Add metric to measure latency for guessing relevant groups

Guessing groups is invoked from some group backends to find the groups
of a user (e.g. from LdapGroupBackend). The groups of the calling user
are needed for the change etag computation. Hence if guessing groups is
slow this has direct impact on the speed of the change etag computation
which makes it interesting to know how much latency we spend on guessing
relevant groups.

Change-Id: Ibae879a702af007f1a326d01d15553487926bc4a
Signed-off-by: Edwin Kempin <ekempin@google.com>

Documentation/metrics.txt

diff --git a/Documentation/metrics.txt b/Documentation/metrics.txt
index 37d6d01..db6f883 100644
--- a/Documentation/metrics.txt
+++ b/Documentation/metrics.txt
@@ -177,6 +177,10 @@
 * `plugin/latency`: Latency for plugin invocation.
 * `plugin/error_count`: Number of plugin errors.
 
+=== Group
+
+* `group/guess_relevant_groups_latency`: Latency for guessing relevant groups.
+
 === Replication Plugin
 
 * `plugins/replication/replication_latency`: Time spent pushing to remote

java/com/google/gerrit/server/project/ProjectCacheImpl.java

diff --git a/java/com/google/gerrit/server/project/ProjectCacheImpl.java b/java/com/google/gerrit/server/project/ProjectCacheImpl.java
index aa455e6..a930037 100644
--- a/java/com/google/gerrit/server/project/ProjectCacheImpl.java
+++ b/java/com/google/gerrit/server/project/ProjectCacheImpl.java
@@ -25,6 +25,10 @@
 import com.google.common.flogger.FluentLogger;
 import com.google.gerrit.index.project.ProjectIndexer;
 import com.google.gerrit.lifecycle.LifecycleModule;
+import com.google.gerrit.metrics.Description;
+import com.google.gerrit.metrics.Description.Units;
+import com.google.gerrit.metrics.MetricMaker;
+import com.google.gerrit.metrics.Timer0;
 import com.google.gerrit.reviewdb.client.AccountGroup;
 import com.google.gerrit.reviewdb.client.Project;
 import com.google.gerrit.server.cache.CacheModule;
@@ -89,6 +93,7 @@
   private final Lock listLock;
   private final ProjectCacheClock clock;
   private final Provider<ProjectIndexer> indexer;
+  private final Timer0 guessRelevantGroupsLatency;
 
   @Inject
   ProjectCacheImpl(
@@ -97,7 +102,8 @@
       @Named(CACHE_NAME) LoadingCache<String, ProjectState> byName,
       @Named(CACHE_LIST) LoadingCache<ListKey, ImmutableSortedSet<Project.NameKey>> list,
       ProjectCacheClock clock,
-      Provider<ProjectIndexer> indexer) {
+      Provider<ProjectIndexer> indexer,
+      MetricMaker metricMaker) {
     this.allProjectsName = allProjectsName;
     this.allUsersName = allUsersName;
     this.byName = byName;
@@ -105,6 +111,13 @@
     this.listLock = new ReentrantLock(true /* fair */);
     this.clock = clock;
     this.indexer = indexer;
+
+    this.guessRelevantGroupsLatency =
+        metricMaker.newTimer(
+            "group/guess_relevant_groups_latency",
+            new Description("Latency for guessing relevant groups")
+                .setCumulative()
+                .setUnit(Units.NANOSECONDS));
   }
 
   @Override
@@ -234,15 +247,17 @@
 
   @Override
   public Set<AccountGroup.UUID> guessRelevantGroupUUIDs() {
-    return all()
-        .stream()
-        .map(n -> byName.getIfPresent(n.get()))
-        .filter(Objects::nonNull)
-        .flatMap(p -> p.getConfig().getAllGroupUUIDs().stream())
-        // getAllGroupUUIDs shouldn't really return null UUIDs, but harden
-        // against them just in case there is a bug or corner case.
-        .filter(id -> id != null && id.get() != null)
-        .collect(toSet());
+    try (Timer0.Context ignored = guessRelevantGroupsLatency.start()) {
+      return all()
+          .stream()
+          .map(n -> byName.getIfPresent(n.get()))
+          .filter(Objects::nonNull)
+          .flatMap(p -> p.getConfig().getAllGroupUUIDs().stream())
+          // getAllGroupUUIDs shouldn't really return null UUIDs, but harden
+          // against them just in case there is a bug or corner case.
+          .filter(id -> id != null && id.get() != null)
+          .collect(toSet());
+    }
   }
 
   @Override