diff --git a/rest-api-spec/src/main/resources/rest-api-spec/test/search.aggregation/300_pipeline.yml b/rest-api-spec/src/main/resources/rest-api-spec/test/search.aggregation/300_pipeline.yml
index 562b8c2c6e1..a5c575ab5ac 100644
--- a/rest-api-spec/src/main/resources/rest-api-spec/test/search.aggregation/300_pipeline.yml
+++ b/rest-api-spec/src/main/resources/rest-api-spec/test/search.aggregation/300_pipeline.yml
@@ -121,3 +121,72 @@ setup:
               bucket_sort:
                 sort:
                   - the_terms>the_max
+
+---
+"deep cumulative sum":
+  - do:
+      indices.create:
+        index: test
+        body:
+          settings:
+            number_of_replicas: 0
+            number_of_shards: 3
+  - do:
+      bulk:
+        index: test
+        refresh: true
+        body:
+          - '{"index": {}}'
+          - '{"t": "a", "int" :  1, "@timestamp" : "2020-07-16T00:00:00.000Z"}'
+          - '{"index": {}}'
+          - '{"t": "a", "int" : 50, "@timestamp" : "2020-07-17T00:00:00.000Z"}'
+          - '{"index": {}}'
+          - '{"t": "a", "int" : 99, "@timestamp" : "2020-07-18T00:00:00.000Z"}'
+          - '{"index": {}}'
+          - '{"t": "b", "int" :  1, "@timestamp" : "2020-07-16T00:00:00.000Z"}'
+          - '{"index": {}}'
+          - '{"t": "b", "int" : 99, "@timestamp" : "2020-07-17T00:00:00.000Z"}'
+
+  - do:
+      search:
+        rest_total_hits_as_int: true
+        index: test
+        body:
+          size: 0
+          aggs:
+            t:
+              terms:
+                field: t.keyword
+              aggs:
+                by_date:
+                  date_histogram:
+                    field: "@timestamp"
+                    fixed_interval: 1d
+                  aggs:
+                    avg:
+                      avg:
+                        field: int
+                    sum:
+                      cumulative_sum:
+                        buckets_path: avg.value
+  - match: { hits.total: 5 }
+  - length: { aggregations.t.buckets: 2 }
+  - match: { aggregations.t.buckets.0.key: a }
+  - match: { aggregations.t.buckets.1.key: b }
+  - length: { aggregations.t.buckets.0.by_date.buckets: 3 }
+  - match: { aggregations.t.buckets.0.by_date.buckets.0.key_as_string: "2020-07-16T00:00:00.000Z" }
+  - match: { aggregations.t.buckets.0.by_date.buckets.0.avg.value: 1 }
+  - match: { aggregations.t.buckets.0.by_date.buckets.0.sum.value: 1 }
+  - match: { aggregations.t.buckets.0.by_date.buckets.1.key_as_string: "2020-07-17T00:00:00.000Z" }
+  - match: { aggregations.t.buckets.0.by_date.buckets.1.avg.value: 50 }
+  - match: { aggregations.t.buckets.0.by_date.buckets.1.sum.value: 51 }
+  - match: { aggregations.t.buckets.0.by_date.buckets.2.key_as_string: "2020-07-18T00:00:00.000Z" }
+  - match: { aggregations.t.buckets.0.by_date.buckets.2.avg.value: 99 }
+  - match: { aggregations.t.buckets.0.by_date.buckets.2.sum.value: 150 }
+  - length: { aggregations.t.buckets.1.by_date.buckets: 2 }
+  - match: { aggregations.t.buckets.1.by_date.buckets.0.key_as_string: "2020-07-16T00:00:00.000Z" }
+  - match: { aggregations.t.buckets.1.by_date.buckets.0.avg.value: 1 }
+  - match: { aggregations.t.buckets.1.by_date.buckets.0.sum.value: 1 }
+  - match: { aggregations.t.buckets.1.by_date.buckets.1.key_as_string: "2020-07-17T00:00:00.000Z" }
+  - match: { aggregations.t.buckets.1.by_date.buckets.1.avg.value: 99 }
+  - match: { aggregations.t.buckets.1.by_date.buckets.1.sum.value: 100 }
diff --git a/server/src/main/java/org/elasticsearch/search/aggregations/InternalAggregation.java b/server/src/main/java/org/elasticsearch/search/aggregations/InternalAggregation.java
index 2e6eb33b798..9422d6c7bde 100644
--- a/server/src/main/java/org/elasticsearch/search/aggregations/InternalAggregation.java
+++ b/server/src/main/java/org/elasticsearch/search/aggregations/InternalAggregation.java
@@ -164,6 +164,16 @@ public abstract class InternalAggregation implements Aggregation, NamedWriteable
      * 7.8.0.
      */
     public final void mergePipelineTreeForBWCSerialization(PipelineAggregator.PipelineTree pipelineTree) {
+        if (pipelineAggregatorsForBwcSerialization != null) {
+            /*
+             * This method is called once per level on the results but only
+             * has useful pipeline aggregations on the top level. Every level
+             * below the top will always be empty. So if we've already been
+             * called we should bail. This is pretty messy but it is the kind
+             * of weird thing we have to do to deal with bwc serialization....
+             */
+            return;
+        }
         pipelineAggregatorsForBwcSerialization = pipelineTree.aggregators();
         forEachBucket(bucketAggs -> bucketAggs.mergePipelineTreeForBWCSerialization(pipelineTree));
     }
diff --git a/test/framework/src/main/java/org/elasticsearch/test/InternalAggregationTestCase.java b/test/framework/src/main/java/org/elasticsearch/test/InternalAggregationTestCase.java
index 87b38e80ff9..4457ca46775 100644
--- a/test/framework/src/main/java/org/elasticsearch/test/InternalAggregationTestCase.java
+++ b/test/framework/src/main/java/org/elasticsearch/test/InternalAggregationTestCase.java
@@ -460,6 +460,14 @@ public abstract class InternalAggregationTestCase<T extends InternalAggregation>
         assertMergedPipelineTreeForBWCSerialization(agg, pipelineTree);
     }
 
+    public void testMergePipelineTreeTwice() {
+        T agg = createTestInstance();
+        PipelineAggregator.PipelineTree pipelineTree = randomPipelineTree(agg);
+        agg.mergePipelineTreeForBWCSerialization(pipelineTree);
+        agg.mergePipelineTreeForBWCSerialization(randomPipelineTree(agg)); // This should be ignored
+        assertMergedPipelineTreeForBWCSerialization(agg, pipelineTree);
+    }
+
     public static PipelineAggregator.PipelineTree randomPipelineTree(InternalAggregation aggregation) {
         Map<String, PipelineTree> subTree = new HashMap<>();
         aggregation.forEachBucket(bucketAggs -> {