@@ -122,6 +122,7 @@ func New(options Options) io.Closer {
122122logDir :options .LogDir ,
123123tempDir :options .TempDir ,
124124lifecycleUpdate :make (chan struct {},1 ),
125+ lifecycleReported :make (chan codersdk.WorkspaceAgentLifecycle ,1 ),
125126connStatsChan :make (chan * agentsdk.Stats ,1 ),
126127}
127128a .init (ctx )
@@ -150,9 +151,10 @@ type agent struct {
150151sessionToken atomic.Pointer [string ]
151152sshServer * ssh.Server
152153
153- lifecycleUpdate chan struct {}
154- lifecycleMu sync.Mutex // Protects following.
155- lifecycleState codersdk.WorkspaceAgentLifecycle
154+ lifecycleUpdate chan struct {}
155+ lifecycleReported chan codersdk.WorkspaceAgentLifecycle
156+ lifecycleMu sync.RWMutex // Protects following.
157+ lifecycleState codersdk.WorkspaceAgentLifecycle
156158
157159network * tailnet.Conn
158160connStatsChan chan * agentsdk.Stats
@@ -205,9 +207,9 @@ func (a *agent) reportLifecycleLoop(ctx context.Context) {
205207}
206208
207209for r := retry .New (time .Second ,15 * time .Second );r .Wait (ctx ); {
208- a .lifecycleMu .Lock ()
210+ a .lifecycleMu .RLock ()
209211state := a .lifecycleState
210- a .lifecycleMu .Unlock ()
212+ a .lifecycleMu .RUnlock ()
211213
212214if state == lastReported {
213215break
@@ -220,6 +222,11 @@ func (a *agent) reportLifecycleLoop(ctx context.Context) {
220222})
221223if err == nil {
222224lastReported = state
225+ select {
226+ case a .lifecycleReported <- state :
227+ case <- a .lifecycleReported :
228+ a .lifecycleReported <- state
229+ }
223230break
224231}
225232if xerrors .Is (err ,context .Canceled )|| xerrors .Is (err ,context .DeadlineExceeded ) {
@@ -231,13 +238,20 @@ func (a *agent) reportLifecycleLoop(ctx context.Context) {
231238}
232239}
233240
241+ // setLifecycle sets the lifecycle state and notifies the lifecycle loop.
242+ // The state is only updated if it's a valid state transition.
234243func (a * agent )setLifecycle (ctx context.Context ,state codersdk.WorkspaceAgentLifecycle ) {
235244a .lifecycleMu .Lock ()
236- defer a .lifecycleMu .Unlock ()
237-
238- a .logger .Debug (ctx ,"set lifecycle state" ,slog .F ("state" ,state ),slog .F ("previous" ,a .lifecycleState ))
239-
245+ lastState := a .lifecycleState
246+ if slices .Index (codersdk .WorkspaceAgentLifecycleOrder ,lastState )> slices .Index (codersdk .WorkspaceAgentLifecycleOrder ,state ) {
247+ a .logger .Warn (ctx ,"attempted to set lifecycle state to a previous state" ,slog .F ("last" ,lastState ),slog .F ("state" ,state ))
248+ a .lifecycleMu .Unlock ()
249+ return
250+ }
240251a .lifecycleState = state
252+ a .logger .Debug (ctx ,"set lifecycle state" ,slog .F ("state" ,state ),slog .F ("last" ,lastState ))
253+ a .lifecycleMu .Unlock ()
254+
241255select {
242256case a .lifecycleUpdate <- struct {}{}:
243257default :
@@ -297,9 +311,10 @@ func (a *agent) run(ctx context.Context) error {
297311}
298312}
299313
314+ lifecycleState := codersdk .WorkspaceAgentLifecycleReady
300315scriptDone := make (chan error ,1 )
301316scriptStart := time .Now ()
302- err : =a .trackConnGoroutine (func () {
317+ err = a .trackConnGoroutine (func () {
303318defer close (scriptDone )
304319scriptDone <- a .runStartupScript (ctx ,metadata .StartupScript )
305320})
@@ -327,16 +342,17 @@ func (a *agent) run(ctx context.Context) error {
327342if errors .Is (err ,context .Canceled ) {
328343return
329344}
330- execTime := time .Since (scriptStart )
331- lifecycleStatus := codersdk .WorkspaceAgentLifecycleReady
332- if err != nil {
333- a .logger .Warn (ctx ,"startup script failed" ,slog .F ("execution_time" ,execTime ),slog .Error (err ))
334- lifecycleStatus = codersdk .WorkspaceAgentLifecycleStartError
335- }else {
336- a .logger .Info (ctx ,"startup script completed" ,slog .F ("execution_time" ,execTime ))
345+ // Only log if there was a startup script.
346+ if metadata .StartupScript != "" {
347+ execTime := time .Since (scriptStart )
348+ if err != nil {
349+ a .logger .Warn (ctx ,"startup script failed" ,slog .F ("execution_time" ,execTime ),slog .Error (err ))
350+ lifecycleState = codersdk .WorkspaceAgentLifecycleStartError
351+ }else {
352+ a .logger .Info (ctx ,"startup script completed" ,slog .F ("execution_time" ,execTime ))
353+ }
337354}
338-
339- a .setLifecycle (ctx ,lifecycleStatus )
355+ a .setLifecycle (ctx ,lifecycleState )
340356}()
341357}
342358
@@ -604,14 +620,22 @@ func (a *agent) runCoordinator(ctx context.Context, network *tailnet.Conn) error
604620}
605621
606622func (a * agent )runStartupScript (ctx context.Context ,script string )error {
623+ return a .runScript (ctx ,"startup" ,script )
624+ }
625+
626+ func (a * agent )runShutdownScript (ctx context.Context ,script string )error {
627+ return a .runScript (ctx ,"shutdown" ,script )
628+ }
629+
630+ func (a * agent )runScript (ctx context.Context ,lifecycle ,script string )error {
607631if script == "" {
608632return nil
609633}
610634
611- a .logger .Info (ctx ,"runningstartup script" ,slog .F ("script" ,script ))
612- writer ,err := a .filesystem .OpenFile (filepath .Join (a .logDir ,"coder-startup -script.log" ),os .O_CREATE | os .O_RDWR ,0o600 )
635+ a .logger .Info (ctx ,"running script" , slog . F ( "lifecycle" , lifecycle ) ,slog .F ("script" ,script ))
636+ writer ,err := a .filesystem .OpenFile (filepath .Join (a .logDir ,fmt . Sprintf ( "coder-%s -script.log" , lifecycle ) ),os .O_CREATE | os .O_RDWR ,0o600 )
613637if err != nil {
614- return xerrors .Errorf ("openstartup script log file: %w" ,err )
638+ return xerrors .Errorf ("open%s script log file: %w" , lifecycle ,err )
615639}
616640defer func () {
617641_ = writer .Close ()
@@ -772,7 +796,7 @@ func (a *agent) createCommand(ctx context.Context, rawCommand string, env []stri
772796
773797rawMetadata := a .metadata .Load ()
774798if rawMetadata == nil {
775- return nil ,xerrors .Errorf ("no metadata was provided: %w" , err )
799+ return nil ,xerrors .Errorf ("no metadata was provided" )
776800}
777801metadata ,valid := rawMetadata .(agentsdk.Metadata )
778802if ! valid {
@@ -1296,13 +1320,73 @@ func (a *agent) Close() error {
12961320if a .isClosed () {
12971321return nil
12981322}
1323+
1324+ ctx := context .Background ()
1325+ a .setLifecycle (ctx ,codersdk .WorkspaceAgentLifecycleShuttingDown )
1326+
1327+ lifecycleState := codersdk .WorkspaceAgentLifecycleOff
1328+ if metadata ,ok := a .metadata .Load ().(agentsdk.Metadata );ok && metadata .ShutdownScript != "" {
1329+ scriptDone := make (chan error ,1 )
1330+ scriptStart := time .Now ()
1331+ go func () {
1332+ defer close (scriptDone )
1333+ scriptDone <- a .runShutdownScript (ctx ,metadata .ShutdownScript )
1334+ }()
1335+
1336+ var timeout <- chan time.Time
1337+ // If timeout is zero, an older version of the coder
1338+ // provider was used. Otherwise a timeout is always > 0.
1339+ if metadata .ShutdownScriptTimeout > 0 {
1340+ t := time .NewTimer (metadata .ShutdownScriptTimeout )
1341+ defer t .Stop ()
1342+ timeout = t .C
1343+ }
1344+
1345+ var err error
1346+ select {
1347+ case err = <- scriptDone :
1348+ case <- timeout :
1349+ a .logger .Warn (ctx ,"shutdown script timed out" )
1350+ a .setLifecycle (ctx ,codersdk .WorkspaceAgentLifecycleShutdownTimeout )
1351+ err = <- scriptDone // The script can still complete after a timeout.
1352+ }
1353+ execTime := time .Since (scriptStart )
1354+ if err != nil {
1355+ a .logger .Warn (ctx ,"shutdown script failed" ,slog .F ("execution_time" ,execTime ),slog .Error (err ))
1356+ lifecycleState = codersdk .WorkspaceAgentLifecycleShutdownError
1357+ }else {
1358+ a .logger .Info (ctx ,"shutdown script completed" ,slog .F ("execution_time" ,execTime ))
1359+ }
1360+ }
1361+
1362+ // Set final state and wait for it to be reported because context
1363+ // cancellation will stop the report loop.
1364+ a .setLifecycle (ctx ,lifecycleState )
1365+
1366+ // Wait for the lifecycle to be reported, but don't wait forever so
1367+ // that we don't break user expectations.
1368+ ctx ,cancel := context .WithTimeout (ctx ,5 * time .Second )
1369+ defer cancel ()
1370+ lifecycleWaitLoop:
1371+ for {
1372+ select {
1373+ case <- ctx .Done ():
1374+ break lifecycleWaitLoop
1375+ case s := <- a .lifecycleReported :
1376+ if s == lifecycleState {
1377+ break lifecycleWaitLoop
1378+ }
1379+ }
1380+ }
1381+
12991382close (a .closed )
13001383a .closeCancel ()
1384+ _ = a .sshServer .Close ()
13011385if a .network != nil {
13021386_ = a .network .Close ()
13031387}
1304- _ = a .sshServer .Close ()
13051388a .connCloseWait .Wait ()
1389+
13061390return nil
13071391}
13081392