SERVER-44675 Handle race in repl metrics test

1 files changed, 14 insertions, 1 deletions

diff --git a/jstests/replsets/server_status_metrics.js b/jstests/replsets/server_status_metrics.js
index 7b271d197f4..d33f61347bc 100644
--- a/jstests/replsets/server_status_metrics.js
+++ b/jstests/replsets/server_status_metrics.js
@@ -1,7 +1,7 @@
 /**
  * Test replication metrics
  */
-function testSecondaryMetrics(secondary, opCount, baseOpsApplied, baseOpsReceived) {
+function _testSecondaryMetricsHelper(secondary, opCount, baseOpsApplied, baseOpsReceived) {
     var ss = secondary.getDB("test").serverStatus();
     printjson(ss.metrics);
 
@@ -30,6 +30,19 @@ function testSecondaryMetrics(secondary, opCount, baseOpsApplied, baseOpsReceive
     assert.eq(ss.metrics.repl.apply.ops, opCount + baseOpsApplied, "wrong number of applied ops");
 }
 
+// Metrics are racy, e.g. repl.buffer.count could over- or under-reported briefly. Retry on error.
+function testSecondaryMetrics(secondary, opCount, baseOpsApplied, baseOpsReceived) {
+    assert.soon(() => {
+        try {
+            _testSecondaryMetricsHelper(secondary, opCount, baseOpsApplied, baseOpsReceived);
+            return true;
+        } catch (exc) {
+            jsTestLog(`Caught ${exc}, retrying`);
+            return false;
+        }
+    });
+}
+
 var rt = new ReplSetTest({name: "server_status_metrics", nodes: 2, oplogSize: 100});
 rt.startSet();
 rt.initiate();