feat: switch away from websocket to stdout messaging for agent

This commit is contained in:
Aarnav Tale
2025-04-11 12:00:19 -04:00
parent 608fcacdb1
commit c5e6a6c407
19 changed files with 522 additions and 420 deletions
+10 -6
View File
@@ -8,6 +8,11 @@ import (
"github.com/tale/headplane/agent/internal/util"
)
type Register struct {
Type string
ID string
}
func main() {
log := util.GetLogger()
cfg, err := config.Load()
@@ -21,11 +26,10 @@ func main() {
agent.Connect()
defer agent.Shutdown()
ws, err := hpagent.NewSocket(agent, cfg)
if err != nil {
log.Fatal("Failed to create websocket: %s", err)
}
log.Msg(&Register{
Type: "register",
ID: agent.ID,
})
defer ws.StopListening()
ws.FollowMaster()
hpagent.FollowMaster(agent)
}
+3 -10
View File
@@ -8,8 +8,7 @@ type Config struct {
Hostname string
TSControlURL string
TSAuthKey string
HPControlURL string
HPAuthKey string
WorkDir string
}
const (
@@ -17,8 +16,7 @@ const (
HostnameEnv = "HEADPLANE_AGENT_HOSTNAME"
TSControlURLEnv = "HEADPLANE_AGENT_TS_SERVER"
TSAuthKeyEnv = "HEADPLANE_AGENT_TS_AUTHKEY"
HPControlURLEnv = "HEADPLANE_AGENT_HP_SERVER"
HPAuthKeyEnv = "HEADPLANE_AGENT_HP_AUTHKEY"
WorkDirEnv = "HEADPLANE_AGENT_WORK_DIR"
)
// Load reads the agent configuration from environment variables.
@@ -28,8 +26,7 @@ func Load() (*Config, error) {
Hostname: os.Getenv(HostnameEnv),
TSControlURL: os.Getenv(TSControlURLEnv),
TSAuthKey: os.Getenv(TSAuthKeyEnv),
HPControlURL: os.Getenv(HPControlURLEnv),
HPAuthKey: os.Getenv(HPAuthKeyEnv),
WorkDir: os.Getenv(WorkDirEnv),
}
if os.Getenv(DebugEnv) == "true" {
@@ -44,9 +41,5 @@ func Load() (*Config, error) {
return nil, err
}
if err := validateHPReady(c); err != nil {
return nil, err
}
return c, nil
}
+2 -26
View File
@@ -16,16 +16,12 @@ func validateRequired(config *Config) error {
return fmt.Errorf("%s is required", TSControlURLEnv)
}
if config.HPControlURL == "" {
return fmt.Errorf("%s is required", HPControlURLEnv)
}
if config.TSAuthKey == "" {
return fmt.Errorf("%s is required", TSAuthKeyEnv)
}
if config.HPAuthKey == "" {
return fmt.Errorf("%s is required", HPAuthKeyEnv)
if config.WorkDir == "" {
return fmt.Errorf("%s is required", WorkDirEnv)
}
return nil
@@ -50,23 +46,3 @@ func validateTSReady(config *Config) error {
return nil
}
// Pings the Headplane server to make sure it's up and running
func validateHPReady(config *Config) error {
testURL := config.HPControlURL
if strings.HasSuffix(testURL, "/") {
testURL = testURL[:len(testURL)-1]
}
testURL = fmt.Sprintf("%s/healthz", testURL)
resp, err := http.Get(testURL)
if err != nil {
return fmt.Errorf("Failed to connect to HP control server: %s", err)
}
if resp.StatusCode != 200 {
return fmt.Errorf("Failed to connect to HP control server: %s", resp.Status)
}
return nil
}
+26 -23
View File
@@ -1,9 +1,12 @@
package hpagent
import (
"bufio"
"encoding/json"
"os"
"sync"
"github.com/tale/headplane/agent/internal/tsnet"
"github.com/tale/headplane/agent/internal/util"
"tailscale.com/tailcfg"
)
@@ -13,30 +16,32 @@ type RecvMessage struct {
NodeIDs []string
}
// Starts listening for messages from the Headplane master
func (s *Socket) FollowMaster() {
log := util.GetLogger()
type SendMessage struct {
Type string
Data any
}
for {
_, message, err := s.ReadMessage()
if err != nil {
log.Error("Error reading message: %s", err)
return
}
// Starts listening for messages from stdin
func FollowMaster(agent *tsnet.TSAgent) {
log := util.GetLogger()
scanner := bufio.NewScanner(os.Stdin)
for scanner.Scan() {
line := scanner.Bytes()
var msg RecvMessage
err = json.Unmarshal(message, &msg)
err := json.Unmarshal(line, &msg)
if err != nil {
log.Error("Unable to unmarshal message: %s", err)
log.Debug("Full Error: %v", err)
continue
}
log.Debug("Recieved message from master: %v", message)
log.Debug("Recieved message from master: %v", line)
if len(msg.NodeIDs) == 0 {
log.Debug("Message recieved had no node IDs")
log.Debug("Full message: %s", message)
log.Debug("Full message: %s", line)
continue
}
@@ -49,7 +54,7 @@ func (s *Socket) FollowMaster() {
wg.Add(1)
go func(nodeID string) {
defer wg.Done()
result, err := s.Agent.GetStatusForPeer(nodeID)
result, err := agent.GetStatusForPeer(nodeID)
if err != nil {
log.Error("Unable to get status for node %s: %s", nodeID, err)
return
@@ -70,15 +75,13 @@ func (s *Socket) FollowMaster() {
// Send the results back to the Headplane master
log.Debug("Sending status back to master: %v", results)
err = s.SendStatus(results)
if err != nil {
log.Error("Error sending status: %s", err)
return
}
log.Msg(&SendMessage{
Type: "status",
Data: results,
})
}
if err := scanner.Err(); err != nil {
log.Fatal("Error reading from stdin: %s", err)
}
}
// Stops listening for messages from the Headplane master
func (s *Socket) StopListening() {
s.Close()
}
-11
View File
@@ -1,11 +0,0 @@
package hpagent
import (
"tailscale.com/tailcfg"
)
// Sends the status to the Headplane master
func (s *Socket) SendStatus(status map[string]*tailcfg.HostinfoView) error {
err := s.WriteJSON(status)
return err
}
-67
View File
@@ -1,67 +0,0 @@
package hpagent
import (
"fmt"
"net/http"
"net/url"
"github.com/gorilla/websocket"
"github.com/tale/headplane/agent/internal/config"
"github.com/tale/headplane/agent/internal/tsnet"
"github.com/tale/headplane/agent/internal/util"
)
type Socket struct {
*websocket.Conn
Agent *tsnet.TSAgent
}
// Creates a new websocket connection to the Headplane server.
func NewSocket(agent *tsnet.TSAgent, cfg *config.Config) (*Socket, error) {
log := util.GetLogger()
wsURL, err := httpToWs(cfg.HPControlURL)
if err != nil {
return nil, err
}
headers := http.Header{}
headers.Add("X-Headplane-Tailnet-ID", agent.ID)
auth := fmt.Sprintf("Bearer %s", cfg.HPAuthKey)
headers.Add("Authorization", auth)
log.Info("Dialing WebSocket with master: %s", wsURL)
ws, _, err := websocket.DefaultDialer.Dial(wsURL, headers)
if err != nil {
log.Debug("Failed to dial WebSocket: %s", err)
return nil, err
}
return &Socket{ws, agent}, nil
}
// We need to convert the control URL to a websocket URL
func httpToWs(controlURL string) (string, error) {
log := util.GetLogger()
u, err := url.Parse(controlURL)
if err != nil {
log.Debug("Failed to parse control URL: %s", err)
return "", err
}
if u.Scheme == "http" {
u.Scheme = "ws"
} else if u.Scheme == "https" {
u.Scheme = "wss"
} else {
return "", fmt.Errorf("unsupported scheme: %s", u.Scheme)
}
// We also need to append /_dial to the path
if u.Path[len(u.Path)-1] != '/' {
u.Path += "/"
}
u.Path += "_dial"
return u.String(), nil
}
+23 -3
View File
@@ -2,6 +2,9 @@ package tsnet
import (
"context"
"os"
"path/filepath"
"github.com/tale/headplane/agent/internal/config"
"github.com/tale/headplane/agent/internal/util"
"tailscale.com/client/tailscale"
@@ -17,15 +20,27 @@ type TSAgent struct {
// Creates a new tsnet agent and returns an instance of the server.
func NewAgent(cfg *config.Config) *TSAgent {
log := util.GetLogger()
dir, err := filepath.Abs(cfg.WorkDir)
if err != nil {
log.Fatal("Failed to get absolute path: %s", err)
}
if err := os.MkdirAll(dir, 0700); err != nil {
log.Fatal("Cannot create agent work directory: %s", err)
}
server := &tsnet.Server{
Dir: dir,
Hostname: cfg.Hostname,
ControlURL: cfg.TSControlURL,
AuthKey: cfg.TSAuthKey,
Logf: func(string, ...interface{}) {}, // Disabled by default
Logf: func(string, ...any) {}, // Disabled by default
UserLogf: log.Info,
}
if cfg.Debug {
log := util.GetLogger()
server.Logf = log.Debug
}
@@ -47,8 +62,13 @@ func (s *TSAgent) Connect() {
log.Fatal("Failed to initialize local Tailscale client: %s", err)
}
id, err := status.Self.PublicKey.MarshalText()
if err != nil {
log.Fatal("Failed to marshal public key: %s", err)
}
log.Info("Connected to Tailnet (PublicKey: %s)", status.Self.PublicKey)
s.ID = string(status.Self.ID)
s.ID = string(id)
}
// Shuts down the tsnet agent.
+88 -37
View File
@@ -1,66 +1,117 @@
package util
import (
"log"
"encoding/json"
"fmt"
"os"
"strings"
"sync"
"time"
)
type Logger struct {
debug *log.Logger
info *log.Logger
error *log.Logger
type LogLevel string
const (
LevelInfo LogLevel = "info"
LevelDebug LogLevel = "debug"
LevelError LogLevel = "error"
LevelFatal LogLevel = "fatal"
LevelMsg LogLevel = "msg"
)
type LogMessage struct {
Level LogLevel
Time string
Message any
}
var lock = &sync.Mutex{}
var logger *Logger
type Logger struct {
debugEnabled bool
encoder *json.Encoder
pool *sync.Pool
}
var logger = NewLogger()
func GetLogger() *Logger {
if logger == nil {
lock.Lock()
defer lock.Unlock()
if logger == nil {
logger = NewLogger()
}
}
return logger
}
func NewLogger() *Logger {
// Create a new Logger for stdout and stderr
// Errors still go to both stdout and stderr
enc := json.NewEncoder(os.Stdout)
enc.SetEscapeHTML(false)
return &Logger{
debug: nil,
info: log.New(os.Stdout, "[INFO] ", log.LstdFlags),
error: log.New(os.Stderr, "[ERROR] ", log.LstdFlags),
encoder: enc,
pool: &sync.Pool{
New: func() any {
return &LogMessage{}
},
},
}
}
func (logger *Logger) SetDebug(debug bool) {
if debug {
logger.Info("Enabling Debug logging for headplane-agent")
logger.Info("Be careful, this will spam a lot of information")
logger.debug = log.New(os.Stdout, "[DEBUG] ", log.LstdFlags)
} else {
logger.debug = nil
func (l *Logger) SetDebug(enabled bool) {
if enabled {
l.debugEnabled = true
l.Info("Enabling Debug logging for headplane-agent")
l.Info("Be careful, this will spam a lot of information")
}
}
func (logger *Logger) Info(fmt string, v ...any) {
logger.info.Printf(fmt, v...)
}
func (l *Logger) log(level LogLevel, format string, v ...any) {
msg := fmt.Sprintf(format, v...)
timestamp := time.Now().Format(time.RFC3339)
func (logger *Logger) Debug(fmt string, v ...any) {
if logger.debug != nil {
logger.debug.Printf(fmt, v...)
// Manually construct compact JSON line for performance
line := `{"Level":"` + string(level) +
`","Time":"` + timestamp +
`","Message":"` + escapeString(msg) + `"}` + "\n"
if level == LevelError || level == LevelFatal {
os.Stderr.WriteString(line)
}
// Always write to stdout but also write to stderr for errors
os.Stdout.WriteString(line)
if level == LevelFatal {
os.Exit(1)
}
}
func (logger *Logger) Error(fmt string, v ...any) {
logger.error.Printf(fmt, v...)
func (l *Logger) Debug(format string, v ...any) {
if l.debugEnabled {
l.log(LevelDebug, format, v...)
}
}
func (logger *Logger) Fatal(fmt string, v ...any) {
logger.error.Fatalf(fmt, v...)
func (l *Logger) Info(format string, v ...any) { l.log(LevelInfo, format, v...) }
func (l *Logger) Error(format string, v ...any) { l.log(LevelError, format, v...) }
func (l *Logger) Fatal(format string, v ...any) { l.log(LevelFatal, format, v...) }
func (l *Logger) Msg(obj any) {
entry := l.pool.Get().(*LogMessage)
defer l.pool.Put(entry)
entry.Level = LevelMsg
entry.Time = time.Now().Format(time.RFC3339)
entry.Message = obj
// Because the encoder is tied to STDOUT we get a message
_ = l.encoder.Encode(entry)
// Reset the entry for reuse
entry.Level = ""
entry.Time = ""
entry.Message = nil
}
func escapeString(s string) string {
replacer := strings.NewReplacer(
`"`, `\"`,
`\`, `\\`,
"\n", `\n`,
"\t", `\t`,
)
return replacer.Replace(s)
}