Files
hermes-agent/tools/environments/ssh.py

174 lines
6.9 KiB
Python
Raw Normal View History

2026-02-21 22:31:43 -08:00
"""SSH remote execution environment with ControlMaster connection persistence."""
import logging
import shlex
import shutil
2026-02-21 22:31:43 -08:00
import subprocess
import tempfile
from pathlib import Path
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
from tools.environments.base import BaseEnvironment, _popen_bash
from tools.environments.file_sync import FileSyncManager, iter_sync_files, quoted_rm_command
2026-02-21 22:31:43 -08:00
logger = logging.getLogger(__name__)
def _ensure_ssh_available() -> None:
"""Fail fast with a clear error when the SSH client is unavailable."""
if not shutil.which("ssh"):
raise RuntimeError(
"SSH is not installed or not in PATH. Install OpenSSH client: apt install openssh-client"
)
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
class SSHEnvironment(BaseEnvironment):
2026-02-21 22:31:43 -08:00
"""Run commands on a remote machine over SSH.
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
Spawn-per-call: every execute() spawns a fresh ``ssh ... bash -c`` process.
Session snapshot preserves env vars across calls.
CWD persists via in-band stdout markers.
Uses SSH ControlMaster for connection reuse.
2026-02-21 22:31:43 -08:00
"""
def __init__(self, host: str, user: str, cwd: str = "~",
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
timeout: int = 60, port: int = 22, key_path: str = ""):
2026-02-21 22:31:43 -08:00
super().__init__(cwd=cwd, timeout=timeout)
self.host = host
self.user = user
self.port = port
self.key_path = key_path
self.control_dir = Path(tempfile.gettempdir()) / "hermes-ssh"
self.control_dir.mkdir(parents=True, exist_ok=True)
self.control_socket = self.control_dir / f"{user}@{host}:{port}.sock"
_ensure_ssh_available()
2026-02-21 22:31:43 -08:00
self._establish_connection()
self._remote_home = self._detect_remote_home()
self._ensure_remote_dirs()
self._sync_manager = FileSyncManager(
get_files_fn=lambda: iter_sync_files(f"{self._remote_home}/.hermes"),
upload_fn=self._scp_upload,
delete_fn=self._ssh_delete,
)
self._sync_manager.sync(force=True)
2026-02-21 22:31:43 -08:00
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
self.init_session()
2026-03-12 23:38:11 +05:30
def _build_ssh_command(self, extra_args: list | None = None) -> list:
2026-02-21 22:31:43 -08:00
cmd = ["ssh"]
cmd.extend(["-o", f"ControlPath={self.control_socket}"])
cmd.extend(["-o", "ControlMaster=auto"])
cmd.extend(["-o", "ControlPersist=300"])
cmd.extend(["-o", "BatchMode=yes"])
cmd.extend(["-o", "StrictHostKeyChecking=accept-new"])
cmd.extend(["-o", "ConnectTimeout=10"])
if self.port != 22:
cmd.extend(["-p", str(self.port)])
if self.key_path:
cmd.extend(["-i", self.key_path])
if extra_args:
cmd.extend(extra_args)
cmd.append(f"{self.user}@{self.host}")
return cmd
def _establish_connection(self):
cmd = self._build_ssh_command()
cmd.append("echo 'SSH connection established'")
try:
result = subprocess.run(cmd, capture_output=True, text=True, timeout=15)
if result.returncode != 0:
error_msg = result.stderr.strip() or result.stdout.strip()
raise RuntimeError(f"SSH connection failed: {error_msg}")
except subprocess.TimeoutExpired:
raise RuntimeError(f"SSH connection to {self.user}@{self.host} timed out")
def _detect_remote_home(self) -> str:
"""Detect the remote user's home directory."""
try:
cmd = self._build_ssh_command()
cmd.append("echo $HOME")
result = subprocess.run(cmd, capture_output=True, text=True, timeout=10)
home = result.stdout.strip()
if home and result.returncode == 0:
logger.debug("SSH: remote home = %s", home)
return home
except Exception:
pass
if self.user == "root":
return "/root"
return f"/home/{self.user}"
# ------------------------------------------------------------------
# File sync (via FileSyncManager)
# ------------------------------------------------------------------
def _ensure_remote_dirs(self) -> None:
"""Create base ~/.hermes directory tree on remote in one SSH call."""
base = f"{self._remote_home}/.hermes"
dirs = [base, f"{base}/skills", f"{base}/credentials", f"{base}/cache"]
mkdir_cmd = "mkdir -p " + " ".join(shlex.quote(d) for d in dirs)
cmd = self._build_ssh_command()
cmd.append(mkdir_cmd)
subprocess.run(cmd, capture_output=True, text=True, timeout=10)
# _get_sync_files provided via iter_sync_files in FileSyncManager init
def _scp_upload(self, host_path: str, remote_path: str) -> None:
"""Upload a single file via scp over ControlMaster."""
parent = str(Path(remote_path).parent)
mkdir_cmd = self._build_ssh_command()
mkdir_cmd.append(f"mkdir -p {shlex.quote(parent)}")
subprocess.run(mkdir_cmd, capture_output=True, text=True, timeout=10)
scp_cmd = ["scp", "-o", f"ControlPath={self.control_socket}"]
if self.port != 22:
scp_cmd.extend(["-P", str(self.port)])
if self.key_path:
scp_cmd.extend(["-i", self.key_path])
scp_cmd.extend([host_path, f"{self.user}@{self.host}:{remote_path}"])
result = subprocess.run(scp_cmd, capture_output=True, text=True, timeout=30)
if result.returncode != 0:
raise RuntimeError(f"scp failed: {result.stderr.strip()}")
def _ssh_delete(self, remote_paths: list[str]) -> None:
"""Batch-delete remote files in one SSH call."""
cmd = self._build_ssh_command()
cmd.append(quoted_rm_command(remote_paths))
result = subprocess.run(cmd, capture_output=True, text=True, timeout=10)
if result.returncode != 0:
raise RuntimeError(f"remote rm failed: {result.stderr.strip()}")
def _before_execute(self) -> None:
"""Sync files to remote via FileSyncManager (rate-limited internally)."""
self._sync_manager.sync()
# ------------------------------------------------------------------
# Execution
# ------------------------------------------------------------------
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
def _run_bash(self, cmd_string: str, *, login: bool = False,
timeout: int = 120,
stdin_data: str | None = None) -> subprocess.Popen:
"""Spawn an SSH process that runs bash on the remote host."""
cmd = self._build_ssh_command()
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
if login:
cmd.extend(["bash", "-l", "-c", shlex.quote(cmd_string)])
else:
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
cmd.extend(["bash", "-c", shlex.quote(cmd_string)])
feat(environments): unified spawn-per-call execution layer Replace dual execution model (PersistentShellMixin + per-backend oneshot) with spawn-per-call + session snapshot for all backends except ManagedModal. Core changes: - Every command spawns a fresh bash process; session snapshot (env vars, functions, aliases) captured at init and re-sourced before each command - CWD persists via file-based read (local) or in-band stdout markers (remote) - ProcessHandle protocol + _ThreadedProcessHandle adapter for SDK backends - cancel_fn wired for Modal (sandbox.terminate) and Daytona (sandbox.stop) - Shared utilities extracted: _pipe_stdin, _popen_bash, _load_json_store, _save_json_store, _file_mtime_key, _SYNC_INTERVAL_SECONDS - Rate-limited file sync unified in base _before_execute() with _sync_files() hook - execute_oneshot() removed; all 11 call sites in code_execution_tool.py migrated to execute() - Daytona timeout wrapper replaced with SDK-native timeout parameter - persistent_shell.py deleted (291 lines) Backend-specific: - Local: process-group kill via os.killpg, file-based CWD read - Docker: -e env flags only on init_session, not per-command - SSH: shlex.quote transport, ControlMaster connection reuse - Singularity: apptainer exec with instance://, no forced --pwd - Modal: _AsyncWorker + _ThreadedProcessHandle, cancel_fn -> sandbox.terminate - Daytona: SDK-level timeout (not shell wrapper), cancel_fn -> sandbox.stop - ManagedModal: unchanged (gateway owns execution); docstring added explaining why
2026-04-08 13:38:04 -07:00
return _popen_bash(cmd, stdin_data)
2026-03-12 23:38:11 +05:30
2026-02-21 22:31:43 -08:00
def cleanup(self):
if self.control_socket.exists():
try:
cmd = ["ssh", "-o", f"ControlPath={self.control_socket}",
"-O", "exit", f"{self.user}@{self.host}"]
subprocess.run(cmd, capture_output=True, timeout=5)
except (OSError, subprocess.SubprocessError):
pass
try:
self.control_socket.unlink()
except OSError:
pass