From e8fe1107dcf6e3ce2abea229cb8359207e1112ff Mon Sep 17 00:00:00 2001
From: Ling Jin <7138436+3AceShowHand@users.noreply.github.com>
Date: Wed, 27 Jul 2022 14:05:11 +0800
Subject: [PATCH] cdc rolling upgrade remove status check to prevent false
 detection. (#1996)

---
 pkg/cluster/api/cdcapi.go |  4 ++--
 pkg/cluster/spec/cdc.go   | 12 +-----------
 2 files changed, 3 insertions(+), 13 deletions(-)

diff --git a/pkg/cluster/api/cdcapi.go b/pkg/cluster/api/cdcapi.go
index a57b4470ca..c5c2f8ede6 100644
--- a/pkg/cluster/api/cdcapi.go
+++ b/pkg/cluster/api/cdcapi.go
@@ -195,7 +195,7 @@ func (c *CDCOpenAPIClient) GetAllCaptures() (result []*Capture, err error) {
 		}
 		return nil
 	}, utils.RetryOption{
-		Timeout: 20 * time.Second,
+		Timeout: 10 * time.Second,
 	})
 	return result, err
 }
@@ -260,7 +260,7 @@ func (c *CDCOpenAPIClient) GetStatus() (result ServerStatus, err error) {
 		}
 		return nil
 	}, utils.RetryOption{
-		Timeout: 20 * time.Second,
+		Timeout: 10 * time.Second,
 	})
 
 	return result, err
diff --git a/pkg/cluster/spec/cdc.go b/pkg/cluster/spec/cdc.go
index 4ed7bacbb5..71750d232a 100644
--- a/pkg/cluster/spec/cdc.go
+++ b/pkg/cluster/spec/cdc.go
@@ -237,16 +237,11 @@ func (i *CDCInstance) PreRestart(ctx context.Context, topo Topology, apiTimeoutS
 		return nil
 	}
 
-	if i.Status(ctx, 5*time.Second, tlsCfg) == "Down" {
-		logger.Debugf("cdc pre-restart skipped, instance is down, trigger hard restart, addr: %s", address)
-		return nil
-	}
-
 	start := time.Now()
 	client := api.NewCDCOpenAPIClient(ctx, []string{address}, 5*time.Second, tlsCfg)
 	captures, err := client.GetAllCaptures()
 	if err != nil {
-		logger.Warnf("cdc pre-restart failed, cannot get all captures, trigger hard restart, addr: %s, elapsed: %+v", address, time.Since(start))
+		logger.Warnf("cdc pre-restart skipped, cannot get all captures, trigger hard restart, addr: %s, elapsed: %+v", address, time.Since(start))
 		return nil
 	}
 
@@ -306,13 +301,8 @@ func (i *CDCInstance) PostRestart(ctx context.Context, topo Topology, tlsCfg *tl
 
 	start := time.Now()
 	address := i.GetAddr()
-	if i.Status(ctx, 5*time.Second, tlsCfg) == "Down" {
-		logger.Debugf("cdc post-restart skipped, instance is down, addr: %s, elapsed: %+v", address, time.Since(start))
-		return nil
-	}
 
 	client := api.NewCDCOpenAPIClient(ctx, []string{address}, 5*time.Second, tlsCfg)
-
 	err := client.IsCaptureAlive()
 	if err != nil {
 		logger.Debugf("cdc post-restart finished, get capture status failed, addr: %s, err: %+v, elapsed: %+v", address, err, time.Since(start))