diff --git a/CHANGELOG.md b/CHANGELOG.md
index cadba2e6eed..a9d172a2468 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -26,7 +26,8 @@
 * [BUGFIX] Set appropriate `Content-Type` header for /services endpoint, which previously hard-coded `text/plain`. #4596
 * [BUGFIX] Querier: Disable query scheduler SRV DNS lookup, which removes noisy log messages about "failed DNS SRV record lookup". #4601
 * [BUGFIX] Memberlist: fixed corrupted packets when sending compound messages with more than 255 messages or messages bigger than 64KB. #4601
-* [BUGIX] Query Frontend: If 'LogQueriesLongerThan' is set to < 0, log all queries as described in the docs. #4633
+* [BUGFIX] Query Frontend: If 'LogQueriesLongerThan' is set to < 0, log all queries as described in the docs. #4633
+* [BUGFIX] Distributor: update defaultReplicationStrategy to not fail with extend-write when a single instance is unhealthy. #4636
 
 ## 1.11.0 2021-11-25
 
diff --git a/pkg/ring/replication_strategy.go b/pkg/ring/replication_strategy.go
index 44e05a53833..32eff1384a5 100644
--- a/pkg/ring/replication_strategy.go
+++ b/pkg/ring/replication_strategy.go
@@ -26,19 +26,9 @@ func NewDefaultReplicationStrategy() ReplicationStrategy {
 // - Checks there are enough instances for an operation to succeed.
 // The instances argument may be overwritten.
 func (s *defaultReplicationStrategy) Filter(instances []InstanceDesc, op Operation, replicationFactor int, heartbeatTimeout time.Duration, zoneAwarenessEnabled bool) ([]InstanceDesc, int, error) {
-	// We need a response from a quorum of instances, which is n/2 + 1.  In the
-	// case of a node joining/leaving, the actual replica set might be bigger
-	// than the replication factor, so use the bigger or the two.
-	if len(instances) > replicationFactor {
-		replicationFactor = len(instances)
-	}
-
-	minSuccess := (replicationFactor / 2) + 1
 	now := time.Now()
 
-	// Skip those that have not heartbeated in a while. NB these are still
-	// included in the calculation of minSuccess, so if too many failed instances
-	// will cause the whole write to fail.
+	// Skip those that have not heartbeated in a while.
 	var unhealthy []string
 	for i := 0; i < len(instances); {
 		if instances[i].IsHealthy(op, heartbeatTimeout, now) {
@@ -49,6 +39,14 @@ func (s *defaultReplicationStrategy) Filter(instances []InstanceDesc, op Operati
 		}
 	}
 
+	// We need a response from a quorum of instances, which is n/2 + 1.  In the
+	// case of a node joining/leaving with extend-writes enabled, the actual replica
+	// set will be bigger than the replication factor, so use the bigger or the two.
+	if len(instances) > replicationFactor {
+		replicationFactor = len(instances)
+	}
+
+	minSuccess := (replicationFactor / 2) + 1
 	// This is just a shortcut - if there are not minSuccess available instances,
 	// after filtering out dead ones, don't even bother trying.
 	if len(instances) < minSuccess {
diff --git a/pkg/ring/replication_strategy_test.go b/pkg/ring/replication_strategy_test.go
index 1fe5d0e9187..75af12671ce 100644
--- a/pkg/ring/replication_strategy_test.go
+++ b/pkg/ring/replication_strategy_test.go
@@ -68,14 +68,14 @@ func TestRingReplicationStrategy(t *testing.T) {
 			replicationFactor:  3,
 			liveIngesters:      3,
 			deadIngesters:      1,
-			expectedMaxFailure: 0,
+			expectedMaxFailure: 1,
 		},
 
 		{
-			replicationFactor: 3,
-			liveIngesters:     2,
-			deadIngesters:     2,
-			expectedError:     "at least 3 live replicas required, could only find 2 - unhealthy instances: dead1,dead2",
+			replicationFactor:  3,
+			liveIngesters:      2,
+			deadIngesters:      2,
+			expectedMaxFailure: 0,
 		},
 	} {
 		ingesters := []InstanceDesc{}