From 95387c3b5effc6552b72f2b26d077dbae303c90a Mon Sep 17 00:00:00 2001
From: Madhu Rajanna <madhupr007@gmail.com>
Date: Thu, 18 Mar 2021 12:46:36 +0530
Subject: [PATCH] rbd: check for peer site status

Do resync if the image is in unknow or in error
state.

Check for the current image state for up+stopped
or up+replaying and also all peer site status
should be un up+stopped to confirm that resyncing
is done and image can be promoted and used.

Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
---
 internal/rbd/mirror.go                      |  6 +++++
 internal/rbd/replicationcontrollerserver.go | 27 ++++++++++++++++-----
 2 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/internal/rbd/mirror.go b/internal/rbd/mirror.go
index d37b85333..2c231ac94 100644
--- a/internal/rbd/mirror.go
+++ b/internal/rbd/mirror.go
@@ -118,6 +118,12 @@ type imageMirrorStatus struct {
 	State       string `json:"state"` // rbd image state
 	Description string `json:"description"`
 	LastUpdate  string `json:"last_update"`
+	PeerSites   []struct {
+		SiteName    string `json:"site_name"`
+		State       string `json:"state"`
+		Description string `json:"description"`
+		LastUpdate  string `json:"last_update"`
+	} `json:"peer_sites"`
 }
 
 // FIXME: once https://github.com/ceph/go-ceph/issues/460 is fixed use go-ceph.
diff --git a/internal/rbd/replicationcontrollerserver.go b/internal/rbd/replicationcontrollerserver.go
index bf440c837..c68d88baf 100644
--- a/internal/rbd/replicationcontrollerserver.go
+++ b/internal/rbd/replicationcontrollerserver.go
@@ -20,6 +20,7 @@ import (
 	"context"
 	"errors"
 	"strconv"
+	"strings"
 
 	"github.com/ceph/ceph-csi/internal/util"
 
@@ -50,6 +51,9 @@ const (
 	// running and stopped means the image is not a target for replication from
 	// another cluster
 	upAndStopped imageMirroringState = "up+stopped"
+
+	// If the state is error means image need resync.
+	errorState imageMirroringState = "error"
 )
 
 const (
@@ -417,12 +421,6 @@ func (rs *ReplicationServer) ResyncVolume(ctx context.Context,
 		return nil, status.Error(codes.InvalidArgument, "image is in primary state")
 	}
 
-	err = rbdVol.resyncImage()
-	if err != nil {
-		util.ErrorLog(ctx, err.Error())
-		return nil, status.Error(codes.Internal, err.Error())
-	}
-	// TODO: check the image state and return its ready to use or not
 	mirrorStatus, err := rbdVol.getImageMirroingStatus()
 	if err != nil {
 		// the image gets recreated after issuing resync in that case return
@@ -439,8 +437,25 @@ func (rs *ReplicationServer) ResyncVolume(ctx context.Context,
 	ready := false
 	state := imageMirroringState(mirrorStatus.State)
 	if state == upAndStopped || state == upAndReplaying {
+		// Make sure the peer site image state is up and stopped
 		ready = true
+		for _, s := range mirrorStatus.PeerSites {
+			if imageMirroringState(s.State) != upAndStopped {
+				util.UsefulLog(ctx, "peer site name=%s mirroring state=%s, description=%s and lastUpdate=%s", s.SiteName, s.State, s.Description, s.LastUpdate)
+				ready = false
+			}
+		}
 	}
+
+	// resync only if the image is in error state
+	if strings.Contains(mirrorStatus.State, string(errorState)) {
+		err = rbdVol.resyncImage()
+		if err != nil {
+			util.ErrorLog(ctx, err.Error())
+			return nil, status.Error(codes.Internal, err.Error())
+		}
+	}
+
 	util.UsefulLog(ctx, "image mirroring state=%s, description=%s and lastUpdate=%s", mirrorStatus.State, mirrorStatus.Description, mirrorStatus.LastUpdate)
 	resp := &replication.ResyncVolumeResponse{
 		Ready: ready,