From 15ba78e620fd0c1e7f505cebd4032979676c7eb2 Mon Sep 17 00:00:00 2001
From: liuxiaolong <liuxiaolong@aiotlink.com>
Date: 星期二, 28 六月 2022 16:12:50 +0800
Subject: [PATCH] scale MaxUserEventSize to 90k
---
agent.go | 362 +++++++++++++++++++++++++--------------------------
1 files changed, 180 insertions(+), 182 deletions(-)
diff --git a/agent.go b/agent.go
index 9cea45c..822583b 100644
--- a/agent.go
+++ b/agent.go
@@ -18,53 +18,64 @@
import (
"context"
+ "encoding/json"
"errors"
"fmt"
"github.com/hashicorp/memberlist"
- "io/ioutil"
- "os"
+ "net"
"strconv"
- //"os"
- "strings"
"time"
- "github.com/hashicorp/serf/cmd/serf/command/agent"
- "github.com/hashicorp/serf/serf"
+ "basic.com/valib/serf.git/cmd/serf/command/agent"
+ "basic.com/valib/serf.git/serf"
//"github.com/apache/servicecomb-service-center/pkg/log"
- "log"
+ "basic.com/valib/logger.git"
)
// Agent warps the serf agent
type Agent struct {
*agent.Agent
- conf *Config
- readyCh chan struct{}
- errorCh chan error
+ ipc *agent.AgentIPC
+ conf *Config
+ readyCh chan struct{}
+ errorCh chan error
+ handleEv HandleEventFunc
+}
+
+//鐢ㄦ埛鑷畾涔変簨浠跺鐞�
+type HandleEventFunc func(event serf.Event)
+
+type NodeInfo struct {
+ ClusterID string `json:"clusterID"`
+ NodeID string `json:"nodeID"`
+ NodeAddress string `json:"nodeAddress"`
+ IsAlive int `json:"isAlive"`
}
// Create create serf agent with config
-func Create(conf *Config) (*Agent, error) {
+func Create(conf *Config, snapshotPath string) (*Agent, error) {
// config cover to serf config
- serfConf, err := conf.convertToSerf()
+ serfConf, err := conf.convertToSerf(snapshotPath)
if err != nil {
return nil, err
}
// create serf agent with serf config
- fmt.Println("conf.Config.EncryptKey:",conf.EncryptKey)
- serfAgent, err := agent.Create(conf.Config, serfConf, nil)
+ logger.Info("conf.Config.EncryptKey:", conf.EncryptKey)
+ serfAgent, err := agent.Create(conf.Config, serfConf, logger.GetLogFile())
if err != nil {
return nil, err
}
// Create the keyring
keyring, err := memberlist.NewKeyring(nil, []byte(conf.EncryptKey))
if err != nil {
- fmt.Printf("Failed to restore keyring: %s", err)
+ logger.Error("Failed to restore keyring: %s", err)
return nil, err
}
serfConf.MemberlistConfig.Keyring = keyring
- fmt.Printf("[INFO] agent: Restored keyring with %d keys from %s",
+
+ logger.Info("[INFO] agent: Restored keyring with %d keys from %s",
len(conf.EncryptKey), conf.EncryptKey)
return &Agent{
@@ -75,90 +86,108 @@
}, nil
}
-// Start agent
+func (a *Agent) RegisterHandleEventFunc(f HandleEventFunc) {
+ if f != nil {
+ a.handleEv = f
+ }
+}
+
+// Start agent and IPC
func (a *Agent) Start(ctx context.Context) {
+ a.RegisterEventHandler(a)
err := a.Agent.Start()
if err != nil {
- log.Println(err, "start serf agent failed")
+ logger.Error(err, "start serf agent failed")
a.errorCh <- err
return
}
- a.RegisterEventHandler(a)
+
+ ipc := a.startIPC()
+ a.ipc = ipc
+ go func() {
+ for {
+ select {
+ case <-ctx.Done():
+ a.Agent.Shutdown()
+ a.ipc.Shutdown()
+ return
+ default:
+ time.Sleep(1*time.Second)
+ }
+ }
+ }()
err = a.retryJoin(ctx)
if err != nil {
- log.Println(err, "start serf agent failed")
+ logger.Error(err, "start serf agent failed")
if err != ctx.Err() && a.errorCh != nil {
a.errorCh <- err
}
}
+
+ go a.BroadcastMemberlist(BroadcastInterval * time.Second)
}
+func (a *Agent) startIPC() *agent.AgentIPC {
+ // Parse the bind address information
+ bindIP, bindPort, err := a.conf.AddrParts(a.conf.BindAddr)
+ bindAddr := &net.TCPAddr{IP: net.ParseIP(bindIP), Port: bindPort}
+
+ // Setup the RPC listener
+ rpcListener, err := net.Listen("tcp", a.conf.RPCAddr)
+ if err != nil {
+ logger.Error("Error starting RPC listener:", err)
+ return nil
+ }
+ ipc := agent.NewAgentIPC(a.Agent, a.conf.RPCAuthKey, rpcListener, logger.GetLogFile(), nil)
+ logger.Debug("RPC addr:", a.conf.RPCAddr)
+ logger.Debug("Snapshot:", a.conf.SnapshotPath)
+ logger.Debug("Profile:", a.conf.Profile)
+ logger.Debug("Message Compression Enabled:", a.conf.EnableCompression)
+ logger.Debug("bindAddr:", bindAddr)
+ return ipc
+}
// HandleEvent Handles serf.EventMemberJoin events,
// which will wait for members to join until the number of group members is equal to "groupExpect"
// when the startup mode is "ModeCluster",
// used for logical grouping of serf nodes
func (a *Agent) HandleEvent(event serf.Event) {
-
- switch ev := event.(type) {
- case serf.UserEvent:
- fmt.Println(string(ev.Payload))
- var tmpstringslice []string
- tmpstringslice = append(tmpstringslice, string(ev.Payload))
- fmt.Println(tmpstringslice)
- results, err := DoExecute(tmpstringslice)
-
- for _, result := range results {
- fmt.Println(result, "results err: ", err)
- }
-
- case *serf.Query:
- //bak file and send resp
- filename, err := BakDbFile()
- if err != nil {
- fmt.Println("bak db file error!")
- return
- }
- fmt.Println(filename)
-
- filebuf, err := ioutil.ReadFile(filename)
- fmt.Println("filebuf: ", len(filebuf))
- if err != nil {
- fmt.Printf("file to []bytes error: %s\n", err)
- return
- }
-
- err = os.Remove(filename)
- if err != nil {
- fmt.Printf("remove file%s\n failed", filename)
- return
- }
-
- fmt.Println("query payload: ", len(ev.Payload))
- if query, ok := event.(*serf.Query); ok {
- if err := query.Respond(filebuf); err != nil {
- fmt.Printf("err: %s\n", err)
- return
- }
- }
-
- default:
- fmt.Printf("Unknown event type: %s\n", ev.EventType().String())
+ if a.handleEv != nil {
+ a.handleEv(event)
}
+}
- //if event.EventType() != serf.EventMemberJoin {
- // fmt.Printf("event.EventType() != serf.EventMemberJoin")
- // return
- //}
- //
- //if a.conf.Mode == ModeCluster {
- // if len(a.GroupMembers(a.conf.ClusterID)) < groupExpect {
- // fmt.Printf("len(a.GroupMembers(a.conf.ClusterID)) < groupExpect")
- // return
- // }
- //}
- //a.DeregisterEventHandler(a)
- //close(a.readyCh)
+
+func (a *Agent) BroadcastMemberlist(delay time.Duration) {
+ //serf := a.serf
+ serf := a.Agent.Serf()
+ mb := serf.LocalMember()
+ mblist := serf.Memberlist()
+ logger.Info("mb:", mb)
+
+ // copy local node
+ localNode := *mblist.LocalNode()
+ nodeID := a.conf.NodeName
+ nodeAddress := localNode.Address()
+ clusterID := mb.Tags[tagKeyClusterID]
+ isAlive := int(mb.Status)
+
+ message, _ := json.Marshal(NodeInfo{
+ clusterID,
+ nodeID,
+ nodeAddress,
+ isAlive,
+ })
+
+ // replace node address
+ localNode.Addr = net.ParseIP(BroadcastIP)
+ //localNode.Addr = net.IPv4(255,255,255,255)
+ localNode.Port = BroadcastPort
+ for {
+ // logger.Info("localNode: %v %v\n", nodeName, nodeAddress)
+ mblist.SendBestEffort(&localNode, []byte(message))
+ time.Sleep(delay)
+ }
}
// Ready Returns a channel that will be closed when serf is ready
@@ -174,8 +203,8 @@
// Stop serf agent
func (a *Agent) Stop() {
if a.errorCh != nil {
- a.Leave()
- a.Shutdown()
+ logger.Info("a.Shutdown()", a.Leave())
+ logger.Info("a.Shutdown()", a.Shutdown())
close(a.errorCh)
a.errorCh = nil
}
@@ -196,7 +225,7 @@
serfAgent := a.Agent.Serf()
if serfAgent != nil {
for _, member := range serfAgent.Members() {
- log.Printf("member = %s, clusterID = %s", member.Name, member.Tags[tagKeyClusterID])
+ logger.Info("member = %s, clusterID = %s", member.Name, member.Tags[tagKeyClusterID])
if member.Tags[tagKeyClusterID] == clusterID {
members = append(members, member)
}
@@ -241,7 +270,7 @@
func (a *Agent) retryJoin(ctx context.Context) (err error) {
if len(a.conf.RetryJoin) == 0 {
- log.Printf("retry join mumber %d", len(a.conf.RetryJoin))
+ logger.Error("retry join mumber %d", len(a.conf.RetryJoin))
return nil
}
@@ -249,13 +278,13 @@
attempt := 0
ticker := time.NewTicker(a.conf.RetryInterval)
for {
- log.Printf("serf: Joining cluster...(replay: %v)", a.conf.ReplayOnJoin)
+ logger.Info("serf: Joining cluster...(replay: %v)", a.conf.ReplayOnJoin)
var n int
// Try to join the specified serf nodes
n, err = a.Join(a.conf.RetryJoin, a.conf.ReplayOnJoin)
if err == nil {
- log.Printf("serf: Join completed. Synced with %d initial agents", n)
+ logger.Error("serf: Join completed. Synced with %d initial agents", n)
break
}
attempt++
@@ -265,7 +294,7 @@
// else agent will try to join other nodes until successful always
if a.conf.RetryMaxAttempts > 0 && attempt > a.conf.RetryMaxAttempts {
err = errors.New("serf: maximum retry join attempts made, exiting")
- log.Println(err, err.Error())
+ logger.Error(err, err.Error())
break
}
select {
@@ -281,136 +310,105 @@
return
}
-
-
-//GetDbFromCluster get the newest database after join cluster
-//dbPathWrite the path where to write after got a database,
-func (a *Agent)GetDbFromCluster(dbPathWrite string) {
- //members: get name of first member
- mbs := a.GroupMembers(a.conf.ClusterID)
- var specmembername string
- for _, m := range mbs {
- if m.Addr.String() != a.conf.BindAddr {
- specmembername = m.Name
- break
- }
- }
- fmt.Println(specmembername)
-
- //query: get db file.
- params := serf.QueryParam{
- FilterNodes: strings.Fields(specmembername),
- }
-
- resp, err := a.Query("getDatabase", []byte(""), ¶ms)
- if err == nil || !strings.Contains(err.Error(), "cannot contain") {
- fmt.Println("err: ", err)
- }
-
- go func() {
- respCh := resp.ResponseCh()
- for {
- select {
- case r := <-respCh:
- fmt.Println("x length is: ", len(r.Payload))
-
- // // byte to file.
- Dbconn.Close()
- Dbconn = nil
- err = ioutil.WriteFile(dbPathWrite, r.Payload, 0644)
- if err != nil {
- fmt.Println("query byte to file error!", err)
- }
- err := GetConn()
- if err != nil {
- fmt.Println("create db conn of test.db error: ", err)
- }
- return
- }
- }
- }()
-}
-
-//SyncSql boardcast sql to cluster
-func (a *Agent)SyncSql(sqlOp string) {
- // event : use to send command to operate db.
- err := a.UserEvent("SyncSql", []byte(sqlOp), false)
- if err == nil || !strings.Contains(err.Error(), "cannot contain") {
- fmt.Println("err: ", err)
- }
-}
-
//Init serf Init
-//web鍚庡彴鏀跺埌鍒涘缓闆嗙兢鐨勮姹傦紝
-func Init(clusterID string, password string, nodeID string) (*Agent, error) {
+func Init(clusterID string, password string, nodeID string, addrs []string, snapshotPath string, c *Config) (*Agent, error) {
+ agent, err := InitNode(clusterID, password, nodeID, snapshotPath, c)
+ if err != nil {
+ logger.Error("InitNode failed, error: %s", err)
+ return agent, err
+ }
+
+ err = agent.JoinByNodeAddrs(addrs)
+ if err != nil {
+ logger.Error("JoinByNodeIP failed, error: %s", err)
+ return agent, err
+ }
+
+ return agent, err
+}
+
+//InitNode web鍚庡彴鏀跺埌鍒涘缓闆嗙兢鐨勮姹傦紝
+func InitNode(clusterID string, password string, nodeID string, snapshotPath string, c *Config) (*Agent, error) {
conf := DefaultConfig()
- fmt.Println("clusterID:", clusterID, "password:", password, "nodeID:", nodeID)
- //conf.ClusterID = clusterID
+ conf.MergeConf(c)
+
+ logger.Info("clusterID:", clusterID, "password:", password, "nodeID:", nodeID)
+ conf.ClusterID = clusterID
conf.NodeName = nodeID
if password == "" {
conf.EncryptKey = DefaultEncryptKey
- }else{
+ } else {
if len(password) >= 16 {
password = password[:16]
- }else{
+ } else {
password = fmt.Sprintf("%016s", password)[:16]
//return nil, fmt.Errorf("error password")
}
conf.EncryptKey = password
}
- agent, err := Create(conf)
+ agent, err := Create(conf, snapshotPath)
if err != nil {
- fmt.Printf("create agent failed, error: %s", err)
+ logger.Error("create agent failed, error: %s", err)
return agent, err
}
- agent.Start(context.Background())
+ agent.Start(conf.Ctx)
//<- agent.readyCh
- go func() {
- agent.ShutdownCh()
- }()
+
time.Sleep(time.Second)
- fmt.Println("Stats:",agent.Agent.Serf().Stats())
- fmt.Println("EncryptionEnabled:",agent.Agent.Serf().EncryptionEnabled())
- fmt.Printf("create agent sucess!!")
+ logger.Info("Stats:", agent.Agent.Serf().Stats())
+ logger.Info("EncryptionEnabled:", agent.Agent.Serf().EncryptionEnabled())
+ logger.Info("create agent sucess!!")
return agent, nil
}
-func (a *Agent) JoinByNodeIP(ip string) error {
- n, err := a.Agent.Join([]string{ip + ":" + strconv.Itoa(DefaultBindPort)}, true)
- if err != nil || n == 0{
- a.Stop()
- fmt.Println("Stop node")
- return fmt.Errorf("Error Encrypt Key!")
+func (a *Agent) JoinByNodeAddrs(addrs []string) error {
+ var nodes []string
+
+ if len(addrs) == 0 {
+ return fmt.Errorf("No Nodes To Join!")
+ }
+ for _, addr := range addrs {
+ nodes = append(nodes, addr)
}
- return err
+ a.Agent.Join(nodes, true)
+
+ return nil
}
-type Node struct {
- clusterID string
- NodeID string
- IP string
- isAlive int //StatusNone:0, StatusAlive:1, StatusLeaving:2, StatusLeft:3, StatusFailed:4
-}
+//func (a *Agent) JoinByNodeIP(ips []string) error {
+// var nodes []string
+//
+// if len(ips) == 0 {
+// return fmt.Errorf("No Nodes To Join!")
+// }
+// for _, ip := range ips {
+// node := fmt.Sprintf("%s:%d", ip, DefaultBindPort)
+// nodes = append(nodes, node)
+// }
+//
+// n, err := a.Agent.Join(nodes, true)
+// if err != nil || n == 0 {
+// return fmt.Errorf("Error Encrypt Key!")
+// }
+//
+// return err
+//}
-func (a *Agent) GetNodes() (nodes []Node) {
- var node Node
- fmt.Println("a.conf.ClusterID:",a.conf.ClusterID)
+func (a *Agent) GetNodes() (nodes []NodeInfo) {
+ var node NodeInfo
+ logger.Info("a.conf.ClusterID:", a.conf.ClusterID)
mbs := a.GroupMembers(a.conf.ClusterID)
for _, mb := range mbs {
node.NodeID = mb.Name
- node.IP = mb.Addr.String()
- node.isAlive = int(mb.Status)
- node.clusterID = mb.Tags[tagKeyClusterID]
+ node.NodeAddress = mb.Addr.String() + ":" + strconv.Itoa(int(mb.Port))
+ node.IsAlive = int(mb.Status)
+ node.ClusterID = mb.Tags[tagKeyClusterID]
nodes = append(nodes, node)
}
return nodes
}
-
-
-
-
--
Gitblit v1.8.0