@@ -16,7 +16,6 @@ import (
1616 "os"
1717 "os/user"
1818 "path/filepath"
19- "reflect"
2019 "sort"
2120 "strconv"
2221 "strings"
@@ -60,7 +59,7 @@ type Options struct {
6059 ReconnectingPTYTimeout time.Duration
6160 EnvironmentVariables map [string ]string
6261 Logger slog.Logger
63- AgentPorts map [int ]string
62+ IgnorePorts map [int ]string
6463 SSHMaxTimeout time.Duration
6564 TailnetListenPort uint16
6665}
@@ -76,7 +75,12 @@ type Client interface {
7675 PatchStartupLogs (ctx context.Context , req agentsdk.PatchStartupLogs ) error
7776}
7877
79- func New (options Options ) io.Closer {
78+ type Agent interface {
79+ HTTPDebug () http.Handler
80+ io.Closer
81+ }
82+
83+ func New (options Options ) Agent {
8084 if options .ReconnectingPTYTimeout == 0 {
8185 options .ReconnectingPTYTimeout = 5 * time .Minute
8286 }
@@ -112,7 +116,7 @@ func New(options Options) io.Closer {
112116 tempDir : options .TempDir ,
113117 lifecycleUpdate : make (chan struct {}, 1 ),
114118 lifecycleReported : make (chan codersdk.WorkspaceAgentLifecycle , 1 ),
115- ignorePorts : options .AgentPorts ,
119+ ignorePorts : options .IgnorePorts ,
116120 connStatsChan : make (chan * agentsdk.Stats , 1 ),
117121 sshMaxTimeout : options .SSHMaxTimeout ,
118122 }
@@ -161,7 +165,7 @@ type agent struct {
161165}
162166
163167func (a * agent ) init (ctx context.Context ) {
164- sshSrv , err := agentssh .NewServer (ctx , a .logger .Named ("ssh-server" ), a .sshMaxTimeout )
168+ sshSrv , err := agentssh .NewServer (ctx , a .logger .Named ("ssh-server" ), a .filesystem , a . sshMaxTimeout , "" )
165169 if err != nil {
166170 panic (err )
167171 }
@@ -648,6 +652,7 @@ func (a *agent) createTailnet(ctx context.Context, derpMap *tailcfg.DERPMap) (_
648652 }
649653 break
650654 }
655+ logger .Debug (ctx , "accepted conn" , slog .F ("remote" , conn .RemoteAddr ().String ()))
651656 wg .Add (1 )
652657 closed := make (chan struct {})
653658 go func () {
@@ -676,6 +681,7 @@ func (a *agent) createTailnet(ctx context.Context, derpMap *tailcfg.DERPMap) (_
676681 var msg codersdk.WorkspaceAgentReconnectingPTYInit
677682 err = json .Unmarshal (data , & msg )
678683 if err != nil {
684+ logger .Warn (ctx , "failed to unmarshal init" , slog .F ("raw" , data ))
679685 return
680686 }
681687 _ = a .handleReconnectingPTY (ctx , logger , msg , conn )
@@ -967,6 +973,7 @@ func (a *agent) handleReconnectingPTY(ctx context.Context, logger slog.Logger, m
967973
968974 connectionID := uuid .NewString ()
969975 logger = logger .With (slog .F ("id" , msg .ID ), slog .F ("connection_id" , connectionID ))
976+ logger .Debug (ctx , "starting handler" )
970977
971978 defer func () {
972979 if err := retErr ; err != nil {
@@ -1034,20 +1041,20 @@ func (a *agent) handleReconnectingPTY(ctx context.Context, logger slog.Logger, m
10341041 // 1. The timeout completed.
10351042 // 2. The parent context was canceled.
10361043 <- ctx .Done ()
1044+ logger .Debug (ctx , "context done" , slog .Error (ctx .Err ()))
10371045 _ = process .Kill ()
10381046 }()
1039- go func () {
1040- // If the process dies randomly, we should
1041- // close the pty.
1042- _ = process .Wait ()
1043- rpty .Close ()
1044- }()
1047+ // We don't need to separately monitor for the process exiting.
1048+ // When it exits, our ptty.OutputReader() will return EOF after
1049+ // reading all process output.
10451050 if err = a .trackConnGoroutine (func () {
10461051 buffer := make ([]byte , 1024 )
10471052 for {
1048- read , err := rpty .ptty .Output ().Read (buffer )
1053+ read , err := rpty .ptty .OutputReader ().Read (buffer )
10491054 if err != nil {
10501055 // When the PTY is closed, this is triggered.
1056+ // Error is typically a benign EOF, so only log for debugging.
1057+ logger .Debug (ctx , "unable to read pty output, command exited?" , slog .Error (err ))
10511058 break
10521059 }
10531060 part := buffer [:read ]
@@ -1059,8 +1066,15 @@ func (a *agent) handleReconnectingPTY(ctx context.Context, logger slog.Logger, m
10591066 break
10601067 }
10611068 rpty .activeConnsMutex .Lock ()
1062- for _ , conn := range rpty .activeConns {
1063- _ , _ = conn .Write (part )
1069+ for cid , conn := range rpty .activeConns {
1070+ _ , err = conn .Write (part )
1071+ if err != nil {
1072+ logger .Debug (ctx ,
1073+ "error writing to active conn" ,
1074+ slog .F ("other_conn_id" , cid ),
1075+ slog .Error (err ),
1076+ )
1077+ }
10641078 }
10651079 rpty .activeConnsMutex .Unlock ()
10661080 }
@@ -1138,7 +1152,7 @@ func (a *agent) handleReconnectingPTY(ctx context.Context, logger slog.Logger, m
11381152 logger .Warn (ctx , "read conn" , slog .Error (err ))
11391153 return nil
11401154 }
1141- _ , err = rpty .ptty .Input ().Write ([]byte (req .Data ))
1155+ _ , err = rpty .ptty .InputWriter ().Write ([]byte (req .Data ))
11421156 if err != nil {
11431157 logger .Warn (ctx , "write to pty" , slog .Error (err ))
11441158 return nil
@@ -1221,11 +1235,11 @@ func (a *agent) startReportingConnectionStats(ctx context.Context) {
12211235 // Convert from microseconds to milliseconds.
12221236 stats .ConnectionMedianLatencyMS /= 1000
12231237
1224- lastStat := a . latestStat . Load ()
1225- if lastStat != nil && reflect . DeepEqual ( lastStat , stats ) {
1226- a . logger . Info ( ctx , "skipping stat because nothing changed" )
1227- return
1228- }
1238+ // Collect agent metrics.
1239+ // Agent metrics are changing all the time, so there is no need to perform
1240+ // reflect.DeepEqual to see if stats should be transferred.
1241+ stats . Metrics = collectMetrics ()
1242+
12291243 a .latestStat .Store (stats )
12301244
12311245 select {
@@ -1267,6 +1281,27 @@ func (a *agent) isClosed() bool {
12671281 }
12681282}
12691283
1284+ func (a * agent ) HTTPDebug () http.Handler {
1285+ return http .HandlerFunc (func (w http.ResponseWriter , r * http.Request ) {
1286+ a .closeMutex .Lock ()
1287+ network := a .network
1288+ a .closeMutex .Unlock ()
1289+
1290+ if network == nil {
1291+ w .WriteHeader (http .StatusOK )
1292+ _ , _ = w .Write ([]byte ("network is not ready yet" ))
1293+ return
1294+ }
1295+
1296+ if r .URL .Path == "/debug/magicsock" {
1297+ network .MagicsockServeHTTPDebug (w , r )
1298+ } else {
1299+ w .WriteHeader (http .StatusNotFound )
1300+ _ , _ = w .Write ([]byte ("404 not found" ))
1301+ }
1302+ })
1303+ }
1304+
12701305func (a * agent ) Close () error {
12711306 a .closeMutex .Lock ()
12721307 defer a .closeMutex .Unlock ()
@@ -1358,7 +1393,7 @@ type reconnectingPTY struct {
13581393 circularBuffer * circbuf.Buffer
13591394 circularBufferMutex sync.RWMutex
13601395 timeout * time.Timer
1361- ptty pty.PTY
1396+ ptty pty.PTYCmd
13621397}
13631398
13641399// Close ends all connections to the reconnecting
0 commit comments