From fa89999d919c1a1d2e8c348f8acf4ca69623aeac Mon Sep 17 00:00:00 2001
From: Marina Biryukova <m.biryukova@yadro.com>
Date: Mon, 22 Jul 2024 11:17:26 +0300
Subject: [PATCH] [#242] pool: Log error that caused healthy status change

Signed-off-by: Marina Biryukova <m.biryukova@yadro.com>
---
 pool/mock_test.go |  6 +++---
 pool/pool.go      | 31 ++++++++++++++++++-------------
 2 files changed, 21 insertions(+), 16 deletions(-)

diff --git a/pool/mock_test.go b/pool/mock_test.go
index 03a981b..d555afd 100644
--- a/pool/mock_test.go
+++ b/pool/mock_test.go
@@ -195,9 +195,9 @@ func (m *mockClient) dial(context.Context) error {
 	return nil
 }
 
-func (m *mockClient) restartIfUnhealthy(ctx context.Context) (healthy bool, changed bool) {
-	_, err := m.endpointInfo(ctx, prmEndpointInfo{})
-	healthy = err == nil
+func (m *mockClient) restartIfUnhealthy(ctx context.Context) (changed bool, err error) {
+	_, err = m.endpointInfo(ctx, prmEndpointInfo{})
+	healthy := err == nil
 	changed = healthy != m.isHealthy()
 	if healthy {
 		m.setHealthy()
diff --git a/pool/pool.go b/pool/pool.go
index 651702b..2f9a651 100644
--- a/pool/pool.go
+++ b/pool/pool.go
@@ -86,7 +86,7 @@ type client interface {
 	// see clientWrapper.dial.
 	dial(ctx context.Context) error
 	// see clientWrapper.restartIfUnhealthy.
-	restartIfUnhealthy(ctx context.Context) (bool, bool)
+	restartIfUnhealthy(ctx context.Context) (bool, error)
 	// see clientWrapper.close.
 	close() error
 }
@@ -373,11 +373,11 @@ func (c *clientWrapper) dial(ctx context.Context) error {
 }
 
 // restartIfUnhealthy checks healthy status of client and recreate it if status is unhealthy.
-// Return current healthy status and indicating if status was changed by this function call.
-func (c *clientWrapper) restartIfUnhealthy(ctx context.Context) (healthy, changed bool) {
+// Indicating if status was changed by this function call and returns error that caused unhealthy status.
+func (c *clientWrapper) restartIfUnhealthy(ctx context.Context) (changed bool, err error) {
 	var wasHealthy bool
-	if _, err := c.endpointInfo(ctx, prmEndpointInfo{}); err == nil {
-		return true, false
+	if _, err = c.endpointInfo(ctx, prmEndpointInfo{}); err == nil {
+		return false, nil
 	} else if !errors.Is(err, errPoolClientUnhealthy) {
 		wasHealthy = true
 	}
@@ -403,22 +403,22 @@ func (c *clientWrapper) restartIfUnhealthy(ctx context.Context) (healthy, change
 		GRPCDialOptions: c.prm.dialOptions,
 	}
 
-	if err := cl.Dial(ctx, prmDial); err != nil {
+	if err = cl.Dial(ctx, prmDial); err != nil {
 		c.setUnhealthyOnDial()
-		return false, wasHealthy
+		return wasHealthy, err
 	}
 
 	c.clientMutex.Lock()
 	c.client = &cl
 	c.clientMutex.Unlock()
 
-	if _, err := cl.EndpointInfo(ctx, sdkClient.PrmEndpointInfo{}); err != nil {
+	if _, err = cl.EndpointInfo(ctx, sdkClient.PrmEndpointInfo{}); err != nil {
 		c.setUnhealthy()
-		return false, wasHealthy
+		return wasHealthy, err
 	}
 
 	c.setHealthy()
-	return true, !wasHealthy
+	return !wasHealthy, nil
 }
 
 func (c *clientWrapper) getClient() (*sdkClient.Client, error) {
@@ -2198,7 +2198,8 @@ func (p *Pool) updateInnerNodesHealth(ctx context.Context, i int, bufferWeights
 			tctx, c := context.WithTimeout(ctx, options.nodeRequestTimeout)
 			defer c()
 
-			healthy, changed := cli.restartIfUnhealthy(tctx)
+			changed, err := cli.restartIfUnhealthy(tctx)
+			healthy := err == nil
 			if healthy {
 				bufferWeights[j] = options.nodesParams[i].weights[j]
 			} else {
@@ -2207,8 +2208,12 @@ func (p *Pool) updateInnerNodesHealth(ctx context.Context, i int, bufferWeights
 			}
 
 			if changed {
-				p.log(zap.DebugLevel, "health has changed",
-					zap.String("address", cli.address()), zap.Bool("healthy", healthy))
+				fields := []zap.Field{zap.String("address", cli.address()), zap.Bool("healthy", healthy)}
+				if err != nil {
+					fields = append(fields, zap.String("reason", err.Error()))
+				}
+
+				p.log(zap.DebugLevel, "health has changed", fields...)
 				healthyChanged.Store(true)
 			}
 		}(j, cli)