set cluster bind address consistently via the same mechanism

grafana · Dec 1, 2017 · 81eea5f · 81eea5f
1 parent ae3c5da
commit 81eea5f
Show file tree

Hide file tree

Showing 9 changed files with 30 additions and 34 deletions.
diff --git a/cluster/config.go b/cluster/config.go
@@ -17,13 +17,12 @@ var (
 	peersStr           string
 	mode               string
 	maxPrio            int
-	clusterPort        int
-	clusterHost        net.IP
-	clusterBindAddr    string
 	httpTimeout        time.Duration
 	minAvailableShards int
 
 	swimUseConfig               string
+	swimBindAddrStr             string
+	swimBindAddr                *net.TCPAddr
 	swimTCPTimeout              time.Duration
 	swimIndirectChecks          int
 	swimRetransmitMult          int
@@ -47,7 +46,6 @@ func ConfigSetup() {
 	clusterCfg := flag.NewFlagSet("cluster", flag.ExitOnError)
 	clusterCfg.StringVar(&ClusterName, "name", "metrictank", "Unique name of the cluster.")
 	clusterCfg.BoolVar(&primary, "primary-node", false, "the primary node writes data to cassandra. There should only be 1 primary node per shardGroup.")
-	clusterCfg.StringVar(&clusterBindAddr, "bind-addr", "0.0.0.0:7946", "TCP Address to listen on for cluster communication")
 	clusterCfg.StringVar(&peersStr, "peers", "", "TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances")
 	clusterCfg.StringVar(&mode, "mode", "single", "Operating mode of cluster. (single|multi)")
 	clusterCfg.DurationVar(&httpTimeout, "http-timeout", time.Second*60, "How long to wait before aborting http requests to cluster peers and returning a http 503 service unavailable")
@@ -57,6 +55,7 @@ func ConfigSetup() {
 
 	swimCfg := flag.NewFlagSet("swim", flag.ExitOnError)
 	swimCfg.StringVar(&swimUseConfig, "use-config", "default-lan", "config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. see https://godoc.org/github.com/hashicorp/memberlist#Config . Note all our swim settings correspond to default-lan")
+	swimCfg.StringVar(&swimBindAddrStr, "bind-addr", "0.0.0.0:7946", "binding TCP Address for UDP and TCP gossip")
 	swimCfg.DurationVar(&swimTCPTimeout, "tcp-timeout", 10*time.Second, "timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes")
 	swimCfg.IntVar(&swimIndirectChecks, "indirect-checks", 3, "number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails")
 	swimCfg.IntVar(&swimRetransmitMult, "retransmit-mult", 4, "multiplier for number of retransmissions for gossip messages. Retransmits = RetransmitMult * log(N+1)")
@@ -80,18 +79,16 @@ func ConfigProcess() {
 		log.Fatal(4, "CLU Config: invalid cluster operating mode")
 	}
 
-	addr, err := net.ResolveTCPAddr("tcp", clusterBindAddr)
+	var err error
+	swimBindAddr, err = net.ResolveTCPAddr("tcp", swimBindAddrStr)
 	if err != nil {
-		log.Fatal(4, "CLU Config: bind-addr is not a valid TCP address: %s", err.Error())
+		log.Fatal(4, "CLU Config: swim-bind-addr is not a valid TCP address: %s", err.Error())
 	}
 
 	if httpTimeout == 0 {
 		log.Fatal(4, "CLU Config: http-timeout must be a non-zero duration string like 60s")
 	}
 
-	clusterHost = addr.IP
-	clusterPort = addr.Port
-
 	Mode = ModeType(mode)
 
 	client = http.Client{

diff --git a/cluster/manager.go b/cluster/manager.go
@@ -3,7 +3,6 @@ package cluster
 import (
 	"crypto/sha256"
 	"encoding/json"
-	"net"
 	"strings"
 	"sync"
 	"time"
@@ -83,6 +82,9 @@ func NewMemberlistManager(thisNode Node) *MemberlistManager {
 	switch swimUseConfig {
 	case "none":
 		mgr.cfg = memberlist.DefaultLANConfig() // use this as base so that the other settings have proper defaults
+		mgr.cfg.BindPort = swimBindAddr.Port
+		mgr.cfg.BindAddr = swimBindAddr.IP.String()
+		mgr.cfg.AdvertisePort = swimBindAddr.Port
 		mgr.cfg.TCPTimeout = swimTCPTimeout
 		mgr.cfg.IndirectChecks = swimIndirectChecks
 		mgr.cfg.RetransmitMult = swimRetransmitMult
@@ -107,9 +109,6 @@ func NewMemberlistManager(thisNode Node) *MemberlistManager {
 	default:
 		panic("invalid swimUseConfig. should already have been validated")
 	}
-	mgr.cfg.BindPort = clusterPort
-	mgr.cfg.BindAddr = clusterHost.String()
-	mgr.cfg.AdvertisePort = clusterPort
 	mgr.cfg.Events = mgr
 	mgr.cfg.Delegate = mgr
 	h := sha256.New()

diff --git a/docker/docker-chaos/metrictank.ini b/docker/docker-chaos/metrictank.ini
@@ -195,8 +195,6 @@ name = metrictank
 primary-node = true
 # maximum priority before a node should be considered not-ready.
 max-priority = 10
-# the TCP/UDP address to listen on for the gossip protocol.
-bind-addr = 0.0.0.0:7946
 # TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances.
 # If no port is specified, it is assumed the other nodes are using the same port this node is listening on.
 peers =
@@ -210,14 +208,16 @@ http-timeout = 60s
 ## SWIM clustering settings ##
 # only relevant when using cluster mode 'multi'
 # for more details, see https://godoc.org/github.com/hashicorp/memberlist#Config
-# all values correspond literally to the memberlist.Config options
+# all values correspond literally to the memberlist.Config options except where noted
 [swim]
 # config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. Note all our swim settings correspond to default-lan")
 # see:
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLANConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLocalConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultWANConfig
 use-config = default-lan
+# binding TCP Address for UDP and TCP gossip (full ip/dns:port combo unlike memberlist.Config)
+bind-addr = 0.0.0.0:7946
 # timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes
 tcp-timeout = 10s
 # number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails

diff --git a/docker/docker-cluster/metrictank.ini b/docker/docker-cluster/metrictank.ini
@@ -195,8 +195,6 @@ name = metrictank
 primary-node = true
 # maximum priority before a node should be considered not-ready.
 max-priority = 10
-# the TCP/UDP address to listen on for the gossip protocol.
-bind-addr = 0.0.0.0:7946
 # TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances.
 # If no port is specified, it is assumed the other nodes are using the same port this node is listening on.
 peers =
@@ -210,14 +208,16 @@ http-timeout = 60s
 ## SWIM clustering settings ##
 # only relevant when using cluster mode 'multi'
 # for more details, see https://godoc.org/github.com/hashicorp/memberlist#Config
-# all values correspond literally to the memberlist.Config options
+# all values correspond literally to the memberlist.Config options except where noted
 [swim]
 # config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. Note all our swim settings correspond to default-lan")
 # see:
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLANConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLocalConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultWANConfig
 use-config = default-lan
+# binding TCP Address for UDP and TCP gossip (full ip/dns:port combo unlike memberlist.Config)
+bind-addr = 0.0.0.0:7946
 # timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes
 tcp-timeout = 10s
 # number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails

diff --git a/docker/docker-dev-custom-cfg-kafka/metrictank.ini b/docker/docker-dev-custom-cfg-kafka/metrictank.ini
@@ -195,8 +195,6 @@ name = metrictank
 primary-node = true
 # maximum priority before a node should be considered not-ready.
 max-priority = 10
-# the TCP/UDP address to listen on for the gossip protocol.
-bind-addr = 0.0.0.0:7946
 # TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances.
 # If no port is specified, it is assumed the other nodes are using the same port this node is listening on.
 peers =
@@ -210,14 +208,16 @@ http-timeout = 60s
 ## SWIM clustering settings ##
 # only relevant when using cluster mode 'multi'
 # for more details, see https://godoc.org/github.com/hashicorp/memberlist#Config
-# all values correspond literally to the memberlist.Config options
+# all values correspond literally to the memberlist.Config options except where noted
 [swim]
 # config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. Note all our swim settings correspond to default-lan")
 # see:
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLANConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLocalConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultWANConfig
 use-config = default-lan
+# binding TCP Address for UDP and TCP gossip (full ip/dns:port combo unlike memberlist.Config)
+bind-addr = 0.0.0.0:7946
 # timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes
 tcp-timeout = 10s
 # number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails

diff --git a/docs/config.md b/docs/config.md
@@ -244,8 +244,6 @@ name = metrictank
 primary-node = true
 # maximum priority before a node should be considered not-ready.
 max-priority = 10
-# the TCP/UDP address to listen on for the gossip protocol.
-bind-addr = 0.0.0.0:7946
 # TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances.
 # If no port is specified, it is assumed the other nodes are using the same port this node is listening on.
 peers =
@@ -262,14 +260,16 @@ http-timeout = 60s
 ```
 # only relevant when using cluster mode 'multi'
 # for more details, see https://godoc.org/github.com/hashicorp/memberlist#Config
-# all values correspond literally to the memberlist.Config options
+# all values correspond literally to the memberlist.Config options except where noted
 [swim]
 # config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. Note all our swim settings correspond to default-lan")
 # see:
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLANConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLocalConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultWANConfig
 use-config = default-lan
+# binding TCP Address for UDP and TCP gossip (full ip/dns:port combo unlike memberlist.Config)
+bind-addr = 0.0.0.0:7946
 # timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes
 tcp-timeout = 10s
 # number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails

diff --git a/metrictank-sample.ini b/metrictank-sample.ini
@@ -198,8 +198,6 @@ name = metrictank
 primary-node = true
 # maximum priority before a node should be considered not-ready.
 max-priority = 10
-# the TCP/UDP address to listen on for the gossip protocol.
-bind-addr = 0.0.0.0:7946
 # TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances.
 # If no port is specified, it is assumed the other nodes are using the same port this node is listening on.
 peers =
@@ -213,14 +211,16 @@ http-timeout = 60s
 ## SWIM clustering settings ##
 # only relevant when using cluster mode 'multi'
 # for more details, see https://godoc.org/github.com/hashicorp/memberlist#Config
-# all values correspond literally to the memberlist.Config options
+# all values correspond literally to the memberlist.Config options except where noted
 [swim]
 # config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. Note all our swim settings correspond to default-lan")
 # see:
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLANConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLocalConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultWANConfig
 use-config = default-lan
+# binding TCP Address for UDP and TCP gossip (full ip/dns:port combo unlike memberlist.Config)
+bind-addr = 0.0.0.0:7946
 # timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes
 tcp-timeout = 10s
 # number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails

diff --git a/scripts/config/metrictank-docker.ini b/scripts/config/metrictank-docker.ini
@@ -195,8 +195,6 @@ name = metrictank
 primary-node = true
 # maximum priority before a node should be considered not-ready.
 max-priority = 10
-# the TCP/UDP address to listen on for the gossip protocol.
-bind-addr = 0.0.0.0:7946
 # TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances.
 # If no port is specified, it is assumed the other nodes are using the same port this node is listening on.
 peers =
@@ -210,14 +208,16 @@ http-timeout = 60s
 ## SWIM clustering settings ##
 # only relevant when using cluster mode 'multi'
 # for more details, see https://godoc.org/github.com/hashicorp/memberlist#Config
-# all values correspond literally to the memberlist.Config options
+# all values correspond literally to the memberlist.Config options except where noted
 [swim]
 # config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. Note all our swim settings correspond to default-lan")
 # see:
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLANConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLocalConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultWANConfig
 use-config = default-lan
+# binding TCP Address for UDP and TCP gossip (full ip/dns:port combo unlike memberlist.Config)
+bind-addr = 0.0.0.0:7946
 # timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes
 tcp-timeout = 10s
 # number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails

diff --git a/scripts/config/metrictank-package.ini b/scripts/config/metrictank-package.ini
@@ -195,8 +195,6 @@ name = metrictank
 primary-node = true
 # maximum priority before a node should be considered not-ready.
 max-priority = 10
-# the TCP/UDP address to listen on for the gossip protocol.
-bind-addr = 0.0.0.0:7946
 # TCP addresses of other nodes, comma separated. use this if you shard your data and want to query other instances.
 # If no port is specified, it is assumed the other nodes are using the same port this node is listening on.
 peers =
@@ -210,14 +208,16 @@ http-timeout = 60s
 ## SWIM clustering settings ##
 # only relevant when using cluster mode 'multi'
 # for more details, see https://godoc.org/github.com/hashicorp/memberlist#Config
-# all values correspond literally to the memberlist.Config options
+# all values correspond literally to the memberlist.Config options except where noted
 [swim]
 # config setting to use. If set, will override all other swim settings. Use none|default-lan|default-local|default-wan. Note all our swim settings correspond to default-lan")
 # see:
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLANConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultLocalConfig
 # * https://godoc.org/github.com/hashicorp/memberlist#DefaultWANConfig
 use-config = default-lan
+# binding TCP Address for UDP and TCP gossip (full ip/dns:port combo unlike memberlist.Config)
+bind-addr = 0.0.0.0:7946
 # timeout for establishing a stream connection with peers for a full state sync, and for stream reads and writes
 tcp-timeout = 10s
 # number of nodes that will be asked to perform an indirect probe of a node in the case a direct probe fails