From 41e267a19e8378e2ad2b2df73c3b4bda24f5b35d Mon Sep 17 00:00:00 2001
From: 6543 <6543@obermui.de>
Date: Tue, 31 Mar 2026 18:53:56 +0200
Subject: [PATCH] Fix agent reusing old shutdownContext on recontect

---
 agent/runner.go         | 14 +++++++++++---
 cmd/agent/core/agent.go | 33 +++++++++------------------------
 2 files changed, 20 insertions(+), 27 deletions(-)

diff --git a/agent/runner.go b/agent/runner.go
index 4220427378f..7e5d19464c9 100644
--- a/agent/runner.go
+++ b/agent/runner.go
@@ -33,6 +33,8 @@ import (
 	"go.woodpecker-ci.org/woodpecker/v3/shared/utils"
 )
 
+const shutdownTimeout = time.Second * 5
+
 type Runner struct {
 	client   rpc.Peer
 	filter   rpc.Filter
@@ -51,15 +53,19 @@ func NewRunner(workEngine rpc.Peer, f rpc.Filter, h string, state *State, backen
 	}
 }
 
+func GetShutdownContext() (context.Context, context.CancelFunc) {
+	return context.WithTimeout(context.Background(), shutdownTimeout)
+}
+
 // TODO: refactor this big function into subfunctions in it's own subpackage
 
 // Run executes a workflow using a backend, tracks its state and reports the state back to the server.
-func (r *Runner) Run(runnerCtx, shutdownCtx context.Context) error {
+func (r *Runner) Run(runnerCtx context.Context) error {
 	log.Debug().Msg("request next execution")
 
 	// Preserve metadata AND cancellation from runnerCtx.
 	meta, _ := metadata.FromOutgoingContext(runnerCtx)
-	ctxMeta := metadata.NewOutgoingContext(shutdownCtx, meta)
+	ctxMeta := metadata.NewOutgoingContext(runnerCtx, meta)
 
 	// Fetch next workflow from the queue
 	workflow, err := r.client.Next(runnerCtx, r.filter)
@@ -192,8 +198,10 @@ func (r *Runner) Run(runnerCtx, shutdownCtx context.Context) error {
 	logger.Debug().Msg("logs and traces uploaded")
 
 	// Update workflow state
-	doneCtx := runnerCtx
+	doneCtx := runnerCtx //nolint:contextcheck
 	if doneCtx.Err() != nil {
+		shutdownCtx, shutdownCtxCancel := GetShutdownContext()
+		defer shutdownCtxCancel()
 		doneCtx = shutdownCtx
 	}
 
diff --git a/cmd/agent/core/agent.go b/cmd/agent/core/agent.go
index d91b76e812a..86927d685c3 100644
--- a/cmd/agent/core/agent.go
+++ b/cmd/agent/core/agent.go
@@ -54,33 +54,14 @@ const (
 	authInterceptorRefreshInterval = time.Minute * 30
 )
 
-const (
-	shutdownTimeout = time.Second * 5
-)
-
-var (
-	stopAgentFunc      context.CancelCauseFunc = func(error) {}
-	shutdownCancelFunc context.CancelFunc      = func() {}
-	shutdownCtx                                = context.Background()
-)
-
 func run(ctx context.Context, c *cli.Command, backends []types.Backend) error {
 	log.Info().Str("version", version.String()).Msg("Starting Woodpecker agent")
 
 	agentCtx, ctxCancel := context.WithCancelCause(ctx)
-	stopAgentFunc = func(err error) {
-		msg := "shutdown of whole agent"
-		if err != nil {
-			log.Error().Err(err).Msg(msg)
-		} else {
-			log.Info().Msg(msg)
-		}
-		stopAgentFunc = func(error) {}
-		shutdownCtx, shutdownCancelFunc = context.WithTimeout(shutdownCtx, shutdownTimeout)
-		ctxCancel(err)
-	}
-	defer stopAgentFunc(nil)
-	defer shutdownCancelFunc()
+	defer func() {
+		log.Info().Msg("shutdown of whole agent")
+		ctxCancel(nil)
+	}()
 
 	serviceWaitingGroup := errgroup.Group{}
 
@@ -107,6 +88,10 @@ func run(ctx context.Context, c *cli.Command, backends []types.Backend) error {
 				go func() {
 					<-agentCtx.Done()
 					log.Info().Msg("shutdown healthcheck server ...")
+
+					shutdownCtx, shutdownCtxCancel := agent.GetShutdownContext()
+					defer shutdownCtxCancel()
+
 					if err := server.Shutdown(shutdownCtx); err != nil { //nolint:contextcheck
 						log.Error().Err(err).Msg("shutdown healthcheck server failed")
 					} else {
@@ -302,7 +287,7 @@ func run(ctx context.Context, c *cli.Command, backends []types.Backend) error {
 				}
 
 				log.Debug().Msg("polling new workflow")
-				if err := runner.Run(agentCtx, shutdownCtx); err != nil {
+				if err := runner.Run(agentCtx); err != nil {
 					if singleWorkflow {
 						log.Error().Err(err).Msg("runner done with error")
 						ctxCancel(nil)