Predict best case overhead during autoscaling

Predicting worst case has the unavoidable consequence
that we will conclude we cannot fulfill requested resources
in the case where there are no resource ranges (only node count ranges),
or where they are too narrow, and thus refrain from scaling down
when we should.

Instead, mostly predict the best case and let the ideal < 1 headroom
absorb the deficiency in what we are actually allocated.

1 files changed, 3 insertions, 10 deletions

diff --git a/container-search/src/test/java/com/yahoo/search/yql/YqlParserTestCase.java b/container-search/src/test/java/com/yahoo/search/yql/YqlParserTestCase.java
index 33cd2f48d46..43aaba7b0f9 100644
--- a/container-search/src/test/java/com/yahoo/search/yql/YqlParserTestCase.java
+++ b/container-search/src/test/java/com/yahoo/search/yql/YqlParserTestCase.java
@@ -976,16 +976,9 @@ public class YqlParserTestCase {
         assertEquals(4, terms.size());
         for (IndexedItem term : terms) {
             switch (term.getIndexedString()) {
-                case "a":
-                case "c":
-                    assertFalse(((Item) term).isRanked());
-                    break;
-                case "b":
-                case "d":
-                    assertTrue(((Item) term).isRanked());
-                    break;
-                default:
-                    fail();
+                case "a", "c" -> assertFalse(((Item) term).isRanked());
+                case "b", "d" -> assertTrue(((Item) term).isRanked());
+                default -> fail();
             }
         }
     }