;
         private final long histogramIntervalMillis;
 
-        private AggregatedDataSummary(double earliestTime, double latestTime, long histogramInterval) {
+        static AggregatedDataSummary noDataSummary(long histogramInterval) {
+            // hasData() uses infinity to mean no data
+            return new AggregatedDataSummary(Double.POSITIVE_INFINITY, Double.POSITIVE_INFINITY, histogramInterval);
+        }
+
+        AggregatedDataSummary(double earliestTime, double latestTime, long histogramInterval) {
             this.earliestTime = earliestTime;
             this.latestTime = latestTime;
             this.histogramIntervalMillis = histogramInterval;

x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/datafeed/extractor/chunked/ChunkedDataExtractorTests.java

@@ -159,7 +159,8 @@ public void testExtractionGivenSpecifiedChunk() throws IOException {
     public void testExtractionGivenSpecifiedChunkAndAggs() throws IOException {
         chunkSpan = TimeValue.timeValueSeconds(1);
         TestDataExtractor extractor = new TestDataExtractor(1000L, 2300L, true, 1000L);
-        extractor.setNextResponse(createSearchResponse(0L, 1000L, 2200L));
+        // 0 hits with non-empty data is possible with rollups
+        extractor.setNextResponse(createSearchResponse(randomFrom(0L, 2L, 10000L), 1000L, 2200L));
 
         InputStream inputStream1 = mock(InputStream.class);
         InputStream inputStream2 = mock(InputStream.class);
@@ -200,7 +201,8 @@ public void testExtractionGivenAutoChunkAndAggs() throws IOException {
         chunkSpan = null;
         TestDataExtractor extractor = new TestDataExtractor(100_000L, 450_000L, true, 200L);
 
-        extractor.setNextResponse(createSearchResponse(0L, 100_000L, 400_000L));
+        // 0 hits with non-empty data is possible with rollups
+        extractor.setNextResponse(createSearchResponse(randomFrom(0L, 2L, 10000L), 100_000L, 400_000L));
 
         InputStream inputStream1 = mock(InputStream.class);
         InputStream inputStream2 = mock(InputStream.class);
@@ -504,6 +506,11 @@ public void testDataSummaryRequestIsFailed() {
         expectThrows(SearchPhaseExecutionException.class, extractor::next);
     }
 
+    public void testNoDataSummaryHasNoData() {
+        ChunkedDataExtractor.DataSummary summary = ChunkedDataExtractor.AggregatedDataSummary.noDataSummary(randomNonNegativeLong());
+        assertFalse(summary.hasData());
+    }
+
     private SearchResponse createSearchResponse(long totalHits, long earliestTime, long latestTime) {
         SearchResponse searchResponse = mock(SearchResponse.class);
         when(searchResponse.status()).thenReturn(RestStatus.OK);