[Backport 2.19-dev] Fix the flaky CalcitePPLTcphIT

integ-test/src/test/java/org/opensearch/sql/calcite/tpch/CalcitePPLTpchIT.java

@@ -18,7 +18,6 @@
 
 import org.json.JSONObject;
 import org.junit.Assume;
-import org.junit.Ignore;
 import org.junit.Test;
 import org.opensearch.sql.ppl.PPLIntegTestCase;
 import org.opensearch.sql.util.Retry;
@@ -144,21 +143,19 @@ public void testQ3() throws IOException {
         rows(4423, 3055.9365, "1995-02-17 00:00:00", 0));
   }
 
-  // TODO: Aggregation push down has a hard-coded limit of 1000 buckets for output, so this query
-  // will not return the correct results with aggregation push down and it's unstable
-  @Ignore
+  @Test
   public void testQ4() throws IOException {
     String ppl = sanitize(loadFromFile("tpch/queries/q4.ppl"));
     JSONObject actual = executeQuery(ppl);
     verifySchemaInOrder(
         actual, schema("o_orderpriority", "string"), schema("order_count", "bigint"));
     verifyDataRows(
         actual,
-        rows("1-URGENT", 7),
+        rows("1-URGENT", 9),
         rows("2-HIGH", 7),
-        rows("3-MEDIUM", 4),
-        rows("4-NOT SPECIFIED", 7),
-        rows("5-LOW", 10));
+        rows("3-MEDIUM", 9),
+        rows("4-NOT SPECIFIED", 8),
+        rows("5-LOW", 12));
   }
 
   @Test
@@ -407,7 +404,6 @@ public void testQ18() throws IOException {
     verifyNumOfRows(actual, 0);
   }
 
-  @Ignore("This IT is easily flaky failure in 2.19.0, but more stable after 2.19.3")
   @Test
   public void testQ19() throws IOException {
     String ppl = sanitize(loadFromFile("tpch/queries/q19.ppl"));

integ-test/src/test/java/org/opensearch/sql/legacy/SQLIntegTestCase.java

@@ -202,6 +202,19 @@ protected synchronized void loadIndex(Index index, RestClient client) throws IOE
       createIndexByRestClient(client, indexName, mapping);
       loadDataByRestClient(client, indexName, dataSet);
     }
+    // loadIndex() could directly return when isIndexExist()=true,
+    // e.g. the index is created in the cluster but data hasn't been flushed.
+    // We block loadIndex() until data loaded to resolve
+    // https://github.com/opensearch-project/sql/issues/4261
+    int countDown = 3; // 1500ms timeout
+    while (countDown != 0 && getDocCount(client, indexName) == 0) {
+      try {
+        Thread.sleep(500);
+        countDown--;
+      } catch (InterruptedException e) {
+        throw new IOException(e);
+      }
+    }
   }
 
   protected synchronized void loadIndex(Index index) throws IOException {

integ-test/src/test/java/org/opensearch/sql/legacy/TestUtils.java

@@ -103,6 +103,21 @@ public static void loadDataByRestClient(
     performRequest(client, request);
   }
 
+  /**
+   * Return how many docs in the index
+   *
+   * @param client client connection
+   * @param indexName index name
+   * @return doc count of the index
+   * @throws IOException
+   */
+  public static int getDocCount(RestClient client, String indexName) throws IOException {
+    Request request = new Request("GET", "/" + indexName + "/_count");
+    Response response = performRequest(client, request);
+    JSONObject jsonObject = new JSONObject(getResponseBody(response));
+    return jsonObject.getInt("count");
+  }
+
   /**
    * Perform a request by REST client.
    *