made startup more durable

Author: andrewnguyen22

controller/controller.go

@@ -3,10 +3,12 @@ package controller
 import (
 	"encoding/json"
 	"errors"
+	"fmt"
 	"net"
 	"os"
 	"os/exec"
 	"path/filepath"
+	"runtime"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -81,7 +83,9 @@ func New(fsm *fsm.StateMachine, c lib.Config, valKey crypto.PrivateKeyI, metrics
 	controller.loadCheckpointsFile()
 	// setup plugin if enabled
 	if c.Plugin != "" {
-		controller.PluginExecute(c.Plugin)
+		if err = controller.PluginExecute(c.Plugin); err != nil {
+			return nil, err
+		}
 		controller.PluginConnectSync()
 	}
 	// initialize the consensus in the controller, passing a reference to itself
@@ -252,24 +256,25 @@ const socketDir = "/tmp/plugin"
 const socketFile = "plugin.sock"
 
 // PluginExecute() executes the plugin control script to start the plugin process
-func (c *Controller) PluginExecute(plugin string) {
+func (c *Controller) PluginExecute(plugin string) lib.ErrorI {
 	if plugin == "" || strings.Contains(plugin, "..") || strings.ContainsRune(plugin, os.PathSeparator) {
-		c.log.Errorf("Invalid plugin name %q", plugin)
-		return
+		return lib.NewError(lib.NoCode, lib.MainModule, fmt.Sprintf("invalid plugin name %q", plugin))
+	}
+	cmdPath, err := resolvePluginCtlPath(plugin)
+	if err != nil {
+		return lib.NewError(lib.NoCode, lib.MainModule, err.Error())
 	}
-	// construct the shell command path: plugin/<plugin>/pluginctl.sh start
-	cmdPath := filepath.Join("plugin", plugin, "pluginctl.sh")
 	// create the command to execute the plugin control script with 'start' argument
 	cmd := exec.Command(cmdPath, "start")
 	// execute the command and capture output
 	output, err := cmd.CombinedOutput()
 	// if an error occurred during execution
 	if err != nil {
-		// log the error and exit
-		c.log.Errorf("Failed to execute plugin %s: %v, output: %s", plugin, err, string(output))
+		return lib.NewError(lib.NoCode, lib.MainModule, fmt.Sprintf("failed to execute plugin %s: %v, output: %s", plugin, err, string(output)))
 	}
 	// log successful plugin execution
 	c.log.Infof("Plugin %s started: %s", plugin, string(output))
+	return nil
 }
 
 // PluginConnectSync() blocking: enables a unix socket file where plugins can interact with the Canopy FSM
@@ -302,6 +307,31 @@ func (c *Controller) PluginConnectSync() {
 	c.FSM.Plugin, c.Mempool.FSM.Plugin = c.Plugin, c.Plugin
 }
 
+func resolvePluginCtlPath(plugin string) (string, error) {
+	relPath := filepath.Join("plugin", plugin, "pluginctl.sh")
+	candidates := []string{
+		relPath,
+	}
+	if exePath, err := os.Executable(); err == nil {
+		exeDir := filepath.Dir(exePath)
+		candidates = append(candidates,
+			filepath.Join(exeDir, relPath),
+			filepath.Join(filepath.Dir(exeDir), relPath),
+		)
+	}
+	if _, sourceFile, _, ok := runtime.Caller(0); ok {
+		repoRoot := filepath.Dir(filepath.Dir(sourceFile))
+		candidates = append(candidates, filepath.Join(repoRoot, relPath))
+	}
+	for _, candidate := range candidates {
+		info, err := os.Stat(candidate)
+		if err == nil && !info.IsDir() {
+			return candidate, nil
+		}
+	}
+	return "", fmt.Errorf("plugin launcher not found for %q; checked: %s", plugin, strings.Join(candidates, ", "))
+}
+
 // INTERNAL CALLS BELOW
 
 // LoadIsOwnRoot() returns if this chain is its own root (base)

controller/controller_test.go

@@ -0,0 +1,33 @@
+package controller
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestResolvePluginCtlPath(t *testing.T) {
+	wd, err := os.Getwd()
+	require.NoError(t, err)
+
+	tempDir := t.TempDir()
+	require.NoError(t, os.Chdir(tempDir))
+	t.Cleanup(func() {
+		require.NoError(t, os.Chdir(wd))
+	})
+
+	path, err := resolvePluginCtlPath("go")
+	require.NoError(t, err)
+	require.True(t, filepath.IsAbs(path))
+	require.FileExists(t, path)
+	require.Equal(t, "pluginctl.sh", filepath.Base(path))
+	require.Equal(t, "go", filepath.Base(filepath.Dir(path)))
+}
+
+func TestResolvePluginCtlPathMissing(t *testing.T) {
+	_, err := resolvePluginCtlPath("missing-plugin")
+	require.Error(t, err)
+	require.ErrorContains(t, err, "plugin launcher not found")
+}

plugin/go/pluginctl.sh

@@ -53,6 +53,11 @@ extract_if_needed() {
     return 1
 }
 
+get_process_cmd() {
+    local pid="$1"
+    ps -p "$pid" -o args= 2>/dev/null || ps -p "$pid" -o command= 2>/dev/null
+}
+
 # Check if the process is running based on PID file
 is_running() {
     # Return 1 if PID file doesn't exist
@@ -68,7 +73,7 @@ is_running() {
     # Check if process exists and is the go-plugin binary
     if ps -p "$pid" > /dev/null 2>&1; then
         # Verify it's actually our binary
-        if ps -p "$pid" -o cmd= | grep -q "go-plugin"; then
+        if get_process_cmd "$pid" | grep -q "go-plugin"; then
             return 0
         fi
     fi

plugin/kotlin/pluginctl.sh

@@ -39,6 +39,11 @@ extract_if_needed() {
     return 1
 }
 
+get_process_cmd() {
+    local pid="$1"
+    ps -p "$pid" -o args= 2>/dev/null || ps -p "$pid" -o command= 2>/dev/null
+}
+
 # Check if the process is running based on PID file
 is_running() {
     # Return 1 if PID file doesn't exist
@@ -54,7 +59,7 @@ is_running() {
     # Check if process exists and is the kotlin-plugin
     if ps -p "$pid" > /dev/null 2>&1; then
         # Verify it's actually our process
-        if ps -p "$pid" -o cmd= | grep -q "canopy-plugin-kotlin"; then
+        if get_process_cmd "$pid" | grep -q "canopy-plugin-kotlin"; then
             return 0
         fi
     fi

plugin/python/pluginctl.sh

@@ -118,6 +118,11 @@ setup_venv_if_needed() {
     return $?
 }
 
+get_process_cmd() {
+    local pid="$1"
+    ps -p "$pid" -o args= 2>/dev/null || ps -p "$pid" -o command= 2>/dev/null
+}
+
 # Check if the process is running based on PID file
 is_running() {
     # Return 1 if PID file doesn't exist
@@ -133,7 +138,7 @@ is_running() {
     # Check if process exists and is running our Python script
     if ps -p "$pid" > /dev/null 2>&1; then
         # Verify it's actually our Python script
-        if ps -p "$pid" -o cmd= | grep -q "python.*main.py"; then
+        if get_process_cmd "$pid" | grep -q "python.*main.py"; then
             return 0
         fi
     fi

plugin/typescript/pluginctl.sh

@@ -35,6 +35,11 @@ extract_if_needed() {
     return 1
 }
 
+get_process_cmd() {
+    local pid="$1"
+    ps -p "$pid" -o args= 2>/dev/null || ps -p "$pid" -o command= 2>/dev/null
+}
+
 # Check if the process is running based on PID file
 is_running() {
     # Return 1 if PID file doesn't exist
@@ -50,7 +55,7 @@ is_running() {
     # Check if process exists and is running our Node.js script
     if ps -p "$pid" > /dev/null 2>&1; then
         # Verify it's actually our Node.js script
-        if ps -p "$pid" -o cmd= | grep -q "node.*dist/main.js"; then
+        if get_process_cmd "$pid" | grep -q "node.*dist/main.js"; then
             return 0
         fi
     fi