ework-aio 0.1.9 → 0.1.11

package/bin/ework-aio

@@ -253,10 +253,11 @@ function startOne(svc: SvcName, foreground: boolean): boolean {
   }
   const envPath = svcEnvPath(svc);
   if (!existsSync(envPath)) {
-    console.error(
-      `${SVC_CONFIG[svc].label}: missing .env at ${envPath}.`
-    );
-    console.error("  Run `ework-aio install --no-start` first to scaffold config + data dir.");
+    console.error(`${SVC_CONFIG[svc].label}: missing .env at ${envPath}.`);
+    console.error("  Run `ework-aio install` first to scaffold config + data dir.");
+    console.error("  If install fails with npm ENOTEMPTY, a stale temp dir is blocking it.");
+    console.error("  Manual fix: sudo rm -rf \"$(npm root -g)/.ework-*\" \"$(npm root -g)\"/*.Trash*");
+    console.error("  Then re-run: ework-aio install");
     return false;
   }
 

package/bin/install.sh

@@ -12,6 +12,13 @@
 
 set -euo pipefail
 
+# ERR trap: when `set -e` kills the script, print the line + last command so
+# the user sees why instead of an empty prompt. Tricky failure modes (root
+# scope + polkit redirect, missing systemd PID 1, journalctl-as-root) all
+# manifest as silent exit at a systemctl call — without this trace, the user
+# just sees the prompt return and assumes install succeeded.
+trap 'rc=$?; if [[ $rc -ne 0 ]]; then echo "${c_red}✗${c_reset} install.sh exited with code $rc at line $LINENO (most recent command: ${BASH_COMMAND})" >&2; echo "  If systemd is unavailable on this host, retry with: ework-aio install --no-start" >&2; echo "  Then start services with: ework-aio start" >&2; fi' ERR
+
 # ─── Pretty output ──────────────────────────────────────────────────────────
 c_reset=$'\033[0m'; c_bold=$'\033[1m'; c_dim=$'\033[2m'
 c_red=$'\033[31m';   c_grn=$'\033[32m'; c_ylw=$'\033[33m'; c_blu=$'\033[34m'
@@ -162,12 +169,59 @@ ensure_pkg() {
 }
 case "$MODE" in
   install)
+    # Pre-clean stale npm rename temp-dirs from any previous failed install.
+    # npm's atomic-install pattern renames the existing package dir to a
+    # `.<pkg>-<rand>` sibling before extracting the new tarball; if a prior
+    # install crashed mid-flight, that temp dir stays behind and the NEXT
+    # install fails with ENOTEMPTY on the rename. Symptoms:
+    #   npm error code ENOTEMPTY
+    #   npm error syscall rename
+    #   npm error path .../node_modules/ework-web
+    #   npm error dest  .../node_modules/.ework-web-Cx2Tkt83
+    clean_npm_stale_dirs() {
+      local npm_root
+      npm_root="$(npm root -g 2>/dev/null)" || return 0
+      [[ -d "$npm_root" ]] || return 0
+      # Stale temp patterns: .ework-web-XXXX, .ework-daemon-XXXX, .opencode-ework-XXXX,
+      # .ework-aio-XXXX, plus the .Trash suffix some npm versions use.
+      local found=()
+      while IFS= read -r p; do
+        [[ -n "$p" ]] && found+=("$p")
+      done < <(find "$npm_root" -maxdepth 1 \( \
+        -name '.ework-web-*'      -o \
+        -name '.ework-daemon-*'   -o \
+        -name '.ework-aio-*'      -o \
+        -name '.opencode-ework-*' -o \
+        -name 'ework-*.Trash*'    -o \
+        -name 'opencode-ework.Trash*' \) 2>/dev/null || true)
+      if [[ ${#found[@]} -gt 0 ]]; then
+        log "Found ${#found[@]} stale npm temp dir(s) under $npm_root from a previous failed install:"
+        printf '  %s\n' "${found[@]}"
+        log "Removing..."
+        local d
+        for d in "${found[@]}"; do
+          rm -rf "$d" 2>/dev/null || warn "could not remove $d (try: sudo rm -rf $d)"
+        done
+      fi
+    }
+
+    npm_install_global() {
+      # One retry after a cleanup pass — covers the ENOTEMPTY case where npm
+      # itself produced a fresh temp dir during this same failed run.
+      if ! npm install -g "$1@latest"; then
+        warn "npm install -g $1@latest failed; cleaning stale temp dirs and retrying once..."
+        clean_npm_stale_dirs
+        npm install -g "$1@latest"
+      fi
+    }
+
+    clean_npm_stale_dirs
     log "Installing/updating npm packages globally..."
     # Install each as a top-level global so bins are linked to PATH.
     # (npm v9 doesn't hoist nested deps' bins from `npm i -g ework-aio`.)
     # Always run `npm install -g` (not gated on presence) so re-runs pick up new versions.
     for pkg in ework-web ework-daemon opencode-ework ework-aio; do
-      npm install -g "$pkg@latest" || die "npm install -g $pkg@latest failed"
+      npm_install_global "$pkg" || die "npm install -g $pkg@latest failed (even after cleanup retry; manual fix: sudo rm -rf $(npm root -g)/.$pkg-* $(npm root -g)/$pkg.Trash*)"
     done
     ok "npm packages ready"
 
@@ -634,22 +688,59 @@ write_unit_file \
 ok "Wrote $WEB_UNIT"
 
 # ─── Reload + start ework-web ──────────────────────────────────────────────
-ctl daemon-reload
-if [[ "$NO_START" == "0" ]]; then
+# systemd calls here are best-effort. If systemd isn't functional on this host
+# (containers without systemd PID 1, polkit redirects under sudo, missing
+# /etc/systemd/system) we still want install to complete so the user can run
+# `ework-aio start` (PID-file mode) against the scaffolded .env.
+SYSTEMD_OK=1
+if ! ctl daemon-reload; then
+  warn "systemctl daemon-reload failed — systemd may be unavailable on this host."
+  SYSTEMD_OK=0
+fi
+
+if [[ "$NO_START" == "0" && "$SYSTEMD_OK" == "1" ]]; then
   log "Starting ework-web..."
-  ctl enable ework-web.service
-  ctl restart ework-web.service
+  ctl enable ework-web.service || { warn "systemctl enable failed"; SYSTEMD_OK=0; }
+  if [[ "$SYSTEMD_OK" == "1" ]]; then
+    ctl restart ework-web.service || { warn "systemctl restart failed"; SYSTEMD_OK=0; }
+  fi
+  if [[ "$SYSTEMD_OK" == "1" ]]; then
+    for i in $(seq 1 60); do
+      if curl -sf -o /dev/null "http://127.0.0.1:$WORK_PORT/login"; then
+        ok "ework-web listening on :$WORK_PORT (after ${i} half-seconds)"
+        break
+      fi
+      sleep 0.5
+      [[ $i -eq 60 ]] && { warn "ework-web did not come up via systemd in 30s. Falling back to PID-file mode."; SYSTEMD_OK=0; }
+    done
+  fi
+else
+  warn "--no-start: unit enabled but not started"
+  ctl enable ework-web.service 2>/dev/null || true
+fi
+
+# ─── PID-file mode fallback for bot bootstrap ──────────────────────────────
+# If systemd couldn't start web (or we're on a host without systemd), bring
+# web up briefly via the PID-file path so the bot user + PAT bootstrap can
+# still complete. We leave web running in PID-file mode so subsequent curl
+# calls to :WORK_PORT succeed.
+if [[ "$NO_START" == "0" && "$SYSTEMD_OK" == "0" ]]; then
+  log "Starting ework-web via PID-file mode (nohup) for bot bootstrap..."
+  WEB_LOG_FILE="$DATA_DIR/run/web.log"
+  mkdir -p "$(dirname "$WEB_LOG_FILE")"
+  setsid nohup "$EWORK_WEB_BIN" >>"$WEB_LOG_FILE" 2>&1 </dev/null &
+  WEB_PID=$!
+  disown "$WEB_PID" 2>/dev/null || true
+  printf '%s\n' "$WEB_PID" > "$DATA_DIR/run/web.pid"
+  ok "ework-web started in PID-file mode (pid $WEB_PID, log $WEB_LOG_FILE)"
   for i in $(seq 1 60); do
     if curl -sf -o /dev/null "http://127.0.0.1:$WORK_PORT/login"; then
-      ok "ework-web listening on :$WORK_PORT (after ${i} half-seconds)"
+      ok "ework-web listening on :$WORK_PORT (after ${i} half-seconds, PID-file mode)"
       break
     fi
     sleep 0.5
-    [[ $i -eq 60 ]] && die "ework-web did not come up in 30s. Check: journalctl $SCOPENAME -u ework-web.service -n 50"
+    [[ $i -eq 60 ]] && die "ework-web did not come up in 30s. Tail of $WEB_LOG_FILE: $(tail -n 20 "$WEB_LOG_FILE" 2>/dev/null)"
   done
-else
-  warn "--no-start: unit enabled but not started"
-  ctl enable ework-web.service
 fi
 
 # ─── Bootstrap bot user + PAT (idempotent) ─────────────────────────────────
@@ -662,10 +753,23 @@ COOKIE_SIG=$(printf '%s' "$WORK_TOKEN_VAL" \
 AUTH_COOKIE="ework_auth=${WORK_TOKEN_VAL}.${COOKIE_SIG}"
 
 BOT_TOKEN=""
+BOT_BOOTSTRAP_OK=1
 if [[ -f "$BOT_TOKEN_FILE" ]]; then
   BOT_TOKEN=$(cat "$BOT_TOKEN_FILE")
   ok "Reusing saved bot token from $BOT_TOKEN_FILE"
 else
+  # Pre-flight: web must be reachable for the bot bootstrap HTTP calls below.
+  # In --no-start mode OR on hosts where systemd couldn't bring web up, skip
+  # the bootstrap entirely — daemon .env still gets written (with empty token)
+  # so the user can `ework-aio start` and re-run install later.
+  if ! curl -sf -o /dev/null "http://127.0.0.1:$WORK_PORT/login"; then
+    warn "ework-web not reachable at :$WORK_PORT — skipping bot bootstrap."
+    warn "Daemon will not be able to talk to web until you re-run install with web running."
+    BOT_BOOTSTRAP_OK=0
+  fi
+fi
+
+if [[ "$BOT_BOOTSTRAP_OK" == "1" && ! -f "$BOT_TOKEN_FILE" ]]; then
   log "Bootstrapping bot user '$BOT_NAME'..."
   BOT_PW=$(openssl rand -hex 24)
   CREATE_CODE=$(curl -sS -o /dev/null -w '%{http_code}' -X POST \
@@ -678,9 +782,11 @@ else
   case "$CREATE_CODE" in
     303) ok "Bot user '$BOT_NAME' created" ;;
     400|409) warn "Bot user '$BOT_NAME' already exists (continuing)" ;;
-    *) die "Failed to create bot user: HTTP $CREATE_CODE" ;;
+    *) warn "Failed to create bot user: HTTP $CREATE_CODE — skipping PAT mint; daemon will not be able to talk to web until re-run"; BOT_BOOTSTRAP_OK=0 ;;
   esac
+fi
 
+if [[ "$BOT_BOOTSTRAP_OK" == "1" && ! -f "$BOT_TOKEN_FILE" ]]; then
   log "Logging in as bot to mint PAT..."
   COOKIE_JAR=$(mktemp)
   LOGIN_CODE=$(curl -sS -c "$COOKIE_JAR" -X POST "http://127.0.0.1:$WORK_PORT/login" \
@@ -689,18 +795,25 @@ else
     -o /dev/null -w '%{http_code}') || LOGIN_CODE=000
   BOT_COOKIE=$(awk '/ework_auth/ {print $7}' "$COOKIE_JAR")
   rm -f "$COOKIE_JAR"
-  [[ "$LOGIN_CODE" == "302" && -n "$BOT_COOKIE" ]] \
-    || die "Bot login failed: HTTP $LOGIN_CODE"
+  if [[ "$LOGIN_CODE" != "302" || -z "$BOT_COOKIE" ]]; then
+    warn "Bot login failed: HTTP $LOGIN_CODE — skipping PAT mint"
+    BOT_BOOTSTRAP_OK=0
+  fi
+fi
 
+if [[ "$BOT_BOOTSTRAP_OK" == "1" && ! -f "$BOT_TOKEN_FILE" ]]; then
   log "Minting PAT..."
   PAT_RES=$(curl -sS -X POST "http://127.0.0.1:$WORK_PORT/me/tokens/create" \
     -H "Cookie: ework_auth=$BOT_COOKIE" \
     --data-urlencode "name=aio-$(date +%s)")
   BOT_TOKEN=$(printf '%s' "$PAT_RES" | grep -oE 'id="t">[a-f0-9]{40}<' | grep -oE '[a-f0-9]{40}' | head -1 || true)
-  [[ -n "$BOT_TOKEN" ]] || die "Could not extract PAT from token-create response"
-  printf '%s' "$BOT_TOKEN" > "$BOT_TOKEN_FILE"
-  chmod 600 "$BOT_TOKEN_FILE"
-  ok "Bot PAT saved to $BOT_TOKEN_FILE"
+  if [[ -z "$BOT_TOKEN" ]]; then
+    warn "Could not extract PAT from token-create response — daemon .env will have empty token"
+  else
+    printf '%s' "$BOT_TOKEN" > "$BOT_TOKEN_FILE"
+    chmod 600 "$BOT_TOKEN_FILE"
+    ok "Bot PAT saved to $BOT_TOKEN_FILE"
+  fi
 fi
 
 # ─── Write ework-daemon .env ───────────────────────────────────────────────
@@ -719,19 +832,30 @@ write_unit_file \
 # (trailing " \"\"" is a no-op separator; kept for future extra env injection)
 ok "Wrote $DAEMON_UNIT"
 
-ctl daemon-reload
-if [[ "$NO_START" == "0" ]]; then
+ctl daemon-reload 2>/dev/null || true
+if [[ "$NO_START" == "0" && "$SYSTEMD_OK" == "1" ]]; then
   log "Starting ework-daemon..."
-  ctl enable ework-daemon.service
-  ctl restart ework-daemon.service
+  ctl enable ework-daemon.service 2>/dev/null || true
+  ctl restart ework-daemon.service 2>/dev/null || warn "ework-daemon restart via systemd failed; use 'ework-aio start daemon' for PID-file mode"
   sleep 1
-  if ctl is-active --quiet ework-daemon.service; then
+  if ctl is-active --quiet ework-daemon.service 2>/dev/null; then
     ok "ework-daemon active"
   else
-    warn "ework-daemon did not report active; check: journalctl $SCOPENAME -u ework-daemon.service -n 50"
+    warn "ework-daemon did not report active; run 'ework-aio start daemon' for PID-file mode"
   fi
 else
-  ctl enable ework-daemon.service
+  ctl enable ework-daemon.service 2>/dev/null || true
+fi
+
+if [[ "$SYSTEMD_OK" == "0" ]]; then
+  hr
+  warn "systemd mode did not come up cleanly on this host."
+  warn "Services are scaffolded but not auto-started under systemd."
+  warn "Use PID-file mode instead:"
+  echo "    ework-aio start          # start web + daemon in background"
+  echo "    ework-aio ps             # check status"
+  echo "    ework-aio logs web       # tail web log"
+  hr
 fi
 
 # ─── Register opencode-ework plugin ────────────────────────────────────────

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ework-aio",
-  "version": "0.1.9",
+  "version": "0.1.11",
   "description": "All-in-one installer for ework (issue tracker) + ework-daemon (AI bridge) + opencode-ework (plugin). One command: npm i -g ework-aio && ework-aio install.",
   "type": "module",
   "license": "MIT",