Improve behaviour when all instances are failed

pkcs11/src/backend/login.rs

@@ -160,14 +160,24 @@ impl LoginCtx {
     }
 
     fn next_instance(&self) -> &InstanceData {
-        for _ in 0..self.slot.instances.len() {
-            let index = self.slot.instance_balancer.fetch_add(1, Relaxed);
-            let index = index % self.slot.instances.len();
-            let instance = &self.slot.instances[index];
+        let index = self.slot.instance_balancer.fetch_add(1, Relaxed);
+        let index = index % self.slot.instances.len();
+        let instance = &self.slot.instances[index];
+        match instance.should_try() {
+            InstanceAttempt::Failed => {}
+            InstanceAttempt::Working | InstanceAttempt::Retry => return instance,
+        }
+        for i in 0..self.slot.instances.len() - 1 {
+            let instance = &self.slot.instances[index + i];
 
             match instance.should_try() {
                 InstanceAttempt::Failed => continue,
-                InstanceAttempt::Working | InstanceAttempt::Retry => return instance,
+                InstanceAttempt::Working | InstanceAttempt::Retry => {
+                    // This not true round-robin in case of multithreaded acces
+                    // This is degraded mode so best-effort is attempted at best
+                    self.slot.instance_balancer.fetch_add(i, Relaxed);
+                    return instance;
+                }
             }
         }
 
@@ -255,7 +265,10 @@ impl LoginCtx {
             retry_count += 1;
             let api_call_clone = api_call.clone();
             match api_call_clone(&instance.config) {
-                Ok(result) => return Ok(result),
+                Ok(result) => {
+                    instance.clear_failed();
+                    return Ok(result);
+                }
 
                 // If the server is in an unusable state, skip retries and try the next one
                 Err(apis::Error::ResponseError(err @ ResponseContent { status: 500, .. }))