fix: persistent daemon startup, plugin integration, mobile fallback

- Hook handler now loads .env.daemon for proper config (plugin URL/secret, bot user ID)
- Hook logs to /tmp/status-watcher.log instead of /dev/null
- Added .env.daemon config file (.gitignored - contains tokens)
- Added start-daemon.sh convenience script
- Plugin mode: mobile fallback updates post message field with formatted markdown
- Fixed unbounded lines array in status-watcher (capped at 50)
- Added session marker to formatter output for restart recovery
- Go plugin: added updatePostMessageForMobile() for dual-render strategy
  (webapp gets custom React component, mobile gets markdown in message field)

Fixes: daemon silently dying, no plugin connection, mobile showing blank posts
This commit is contained in:
sol
2026-03-08 07:42:18 +00:00
parent 0d0e6e9d90
commit 09441b34c1
7 changed files with 268 additions and 12 deletions

1
.gitignore vendored
View File

@@ -3,3 +3,4 @@ coverage/
*.log
plugin/server/dist/
plugin/webapp/node_modules/
.env.daemon

View File

@@ -22,6 +22,9 @@ const { spawn } = require('child_process');
// PID file location (must match watcher-manager.js default)
const PID_FILE = process.env.PID_FILE || '/tmp/status-watcher.pid';
// Log file location
const LOG_FILE = process.env.LIVESTATUS_LOG_FILE || '/tmp/status-watcher.log';
// Path to watcher-manager.js (relative to this hook file's location)
// Hook is in: workspace/hooks/status-watcher-hook/handler.js
// Watcher is in: workspace/projects/openclaw-live-status/src/watcher-manager.js
@@ -30,6 +33,12 @@ const WATCHER_PATH = path.resolve(
'../../projects/openclaw-live-status/src/watcher-manager.js',
);
// Path to .env.daemon config file
const ENV_DAEMON_PATH = path.resolve(
__dirname,
'../../projects/openclaw-live-status/.env.daemon',
);
/**
* Check if a process is alive given its PID.
* Returns true if process exists and is running.
@@ -60,6 +69,31 @@ function isWatcherRunning() {
}
}
/**
* Load .env.daemon file into an env object (key=value lines, ignoring comments).
* Returns merged env: process.env + .env.daemon overrides.
*/
function loadDaemonEnv() {
const env = Object.assign({}, process.env);
try {
// eslint-disable-next-line security/detect-non-literal-fs-filename
const content = fs.readFileSync(ENV_DAEMON_PATH, 'utf8');
for (const line of content.split('\n')) {
const trimmed = line.trim();
if (!trimmed || trimmed.startsWith('#')) continue;
const eqIdx = trimmed.indexOf('=');
if (eqIdx <= 0) continue;
const key = trimmed.slice(0, eqIdx).trim();
const val = trimmed.slice(eqIdx + 1).trim();
env[key] = val; // eslint-disable-line security/detect-object-injection
}
console.log('[status-watcher-hook] Loaded daemon config from', ENV_DAEMON_PATH);
} catch (_err) {
console.warn('[status-watcher-hook] No .env.daemon found at', ENV_DAEMON_PATH, '— using process.env only');
}
return env;
}
/**
* Spawn the watcher daemon as a detached background process.
* The parent (this hook handler) does not wait for it.
@@ -73,14 +107,32 @@ function spawnWatcher() {
console.log('[status-watcher-hook] Starting Live Status v4 watcher daemon...');
// Load .env.daemon for proper config (plugin URL/secret, bot user ID, etc.)
const daemonEnv = loadDaemonEnv();
// Open log file for append — never /dev/null
let logFd;
try {
logFd = fs.openSync(LOG_FILE, 'a');
console.log('[status-watcher-hook] Logging to', LOG_FILE);
} catch (err) {
console.error('[status-watcher-hook] Cannot open log file', LOG_FILE, ':', err.message);
logFd = 'ignore';
}
const child = spawn(process.execPath, [WATCHER_PATH, 'start'], {
detached: true,
stdio: 'ignore',
env: process.env,
stdio: ['ignore', logFd, logFd],
env: daemonEnv,
});
child.unref();
// Close the fd in the parent process (child inherits its own copy)
if (typeof logFd === 'number') {
try { fs.closeSync(logFd); } catch (_e) { /* ignore */ }
}
console.log(
'[status-watcher-hook] Watcher daemon spawned (PID will be written to',
PID_FILE + ')',

View File

@@ -116,7 +116,7 @@ func (p *Plugin) handleCreateSession(w http.ResponseWriter, r *http.Request) {
ChannelId: req.ChannelID,
RootId: req.RootID,
Type: "custom_livestatus",
Message: "", // Custom post types don't need a message
Message: "Agent session active",
}
post.AddProp("session_key", req.SessionKey)
post.AddProp("agent_id", req.AgentID)
@@ -127,6 +127,10 @@ func (p *Plugin) handleCreateSession(w http.ResponseWriter, r *http.Request) {
writeJSON(w, http.StatusInternalServerError, map[string]string{"error": appErr.Error()})
return
}
if createdPost == nil {
writeJSON(w, http.StatusInternalServerError, map[string]string{"error": "CreatePost returned nil without error"})
return
}
// Store session data
sessionData := SessionData{
@@ -193,9 +197,14 @@ func (p *Plugin) handleUpdateSession(w http.ResponseWriter, r *http.Request, ses
return
}
// Broadcast update via WebSocket (no Mattermost post API call!)
// Broadcast update via WebSocket (for webapp — instant, no API call)
p.broadcastUpdate(existing.ChannelID, *existing)
// Mobile fallback: update the post message field with formatted markdown.
// Mobile app doesn't render custom post types, so it shows the message field.
// The webapp plugin overrides rendering entirely, so "(edited)" is invisible on web.
go p.updatePostMessageForMobile(*existing)
writeJSON(w, http.StatusOK, map[string]string{"status": "ok"})
}
@@ -229,6 +238,100 @@ func (p *Plugin) handleDeleteSession(w http.ResponseWriter, r *http.Request, ses
writeJSON(w, http.StatusOK, map[string]string{"status": "done"})
}
// updatePostMessageForMobile updates the post's Message field with formatted markdown.
// This provides a mobile fallback — mobile apps don't render custom post type components
// but DO display the Message field. On webapp, the plugin's React component overrides
// rendering entirely, so the "(edited)" indicator is invisible.
func (p *Plugin) updatePostMessageForMobile(data SessionData) {
post, appErr := p.API.GetPost(data.PostID)
if appErr != nil || post == nil {
return
}
newMessage := formatStatusMarkdown(data)
if post.Message == newMessage {
return // Skip API call if content hasn't changed
}
post.Message = newMessage
_, updateErr := p.API.UpdatePost(post)
if updateErr != nil {
p.API.LogDebug("Failed to update post message for mobile", "postId", data.PostID, "error", updateErr.Error())
}
}
// formatStatusMarkdown generates a markdown blockquote status view for mobile clients.
func formatStatusMarkdown(data SessionData) string {
// Status icon
var statusIcon string
switch data.Status {
case "active":
statusIcon = "[ACTIVE]"
case "done":
statusIcon = "[DONE]"
case "error":
statusIcon = "[ERROR]"
case "interrupted":
statusIcon = "[INTERRUPTED]"
default:
statusIcon = "[UNKNOWN]"
}
// Elapsed time
elapsed := formatElapsedMs(data.ElapsedMs)
// Build lines
result := fmt.Sprintf("> **%s** `%s` | %s\n", statusIcon, data.AgentID, elapsed)
// Show last N status lines (keep it compact for mobile)
maxLines := 15
lines := data.Lines
if len(lines) > maxLines {
lines = lines[len(lines)-maxLines:]
}
for _, line := range lines {
if len(line) > 120 {
line = line[:117] + "..."
}
result += "> " + line + "\n"
}
// Token count for completed sessions
if data.Status != "active" && data.TokenCount > 0 {
result += fmt.Sprintf("> **[%s]** %s | %s tokens\n", strings.ToUpper(data.Status), elapsed, formatTokenCount(data.TokenCount))
}
return result
}
// formatElapsedMs formats milliseconds as human-readable duration.
func formatElapsedMs(ms int64) string {
if ms < 0 {
ms = 0
}
s := ms / 1000
m := s / 60
h := m / 60
if h > 0 {
return fmt.Sprintf("%dh%dm", h, m%60)
}
if m > 0 {
return fmt.Sprintf("%dm%ds", m, s%60)
}
return fmt.Sprintf("%ds", s)
}
// formatTokenCount formats a token count compactly.
func formatTokenCount(count int) string {
if count >= 1000000 {
return fmt.Sprintf("%.1fM", float64(count)/1000000)
}
if count >= 1000 {
return fmt.Sprintf("%.1fk", float64(count)/1000)
}
return fmt.Sprintf("%d", count)
}
// Helper functions
func readJSON(r *http.Request, v interface{}) error {

View File

@@ -1,12 +1,25 @@
package main
import (
"encoding/json"
"github.com/mattermost/mattermost/server/public/model"
)
// broadcastUpdate sends a WebSocket event to all clients viewing the given channel.
// The event is auto-prefixed by the SDK to "custom_com.openclaw.livestatus_update".
// All values must be gob-serializable primitives (string, int64, []string, etc.).
// Complex types like []SessionData must be JSON-encoded to a string first.
func (p *Plugin) broadcastUpdate(channelID string, data SessionData) {
// Serialize children to JSON string — gob cannot encode []SessionData
var childrenJSON string
if len(data.Children) > 0 {
b, err := json.Marshal(data.Children)
if err == nil {
childrenJSON = string(b)
}
}
payload := map[string]interface{}{
"post_id": data.PostID,
"session_key": data.SessionKey,
@@ -15,7 +28,7 @@ func (p *Plugin) broadcastUpdate(channelID string, data SessionData) {
"lines": data.Lines,
"elapsed_ms": data.ElapsedMs,
"token_count": data.TokenCount,
"children": data.Children,
"children_json": childrenJSON,
"start_time_ms": data.StartTimeMs,
}

View File

@@ -76,6 +76,8 @@ function format(sessionState, opts = {}) {
return '> ' + l;
})
.join('\n');
// Append invisible session marker for restart recovery (search by marker)
body += '\n<!-- sw:' + sessionState.sessionKey + ' -->';
}
return body;
}

View File

@@ -18,6 +18,8 @@ const path = require('path');
const { EventEmitter } = require('events');
const { resolve: resolveLabel } = require('./tool-labels');
const MAX_LINES_BUFFER = 50; // Cap state.lines to prevent memory leaks on long sessions
class StatusWatcher extends EventEmitter {
/**
* @param {object} opts
@@ -260,6 +262,11 @@ class StatusWatcher extends EventEmitter {
this._parseLine(sessionKey, state, line);
}
// Cap lines buffer to prevent unbounded growth on long sessions
if (state.lines.length > MAX_LINES_BUFFER) {
state.lines = state.lines.slice(-MAX_LINES_BUFFER);
}
// Update activity timestamp
state.lastActivityAt = Date.now();

78
start-daemon.sh Executable file
View File

@@ -0,0 +1,78 @@
#!/bin/bash
# start-daemon.sh — Start the live-status watcher daemon with proper config.
# This script is the canonical way to start the daemon. It loads env vars,
# ensures only one instance runs, and redirects logs properly.
set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
LOG_FILE="${LIVESTATUS_LOG_FILE:-/tmp/status-watcher.log}"
PID_FILE="${PID_FILE:-/tmp/status-watcher.pid}"
ENV_FILE="${SCRIPT_DIR}/.env.daemon"
# Load .env.daemon if it exists
if [ -f "$ENV_FILE" ]; then
set -a
# shellcheck disable=SC1090
source "$ENV_FILE"
set +a
fi
# Required env vars with defaults
export MM_BOT_TOKEN="${MM_BOT_TOKEN:?MM_BOT_TOKEN is required}"
export MM_BASE_URL="${MM_BASE_URL:-https://slack.solio.tech}"
export MM_BOT_USER_ID="${MM_BOT_USER_ID:-eqtkymoej7rw7dp8xbh7hywzrr}"
export TRANSCRIPT_DIR="${TRANSCRIPT_DIR:-/home/node/.openclaw/agents}"
export LOG_LEVEL="${LOG_LEVEL:-info}"
export IDLE_TIMEOUT_S="${IDLE_TIMEOUT_S:-60}"
export SESSION_POLL_MS="${SESSION_POLL_MS:-2000}"
export MAX_ACTIVE_SESSIONS="${MAX_ACTIVE_SESSIONS:-20}"
export MAX_STATUS_LINES="${MAX_STATUS_LINES:-20}"
export HEALTH_PORT="${HEALTH_PORT:-9090}"
export PID_FILE
export OFFSET_FILE="${OFFSET_FILE:-/tmp/status-watcher-offsets.json}"
# Plugin config (optional but recommended)
export PLUGIN_ENABLED="${PLUGIN_ENABLED:-true}"
export PLUGIN_URL="${PLUGIN_URL:-https://slack.solio.tech/plugins/com.openclaw.livestatus}"
export PLUGIN_SECRET="${PLUGIN_SECRET:-}"
# Kill existing daemon if running
if [ -f "$PID_FILE" ]; then
OLD_PID=$(cat "$PID_FILE" 2>/dev/null || true)
if [ -n "$OLD_PID" ] && kill -0 "$OLD_PID" 2>/dev/null; then
echo "Stopping existing daemon (PID $OLD_PID)..."
kill "$OLD_PID" 2>/dev/null || true
sleep 1
fi
rm -f "$PID_FILE"
fi
# Start daemon with proper logging
echo "Starting status watcher daemon..."
echo " Log file: $LOG_FILE"
echo " PID file: $PID_FILE"
echo " Plugin: ${PLUGIN_ENABLED} (${PLUGIN_URL:-not configured})"
cd "$SCRIPT_DIR"
node src/watcher-manager.js start >> "$LOG_FILE" 2>&1 &
DAEMON_PID=$!
# Wait for PID file
for i in $(seq 1 10); do
if [ -f "$PID_FILE" ]; then
echo "Daemon started (PID $(cat "$PID_FILE"))"
exit 0
fi
sleep 0.5
done
# Check if process is still running
if kill -0 "$DAEMON_PID" 2>/dev/null; then
echo "Daemon started (PID $DAEMON_PID) but PID file not created yet"
exit 0
else
echo "ERROR: Daemon failed to start. Check $LOG_FILE"
tail -20 "$LOG_FILE" 2>/dev/null
exit 1
fi