Merge pull request #3172 from weaveworks/issues/3170-fix-upgrade

weave-kube reclaim: check for upgrade case

prog/kube-peers/main.go

@@ -182,7 +182,7 @@ func main() {
 		logLevel    string
 	)
 	flag.BoolVar(&justReclaim, "reclaim", false, "reclaim IP space from dead peers")
-	flag.BoolVar(&justCheck, "check-peer-exists", false, "return success if peer name is stored in annotation")
+	flag.BoolVar(&justCheck, "check-peer-new", false, "return success if peer name is not stored in annotation")
 	flag.StringVar(&peerName, "peer-name", "unknown", "name of this Weave Net peer")
 	flag.StringVar(&nodeName, "node-name", "unknown", "name of this Kubernetes node")
 	flag.StringVar(&logLevel, "log-level", "info", "logging level (debug, info, warning, error)")
@@ -204,9 +204,9 @@ func main() {
 			common.Log.Fatalf("[kube-peers] Could not check peer list: %v", err)
 		}
 		if exists {
-			os.Exit(0)
-		} else {
 			os.Exit(9)
+		} else {
+			os.Exit(0)
 		}
 	}
 	peers, err := getKubePeers(c)

prog/weave-kube/launch.sh

@@ -95,16 +95,23 @@ if [ -z "$PEERNAME" ]; then
     exit 2
 fi
 
-# If this peer name is not stored in the list, either we are a
-# brand-new peer or we were removed by another peer while temporarily
-# absent. In order to avoid a CRDT clash for the latter case, clean up
-if ! /home/weave/kube-peers -check-peer-exists -peer-name="$PEERNAME" -log-level=debug ; then
-    if [ -f ${DB_PREFIX}data.db ]; then
-        echo "Peer not in list; removing persisted data" >&2
-        rm -f ${DB_PREFIX}data.db
+# Check if peer reclaim is in use - we don't want to delete persisted data on an upgrade
+if /usr/bin/weaveutil get-db-flag "$DB_PREFIX" peer-reclaim >/dev/null ; then
+    # If this peer name is not stored in the list, either we are a
+    # brand-new peer or we were removed by another peer while temporarily
+    # absent. In order to avoid a CRDT clash for the latter case, clean up
+    if /home/weave/kube-peers -check-peer-new -peer-name="$PEERNAME" -log-level=debug ; then
+        if [ -f ${DB_PREFIX}data.db ]; then
+            echo "Peer not in list; removing persisted data" >&2
+            rm -f ${DB_PREFIX}data.db
+        fi
     fi
 fi
 
+# flag that peer reclaim is now in use - note we have to do this before
+# weaver is running as the DB can only be opened by one process at a time
+/usr/bin/weaveutil set-db-flag "$DB_PREFIX" peer-reclaim ok
+
 post_start_actions() {
     # Wait for weave process to become responsive
     while true ; do

prog/weaveutil/db_flag.go

@@ -0,0 +1,54 @@
+// Get/set arbitrary flags in the Weave Net persistence DB
+package main
+
+import (
+	"fmt"
+
+	"github.com/weaveworks/weave/db"
+)
+
+const dbFlagPrefix = "flag:"
+
+func getDBFlag(args []string) error {
+	if len(args) != 2 {
+		cmdUsage("check-db-flag", "<db-prefix> <flag-name>")
+	}
+	dbPrefix := args[0]
+	flagName := args[1]
+
+	d, err := db.NewBoltDBReadOnly(dbPrefix)
+	if err != nil {
+		return err
+	}
+	defer d.Close()
+	var value string
+	nameFound, err := d.Load(dbFlagPrefix+flagName, &value)
+	if err != nil {
+		return err
+	}
+	if !nameFound {
+		return fmt.Errorf("")
+	}
+	fmt.Print(value)
+	return nil
+}
+
+func setDBFlag(args []string) error {
+	if len(args) != 3 {
+		cmdUsage("set-db-flag", "<db-prefix> <flag-name> <flag-value>")
+	}
+	dbPrefix := args[0]
+	flagName := args[1]
+	value := args[2]
+
+	d, err := db.NewBoltDB(dbPrefix)
+	if err != nil {
+		return err
+	}
+	defer d.Close()
+	err = d.Save(dbFlagPrefix+flagName, value)
+	if err != nil {
+		return err
+	}
+	return nil
+}

prog/weaveutil/main.go

@@ -50,6 +50,8 @@ func init() {
 		"swarm-manager-peers":      swarmManagerPeers,
 		"is-docker-plugin-enabled": isDockerPluginEnabled,
 		"rewrite-etc-hosts":        rewriteEtcHosts,
+		"get-db-flag":              getDBFlag,
+		"set-db-flag":              setDBFlag,
 	}
 }
 
@@ -83,7 +85,9 @@ func main() {
 		os.Exit(1)
 	}
 	if err := cmd(os.Args[2:]); err != nil {
-		fmt.Fprintln(os.Stderr, err)
+		if err.Error() != "" {
+			fmt.Fprintln(os.Stderr, err)
+		}
 		os.Exit(1)
 	}
 }