fix: installer starts the service reliably (retry + verify + installer.log), fails loudly; 1.1.6

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

installer/install_common.py

@@ -172,8 +172,26 @@ def wait_for_heartbeat(agent_token: str, hub: str, timeout: float = 120.0) -> bo
     return False
 
 
+def log_path() -> Path:
+    return Path(os.environ.get("GTI_LOG_DIR") or os.path.join(os.environ.get("PROGRAMDATA", r"C:\ProgramData"), "GTI", "FiscalAgent")) / "installer.log"
+
+
+def ilog(message: str) -> None:
+    """Append one timestamped line to installer.log (best effort, never raises, never holds a secret)."""
+    try:
+        path = log_path()
+        path.parent.mkdir(parents=True, exist_ok=True)
+        with path.open("a", encoding="utf-8") as fh:
+            fh.write(time.strftime("%Y-%m-%d %H:%M:%S ") + " ".join(message.split()) + chr(10))
+    except Exception:  # noqa: BLE001
+        pass
+
+
 def _sc(args: list[str]) -> subprocess.CompletedProcess[str]:
-    return subprocess.run(["sc.exe", *args], capture_output=True, text=True)
+    result = subprocess.run(["sc.exe", *args], capture_output=True, text=True)
+    text = " ".join((result.stdout + " " + result.stderr).split())
+    ilog(f"sc {' '.join(a for a in args if not a.lower().startswith('binpath'))} -> rc={result.returncode} {text[:300]}")
+    return result
 
 
 def install_service(install_dir_path: Path) -> None:
@@ -209,6 +227,57 @@ def start_service() -> None:
     _sc(["start", SERVICE_NAME])
 
 
+def service_state() -> str:
+    out = subprocess.run(["sc.exe", "query", SERVICE_NAME], capture_output=True, text=True).stdout
+    for name in ("RUNNING", "START_PENDING", "STOP_PENDING", "STOPPED", "PAUSED"):
+        if name in out:
+            return name
+    return "UNKNOWN"
+
+
+def service_process_alive() -> bool:
+    out = subprocess.run(["sc.exe", "queryex", SERVICE_NAME], capture_output=True, text=True).stdout
+    for line in out.splitlines():
+        if line.strip().upper().startswith("PID"):
+            pid = line.split(":", 1)[1].strip()
+            if pid.isdigit() and int(pid) > 0:
+                tl = subprocess.run(["tasklist", "/FI", f"PID eq {pid}", "/NH"], capture_output=True, text=True).stdout
+                return pid in tl
+    return False
+
+
+def start_and_verify(attempts: int = 4, per_attempt: float = 60.0) -> tuple[bool, str]:
+    """Start the service and prove it: RUNNING for a few seconds AND a live host process.
+    The first start after fresh files often exceeds the SCM's 30 s limit (error 1053: antivirus scans the
+    new onedir), so a failed attempt is diagnosed into installer.log and retried; the second start is fast."""
+    detail = "not started"
+    for attempt in range(1, attempts + 1):
+        ilog(f"start attempt {attempt}/{attempts}; state={service_state()}")
+        if service_state() != "RUNNING":
+            _sc(["start", SERVICE_NAME])
+        deadline = time.time() + per_attempt
+        while time.time() < deadline and service_state() in ("START_PENDING", "STOP_PENDING", "UNKNOWN"):
+            time.sleep(1.5)
+        if service_state() == "RUNNING":
+            time.sleep(6)  # must survive its own startup (preflight) rather than crash right after RUNNING
+            if service_state() == "RUNNING" and service_process_alive():
+                ilog("service RUNNING and host process alive")
+                return True, "RUNNING"
+        state = service_state()
+        query = subprocess.run(["sc.exe", "query", SERVICE_NAME], capture_output=True, text=True).stdout
+        exit_codes = " ".join(line.strip() for line in query.splitlines() if "EXIT_CODE" in line.upper())
+        detail = f"attempt {attempt}: state={state} {exit_codes}".strip()
+        ilog(detail)
+        try:
+            tail = (install_dir() / "logs" / "agent.log").read_text(encoding="utf-8", errors="replace").splitlines()[-3:]
+            ilog("agent.log tail: " + " | ".join(tail)[:400])
+        except Exception:  # noqa: BLE001
+            pass
+        time.sleep(5)
+    ilog(f"service did NOT start: {detail}")
+    return False, detail
+
+
 def stop_service() -> None:
     _sc(["stop", SERVICE_NAME])