[7.1.0] Ensure that the mtime of an AC entry is smaller, not larger, …

…than the CAS blobs it references. (#21416)

7a774ff
got the condition backwards. Note that referential integrity isn't
required for correct operation, but lack of it can slow things down
unnecessarily.

PiperOrigin-RevId: 608354674
Change-Id: If56bb9211a5167d5db9e8eceb6224269758acdb1

src/main/java/com/google/devtools/build/lib/remote/disk/DiskCacheClient.java

@@ -232,6 +232,9 @@ public ListenableFuture<CachedActionResult> downloadActionResult(
     }
 
     return Futures.transformAsync(
+        // Update the mtime on the action result itself before any of the blobs it references.
+        // This ensures that the blobs are always newer than the action result, so that trimming the
+        // cache in LRU order cannot create dangling references.
         Utils.downloadAsActionResult(actionKey, (digest, out) -> download(digest, out, Store.AC)),
         actionResult -> {
           if (actionResult == null) {
@@ -249,11 +252,6 @@ public ListenableFuture<CachedActionResult> downloadActionResult(
             return immediateFuture(null);
           }
 
-          // Update the mtime for the action result itself. This ensures that blobs are older than
-          // the action result, so that trimming the cache in LRU order will not create dangling
-          // references.
-          var unused = refresh(toPath(actionKey.getDigest(), Store.AC));
-
           return immediateFuture(CachedActionResult.disk(actionResult));
         },
         MoreExecutors.directExecutor());