From 45216b1cccc951851517277662aaa327cbfecee4 Mon Sep 17 00:00:00 2001
From: Loic Reyreaud <loic@weaviate.io>
Date: Wed, 8 May 2024 09:07:28 +0200
Subject: [PATCH] make bootstrap return on successful notify

Signed-off-by: Loic Reyreaud <loic@weaviate.io>
---
 cluster/store/bootstrap.go | 24 +++++++++++++++++++++---
 cluster/store/store.go     | 11 +++++++----
 2 files changed, 28 insertions(+), 7 deletions(-)

diff --git a/cluster/store/bootstrap.go b/cluster/store/bootstrap.go
index 52943d1dfcb..87b2f6768cc 100644
--- a/cluster/store/bootstrap.go
+++ b/cluster/store/bootstrap.go
@@ -66,15 +66,33 @@ func (b *Bootstrapper) Do(ctx context.Context, serverPortMap map[string]int, lg
 			return ctx.Err()
 		case <-ticker.C:
 			// try to join an existing cluster
-			if leader, err := b.join(ctx, servers, voter); err == nil {
-				lg.WithField("leader", leader).Info("successfully joined cluster")
+			if leader, err := b.join(ctx, servers, voter); err != nil {
+				lg.WithFields(logrus.Fields{
+					"servers": servers,
+					"action":  "bootstrap",
+					"voter":   voter,
+				}).WithError(err).Warning("failed to join cluster, will notify next if voter")
+			} else {
+				lg.WithFields(logrus.Fields{
+					"action": "bootstrap",
+					"leader": leader,
+				}).Info("successfully joined cluster")
 				return nil
 			}
 
 			if voter {
 				// notify other servers about readiness of this node to be joined
 				if err := b.notify(ctx, servers); err != nil {
-					lg.WithField("servers", servers).WithError(err).Error("notify all peers")
+					lg.WithFields(logrus.Fields{
+						"action":  "bootstrap",
+						"servers": servers,
+					}).WithError(err).Error("notify all peers")
+				} else {
+					lg.WithFields(logrus.Fields{
+						"action":  "bootstrap",
+						"servers": servers,
+					}).Info("notified peers this node is ready to join as voter")
+					return nil
 				}
 			}
 
diff --git a/cluster/store/store.go b/cluster/store/store.go
index 2d3dae3496c..18bee976399 100644
--- a/cluster/store/store.go
+++ b/cluster/store/store.go
@@ -767,7 +767,6 @@ func (st *Store) Remove(id string) error {
 // Notify signals this Store that a node is ready for bootstrapping at the specified address.
 // Bootstrapping will be initiated once the number of known nodes reaches the expected level,
 // which includes this node.
-
 func (st *Store) Notify(id, addr string) (err error) {
 	if !st.open.Load() {
 		return ErrNotOpen
@@ -783,9 +782,10 @@ func (st *Store) Notify(id, addr string) (err error) {
 	st.candidates[id] = addr
 	if len(st.candidates) < st.bootstrapExpect {
 		st.log.WithFields(logrus.Fields{
+			"action": "bootstrap",
 			"expect": st.bootstrapExpect,
 			"got":    st.candidates,
-		}).Debug("number of candidates")
+		}).Debug("number of candidates lower than bootstrap expect param, stopping notify")
 		return nil
 	}
 	candidates := make([]raft.Server, 0, len(st.candidates))
@@ -800,11 +800,14 @@ func (st *Store) Notify(id, addr string) (err error) {
 		i++
 	}
 
-	st.log.WithField("candidates", candidates).Info("starting cluster bootstrapping")
+	st.log.WithFields(logrus.Fields{
+		"action":     "bootstrap",
+		"candidates": candidates,
+	}).Info("starting cluster bootstrapping")
 
 	fut := st.raft.BootstrapCluster(raft.Configuration{Servers: candidates})
 	if err := fut.Error(); err != nil {
-		st.log.WithError(err).Error("bootstrapping cluster")
+		st.log.WithField("action", "bootstrap").WithError(err).Error("could not bootstrapping cluster")
 		if !errors.Is(err, raft.ErrCantBootstrap) {
 			return err
 		}