Enhance diagnostics logging with context-aware timeouts (#1138)

* Enhance diagnostics logging with context-aware timeouts

* Add full system diagnostics logging on supervisor crash

* Clear diagnostics context after logging to prevent memory leaks

* remove pr.txt

Author: adamshiervani

cmd/main.go

@@ -13,6 +13,7 @@ import (
 
 	"github.com/erikdubbelboer/gspt"
 	"github.com/jetkvm/kvm"
+	"github.com/jetkvm/kvm/internal/diagnostics"
 	"github.com/jetkvm/kvm/internal/native"
 	"github.com/jetkvm/kvm/internal/supervisor"
 )
@@ -143,6 +144,13 @@ func supervise() error {
 			}
 		}
 
+		// Log full system diagnostics
+		diag := diagnostics.New(diagnostics.Options{Writer: logFile})
+		diag.LogAll("supervisor_crash")
+
+		// Ensure all data is flushed to disk before copying
+		_ = logFile.Sync()
+
 		createErrorDump(logFile)
 		os.Exit(exiterr.ExitCode())
 	}

internal/diagnostics/diagnostics.go

@@ -3,7 +3,9 @@
 package diagnostics
 
 import (
+	"context"
 	"io"
+	"time"
 
 	"github.com/jetkvm/kvm/internal/logging"
 	"github.com/rs/zerolog"
@@ -39,8 +41,11 @@ type Options struct {
 type Diagnostics struct {
 	logger  *zerolog.Logger
 	options Options
+	ctx     context.Context // set during LogAll for overall timeout
 }
 
+const defaultLogAllTimeout = 10 * time.Second
+
 // New creates a new Diagnostics instance using the default diagnostics logger.
 // If opts.Writer is set, logs are written there instead of the default logger.
 func New(opts Options) *Diagnostics {
@@ -61,7 +66,13 @@ func NewWithLogger(logger *zerolog.Logger, opts Options) *Diagnostics {
 
 // LogAll runs all diagnostic checks and logs the results.
 // The phase parameter distinguishes context (e.g., "crash" vs "handshake" vs "download").
+// LogAll has an overall timeout of 10 seconds; individual commands have a 2-second timeout.
 func (d *Diagnostics) LogAll(phase string) {
+	ctx, cancel := context.WithTimeout(context.Background(), defaultLogAllTimeout)
+	defer cancel()
+	d.ctx = ctx
+	defer func() { d.ctx = nil }()
+
 	d.logger.Error().Str("phase", phase).Msg("=== DIAGNOSTICS ===")
 
 	d.LogSystemInfo()

internal/diagnostics/helpers.go

@@ -12,7 +12,11 @@ const defaultCmdTimeout = 2 * time.Second
 
 // runCmdLog runs a command and logs its output.
 func (d *Diagnostics) runCmdLog(label string, cmd string, args ...string) {
-	ctx, cancel := context.WithTimeout(context.Background(), defaultCmdTimeout)
+	parentCtx := d.ctx
+	if parentCtx == nil {
+		parentCtx = context.Background()
+	}
+	ctx, cancel := context.WithTimeout(parentCtx, defaultCmdTimeout)
 	defer cancel()
 
 	output, err := exec.CommandContext(ctx, cmd, args...).CombinedOutput()
@@ -26,7 +30,11 @@ func (d *Diagnostics) runCmdLog(label string, cmd string, args ...string) {
 
 // runShellLog runs a shell command (for pipelines) and logs its output.
 func (d *Diagnostics) runShellLog(label, script string) {
-	ctx, cancel := context.WithTimeout(context.Background(), defaultCmdTimeout)
+	parentCtx := d.ctx
+	if parentCtx == nil {
+		parentCtx = context.Background()
+	}
+	ctx, cancel := context.WithTimeout(parentCtx, defaultCmdTimeout)
 	defer cancel()
 
 	output, err := exec.CommandContext(ctx, "sh", "-c", script).CombinedOutput()