Fail queries with scroll that explicitely set request_cache

core/src/main/java/org/elasticsearch/action/search/SearchRequest.java

@@ -169,6 +169,10 @@ public ActionRequestValidationException validate() {
             validationException =
                 addValidationError("using [from] is not allowed in a scroll context", validationException);
         }
+        if (requestCache != null && requestCache && scroll() != null) {
+            validationException =
+                addValidationError("[request_cache] cannot be used in a a scroll context", validationException);
+        }
         return validationException;
     }
 

core/src/main/java/org/elasticsearch/indices/IndicesService.java

@@ -1072,6 +1072,12 @@ public void close() {
      * Can the shard request be cached at all?
      */
     public boolean canCache(ShardSearchRequest request, SearchContext context) {
+        // Queries that create a scroll context cannot use the cache.
+        // They modify the search context during their execution so using the cache
+        // may invalidate the scroll for the next query.
+        if (request.scroll() != null) {
+            return false;
+        }
 
         // We cannot cache with DFS because results depend not only on the content of the index but also
         // on the overridden statistics. So if you ran two queries on the same index with different stats
@@ -1080,6 +1086,7 @@ public boolean canCache(ShardSearchRequest request, SearchContext context) {
         if (SearchType.QUERY_THEN_FETCH != context.searchType()) {
             return false;
         }
+
         IndexSettings settings = context.indexShard().indexSettings();
         // if not explicitly set in the request, use the index setting, if not, use the request
         if (request.requestCache() == null) {

docs/reference/migration/migrate_7_0/search.asciidoc

@@ -33,3 +33,10 @@ The Search API returns `400 - Bad request` while it would previously return
 *   the number of slices is too large
 *   keep alive for scroll is too large
 *   number of filters in the adjacency matrix aggregation is too large
+
+
+==== Scroll queries cannot use the request_cache anymore
+
+Setting `request_cache:true` on a query that creates a scroll ('scroll=1m`)
+has been deprecated in 6 and will now return a `400 - Bad request`.
+Scroll queries are not meant to be cached.

rest-api-spec/src/main/resources/rest-api-spec/test/scroll/10_basic.yml

@@ -197,3 +197,20 @@
       clear_scroll:
         scroll_id: $scroll_id
 
+---
+"Scroll cannot used the request cache":
+  - skip:
+      version: " - 6.99.99"
+      reason:  the error message has been added in v7.0.0
+  - do:
+      indices.create:
+        index: test_scroll
+  - do:
+      catch: /\[request_cache\] cannot be used in a a scroll context/
+      search:
+        index: test_scroll
+        scroll: 1m
+        request_cache: true
+        body:
+          query:
+            match_all: {}