skypilot-nightly 1.0.0.dev20250910__py3-none-any.whl → 1.0.0.dev20250913__py3-none-any.whl

sky/provision/seeweb/instance.py

@@ -0,0 +1,806 @@
+"""Seeweb provisioner for SkyPilot / Ray autoscaler.
+
+Prerequisites:
+    pip install ecsapi
+"""
+
+import os
+import subprocess
+import time
+from typing import Any, Dict, List, Optional, Tuple
+
+from sky import authentication as auth
+from sky import sky_logging
+from sky.adaptors import seeweb as seeweb_adaptor
+from sky.provision import common
+from sky.provision.common import ClusterInfo
+from sky.provision.common import InstanceInfo
+from sky.provision.common import ProvisionConfig
+from sky.provision.common import ProvisionRecord
+from sky.utils import command_runner  # Unified SSH helper
+from sky.utils import common_utils
+from sky.utils import status_lib
+
+logger = sky_logging.init_logger(__name__)
+
+# Singleton Seeweb client reused across the module
+_seeweb_client = None
+
+
+def _get_seeweb_client():
+    """Return a singleton Seeweb ECS API client."""
+    global _seeweb_client
+    if _seeweb_client is None:
+        # Initialize via adaptor's cached client
+        _seeweb_client = seeweb_adaptor.client()
+    return _seeweb_client
+
+
+# --------------------------------------------------------------------------- #
+# Useful constants
+# --------------------------------------------------------------------------- #
+_POLL_INTERVAL = 5  # sec
+_MAX_BOOT_TIME = 1200  # sec
+_ACTION_WATCH_MAX_RETRY = 360  # number of polls before giving up
+_ACTION_WATCH_FETCH_EVERY = 5  # seconds between polls
+_API_RETRY_MAX_RETRIES = 5
+_API_RETRY_INITIAL_BACKOFF = 1
+
+
+# --------------------------------------------------------------------------- #
+#  Class required by the Ray backend
+# --------------------------------------------------------------------------- #
+class SeewebNodeProvider:
+    """Minimalist provisioner for Seeweb ECS."""
+
+    def __init__(self, provider_config: ProvisionConfig, cluster_name: str):
+        """provider_config: dict populated by template (plan, image, location,
+                         remote_key_name, optional gpu…)
+        cluster_name   : SkyPilot name on cloud (used in notes)
+        """
+        self.config = provider_config
+        self.cluster_name = cluster_name
+        # Reuse a singleton Seeweb client to avoid repeated authentications/API
+        # object creations across different provider instances.
+        self.ecs = _get_seeweb_client()
+
+    def _get_ssh_user(self) -> str:
+        # Prefer auth config; fallback to template default for Seeweb
+        return (self.config.authentication_config.get('ssh_user') if self.config
+                and self.config.authentication_config else None) or 'ecuser'
+
+    def _get_private_key_path(self) -> str:
+        # Prefer explicit path from auth config; otherwise use SkyPilot key
+        key_path = None
+        if self.config and self.config.authentication_config:
+            key_path = self.config.authentication_config.get('ssh_private_key')
+        if not key_path:
+            key_path, _ = auth.get_or_generate_keys()
+        return os.path.expanduser(key_path)
+
+    # ------------------------------------------------------------------ #
+    # Helper: run a command on the VM via SSH using CommandRunner
+    # ------------------------------------------------------------------ #
+    def _run_remote(self,
+                    server_ip: str,
+                    cmd: str,
+                    *,
+                    timeout: int = 30,
+                    stream_logs: bool = False) -> subprocess.CompletedProcess:
+        """Execute *cmd* on the remote host.
+
+        Uses sky.utils.command_runner.SSHCommandRunner for consistent SSH
+        options across all providers.
+        Returns a subprocess.CompletedProcess-like
+        object with returncode, stdout, stderr.
+        """
+        runner = command_runner.SSHCommandRunner(
+            node=(server_ip, 22),
+            ssh_user=self._get_ssh_user(),
+            ssh_private_key=self._get_private_key_path(),
+        )
+        rc, stdout, stderr = runner.run(cmd,
+                                        stream_logs=stream_logs,
+                                        require_outputs=True,
+                                        connect_timeout=timeout)
+        # Convert to simple namespace for compatibility
+        proc = subprocess.CompletedProcess(args=cmd,
+                                           returncode=rc,
+                                           stdout=stdout.encode(),
+                                           stderr=stderr.encode())
+        return proc
+
+    # --------------------------------------------------------------------- #
+    # 1. bootstrap_instances – no preprocessing needed here
+    # --------------------------------------------------------------------- #
+
+    # --------------------------------------------------------------------- #
+    # 2. run_instances: restart or create until we reach count
+    # --------------------------------------------------------------------- #
+    def run_instances(self, config: Dict, count: int) -> None:
+        existing = self._query_cluster_nodes()
+        del config  # unused
+        running = [
+            s for s in existing if s.status in ('Booted', 'Running', 'RUNNING',
+                                                'Booting', 'PoweringOn')
+        ]
+
+        # a) restart Off servers
+        for srv in (s for s in existing if s.status == 'Booted'):
+            specific_status = self.ecs.fetch_server_status(srv.name)
+            if specific_status == 'SHUTOFF':
+                logger.info(f'Powering on server {srv.name}')
+                self._power_on(srv.name)
+                running.append(srv)
+            if len(running) >= count:
+                break
+
+        # b) create new VMs if missing
+        while len(running) < count:
+            self._create_server()
+            running.append({})  # placeholder
+
+    # --------------------------------------------------------------------- #
+    # 3. terminate_instances
+    # --------------------------------------------------------------------- #
+    def terminate_instances(self) -> None:
+        for srv in self._query_cluster_nodes():
+            logger.info('Deleting server %s …', srv.name)
+            self.ecs.delete_server(srv.name)  # DELETE /servers/{name}
+
+            # Retry deletion with exponential backoff
+            # to handle transient API errors
+            common_utils.retry(self.ecs.delete_server,
+                               max_retries=5,
+                               initial_backoff=1)(srv.name)
+
+    # --------------------------------------------------------------------- #
+    # 4. stop_instances
+    # --------------------------------------------------------------------- #
+    def stop_instances(self) -> None:
+        cluster_nodes = self._query_cluster_nodes()
+
+        for srv in cluster_nodes:
+            specific_status = self.ecs.fetch_server_status(srv.name)
+
+            if specific_status == 'SHUTOFF':
+                logger.info(f'\nServer {srv.name} is already stopped\n')
+                continue
+            elif srv.status in ('Booted', 'Running', 'RUNNING'):
+                # Get specific status to check if server is not already SHUTOFF
+                try:
+                    specific_status = self.ecs.fetch_server_status(srv.name)
+                    # Continue with power off only if
+                    # specific_status is not SHUTOFF
+                    # and general status is not STOPPED
+                    if specific_status != 'SHUTOFF' and srv.status != 'STOPPED':
+                        self._power_off(srv.name)
+                except Exception:  # pylint: disable=broad-except
+                    # Fallback: if we can't get specific
+                    # status, use general status check
+                    if srv.status != 'STOPPED':
+                        self._power_off(srv.name)
+            else:
+                logger.info(f'\nServer {srv.name} has status'
+                            f'{srv.status}, skipping\n')
+        # Wait for all servers to be actually stopped with forced refresh
+        self._wait_for_stop_with_forced_refresh()
+
+    # --------------------------------------------------------------------- #
+    # 5. query_instances
+    # --------------------------------------------------------------------- #
+    def query_instances(self) -> Dict[str, str]:
+        """Query instances status using both fetch_servers()
+        and fetch_server_status().
+
+        Seeweb has two different APIs:
+        - fetch_servers() returns states like 'Booted', 'Booting'
+        - fetch_server_status() returns states like 'SHUTOFF' (stopped)
+
+        We need to use fetch_server_status() to get the correct stopped state.
+        """
+        instances = {}
+        cluster_nodes = self._query_cluster_nodes()
+
+        for server in cluster_nodes:
+            # Always try to get the specific status first for more accuracy
+            try:
+                specific_status = self.ecs.fetch_server_status(server.name)
+                instances[server.name] = specific_status
+            except Exception:  # pylint: disable=broad-except
+                # Fallback to general status if fetch_server_status fails
+                general_status = server.status
+                instances[server.name] = general_status
+
+        return instances
+
+    # --------------------------------------------------------------------- #
+    # 6. wait_instances
+    # --------------------------------------------------------------------- #
+    def wait_instances(self, desired_state: str = 'Booted') -> None:
+        deadline = time.time() + _MAX_BOOT_TIME
+
+        while time.time() < deadline:
+            cluster_nodes = self._query_cluster_nodes()
+
+            # For SHUTOFF state, we need to use fetch_server_status()
+            # to get the real status
+            if desired_state == 'SHUTOFF':
+                all_shutoff = True
+                for server in cluster_nodes:
+                    try:
+                        specific_status = self.ecs.fetch_server_status(
+                            server.name)
+                        if specific_status != 'SHUTOFF':
+                            all_shutoff = False
+                    except Exception:  # pylint: disable=broad-except
+                        all_shutoff = False
+
+                if all_shutoff:
+                    return
+            else:
+                # For other states, use the general status
+                states = {srv.status for srv in cluster_nodes}
+
+                if states <= {desired_state}:
+                    # If all servers are Booted, wait
+                    # for them to be truly stable
+                    if desired_state == 'Booted':
+                        if self._wait_for_all_servers_stable():
+                            return
+                        else:
+                            time.sleep(_POLL_INTERVAL)
+                            continue
+                    return
+
+            time.sleep(_POLL_INTERVAL)
+
+        raise TimeoutError(
+            f'Nodes are not all in state {desired_state} within timeout')
+
+    def _wait_for_all_servers_stable(self, max_wait: int = 600) -> bool:
+        """Waits for all cluster servers to be stable."""
+        logger.info('Checking stability of all cluster servers...')
+
+        start_time = time.time()
+        while time.time() - start_time < max_wait:
+            cluster_nodes = self._query_cluster_nodes()
+            all_stable = True
+
+            for node in cluster_nodes:
+                if node.status == 'Booted':
+                    # Check that server is reachable via ping
+                    if not self._ping_server(node.ipv4):
+                        logger.warning(f'Server {node.name} ({node.ipv4})'
+                                       f'not reachable via ping')
+                        all_stable = False
+                        break
+
+                    # SSH readiness handled by provisioner.wait_for_ssh()
+
+                    logger.info(f'Server {node.name} ({node.ipv4}) is stable')
+
+            if all_stable:
+                logger.info('All servers are stable')
+                # Safety sleep to allow for late reboots
+                logger.info('Waiting 1 second to allow for late reboots...')
+                time.sleep(1)
+                return True
+
+            logger.info('Waiting for all servers to be stable...')
+            time.sleep(1)
+
+        logger.error('Timeout waiting for server stability')
+        return False
+
+    def _ping_server(self, server_ip: str) -> bool:
+        """Check that server is reachable via ping."""
+        try:
+            result = subprocess.run(['ping', '-c', '1', '-W', '5', server_ip],
+                                    capture_output=True,
+                                    timeout=10,
+                                    check=False)
+            return result.returncode == 0
+        except Exception as e:  # pylint: disable=broad-except
+            logger.debug(f'Error pinging {server_ip}: {e}')
+            return False
+
+    def _check_ssh_ready(self, server_ip: str) -> bool:
+        """Check that SSH is available on the server."""
+        try:
+            ssh_user = self._get_ssh_user()
+            private_key_path = self._get_private_key_path()
+            result = subprocess.run([
+                'ssh', '-o', 'ConnectTimeout=10', '-o',
+                'StrictHostKeyChecking=no', '-o',
+                f'UserKnownHostsFile={os.devnull}', '-o',
+                f'GlobalKnownHostsFile={os.devnull}', '-o',
+                'IdentitiesOnly=yes', '-i', private_key_path,
+                f'{ssh_user}@{server_ip}', 'echo "SSH ready"'
+            ],
+                                    capture_output=True,
+                                    timeout=15,
+                                    check=False)
+            return result.returncode == 0
+        except Exception as e:  # pylint: disable=broad-except
+            logger.debug(f'Error checking SSH on {server_ip}: {e}')
+            return False
+
+    # ------------------------------------------------------------------ #
+    # 7. open_ports / cleanup_ports – Seeweb has all ports open by default
+    # ------------------------------------------------------------------ #
+    def open_ports(
+        self,
+        cluster_name_on_cloud: str,
+        ports: List[str],
+        provider_config: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """See sky/provision/__init__.py"""
+        logger.debug(f'Skip opening ports {ports} for Seeweb instances, as all '
+                     'ports are open by default.')
+        del cluster_name_on_cloud, provider_config, ports
+
+    def cleanup_ports(
+        self,
+        cluster_name_on_cloud: str,
+        ports: List[str],
+        provider_config: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        del cluster_name_on_cloud, ports, provider_config  # Unused.
+
+    # ======================  private helpers  ========================= #
+    def _query_cluster_nodes(self):
+        """List servers with notes == cluster_name."""
+        servers = common_utils.retry(
+            self.ecs.fetch_servers,
+            max_retries=_API_RETRY_MAX_RETRIES,
+            initial_backoff=_API_RETRY_INITIAL_BACKOFF)()
+        return [
+            s for s in servers
+            if s.notes and s.notes.startswith(self.cluster_name)
+        ]
+
+    def query_cluster_nodes(self):
+        """Public wrapper for querying cluster nodes for this cluster."""
+        return common_utils.retry(self._query_cluster_nodes,
+                                  max_retries=_API_RETRY_MAX_RETRIES,
+                                  initial_backoff=_API_RETRY_INITIAL_BACKOFF)()
+
+    def _get_head_instance_id(self) -> Optional[str]:
+        """Return head instance id for this cluster.
+
+        Prefer notes == "{cluster}-head"; fallback to first node if none
+        matches (legacy naming).
+        """
+        nodes = common_utils.retry(self._query_cluster_nodes,
+                                   max_retries=_API_RETRY_MAX_RETRIES,
+                                   initial_backoff=_API_RETRY_INITIAL_BACKOFF)()
+        for node in nodes:
+            try:
+                if getattr(node, 'notes', None) == f'{self.cluster_name}-head':
+                    return node.name
+                if getattr(node, 'name', None) and node.name.endswith('-head'):
+                    return node.name
+            except Exception:  # pylint: disable=broad-except
+                continue
+        return nodes[0].name if nodes else None
+
+    def get_head_instance_id(self) -> Optional[str]:
+        """Public wrapper for getting head instance id."""
+        return common_utils.retry(self._get_head_instance_id,
+                                  max_retries=_API_RETRY_MAX_RETRIES,
+                                  initial_backoff=_API_RETRY_INITIAL_BACKOFF)()
+
+    def _create_server(self):
+        """POST /servers with complete payload."""
+        node_type = 'head'
+        payload = {
+            'plan': self.config.node_config.get('plan'),  # e.g. eCS4
+            'image': self.config.node_config.get('image'),  # e.g. ubuntu-2204
+            'location': self.config.node_config.get('location'),  # e.g. it-mi2
+            'notes': f'{self.cluster_name}-{node_type}',
+            'ssh_key': self.config.authentication_config.get('remote_key_name'
+                                                            ),  # remote key
+        }
+
+        # Optional GPU
+        if 'gpu' in self.config.node_config:
+            payload.update({
+                'gpu': self.config.node_config.get('gpu'),
+                'gpu_label': self.config.node_config.get('gpu_label', ''),
+            })
+
+        # Build the request object expected by ecsapi
+        server_create_request_cls = (
+            seeweb_adaptor.ecsapi.ServerCreateRequest  # type: ignore
+        )
+        create_request = server_create_request_cls(**payload)
+
+        logger.info('Creating Seeweb server %s', payload)
+
+        # POST /servers – returns (response, action_id)
+        _, action_id = common_utils.retry(
+            self.ecs.create_server,
+            max_retries=_API_RETRY_MAX_RETRIES,
+            initial_backoff=_API_RETRY_INITIAL_BACKOFF)(
+                create_request, check_if_can_create=False)
+        self.ecs.watch_action(action_id,
+                              max_retry=_ACTION_WATCH_MAX_RETRY,
+                              fetch_every=_ACTION_WATCH_FETCH_EVERY)
+
+    def _power_on(self, server_id: str):
+        try:
+            common_utils.retry(
+                self.ecs.turn_on_server,
+                max_retries=_API_RETRY_MAX_RETRIES,
+                initial_backoff=_API_RETRY_INITIAL_BACKOFF)(server_id)
+        except seeweb_adaptor.SeewebError as e:
+            logger.error(f'Error in _power_on for {server_id}: {e}')
+            raise
+
+    def _power_off(self, server_id: str):
+        try:
+            common_utils.retry(
+                self.ecs.turn_off_server,
+                max_retries=_API_RETRY_MAX_RETRIES,
+                initial_backoff=_API_RETRY_INITIAL_BACKOFF)(server_id)
+        except seeweb_adaptor.SeewebError as e:
+            logger.error(f'\n\nError in _power_off for {server_id}: {e}')
+            raise
+
+    def _wait_action(self, action_id: int):
+        """Poll action until it completes."""
+        while True:
+            action = common_utils.retry(
+                self.ecs.fetch_action,
+                max_retries=_API_RETRY_MAX_RETRIES,
+                initial_backoff=_API_RETRY_INITIAL_BACKOFF)(action_id)
+            if action['status'] in ('completed', 'ok', 'no_content'):
+                return
+            if action['status'] == 'error':
+                raise RuntimeError(f'Seeweb action {action_id} failed')
+            time.sleep(_POLL_INTERVAL)
+
+    def _wait_for_stop_with_forced_refresh(self, max_wait: int = 300) -> None:
+        """Wait for servers to be stopped with
+        aggressive polling and forced refresh."""
+        start_time = time.time()
+        poll_interval = 1  # 1 second for aggressive polling
+
+        while time.time() - start_time < max_wait:
+            # Force refresh by re-fetching cluster nodes
+            cluster_nodes = common_utils.retry(
+                self._query_cluster_nodes,
+                max_retries=_API_RETRY_MAX_RETRIES,
+                initial_backoff=_API_RETRY_INITIAL_BACKOFF)()
+
+            all_stopped = True
+            for server in cluster_nodes:
+                try:
+                    # Always use fetch_server_status() for accurate status
+                    specific_status = common_utils.retry(
+                        self.ecs.fetch_server_status,
+                        max_retries=_API_RETRY_MAX_RETRIES,
+                        initial_backoff=_API_RETRY_INITIAL_BACKOFF)(server.name)
+
+                    if specific_status != 'SHUTOFF':
+                        all_stopped = False
+
+                except Exception:  # pylint: disable=broad-except
+                    all_stopped = False
+
+            if all_stopped:
+                return
+
+            time.sleep(poll_interval)
+
+        raise TimeoutError(f'Servers not stopped within {max_wait} seconds')
+
+
+# =============================================================================
+# Standalone functions required by the provisioning interface
+# =============================================================================
+
+
+def run_instances(region: str, cluster_name_on_cloud: str,
+                  config: ProvisionConfig) -> ProvisionRecord:
+    """Run instances for Seeweb cluster."""
+    provider = SeewebNodeProvider(config, cluster_name_on_cloud)
+    provider.run_instances(config.node_config, config.count)
+
+    # Find the head node using notes convention
+    cluster_nodes = provider.query_cluster_nodes()
+    if not cluster_nodes:
+        raise RuntimeError(
+            f'No nodes found for cluster {cluster_name_on_cloud}')
+    head_node_id = provider.get_head_instance_id()
+    assert head_node_id is not None, 'head_instance_id should not be None'
+
+    return ProvisionRecord(
+        provider_name='Seeweb',
+        region=region,
+        zone=None,  # Seeweb doesn't use zones
+        cluster_name=cluster_name_on_cloud,
+        head_instance_id=head_node_id,
+        resumed_instance_ids=[],  # Empty for now
+        created_instance_ids=[node.name for node in cluster_nodes],
+    )
+
+
+def stop_instances(
+    cluster_name_on_cloud: str,
+    provider_config: Optional[Dict[str, Any]] = None,
+    worker_only: bool = False,
+) -> None:
+    """Stop instances for Seeweb cluster."""
+    del worker_only  # unused - Seeweb doesn't distinguish between head/worker
+    assert provider_config is not None
+
+    # Convert Dict to ProvisionConfig for SeewebNodeProvider
+    config = common.ProvisionConfig(
+        provider_config=provider_config,
+        authentication_config={},
+        docker_config={},
+        node_config=provider_config,
+        count=1,  # Not used for stop operation
+        tags={},
+        resume_stopped_nodes=False,
+        ports_to_open_on_launch=None,
+    )
+    provider = SeewebNodeProvider(config, cluster_name_on_cloud)
+    provider.stop_instances()
+
+
+def terminate_instances(
+    cluster_name_on_cloud: str,
+    provider_config: Optional[Dict[str, Any]] = None,
+    worker_only: bool = False,
+) -> None:
+    """Terminate instances for Seeweb cluster."""
+    del worker_only  # unused - Seeweb doesn't distinguish between head/worker
+    assert provider_config is not None
+    # Convert Dict to ProvisionConfig for SeewebNodeProvider
+    config = common.ProvisionConfig(
+        provider_config=provider_config,
+        authentication_config={},
+        docker_config={},
+        node_config=provider_config,
+        count=1,  # Not used for terminate operation
+        tags={},
+        resume_stopped_nodes=False,
+        ports_to_open_on_launch=None,
+    )
+    provider = SeewebNodeProvider(config, cluster_name_on_cloud)
+    provider.terminate_instances()
+
+
+def wait_instances(
+    region: str,
+    cluster_name_on_cloud: str,
+    state: Optional[status_lib.ClusterStatus],
+) -> None:
+    del region  # unused
+    # Map ClusterStatus to Seeweb string
+    if state == status_lib.ClusterStatus.UP:
+        seeweb_state = 'Booted'
+    elif state == status_lib.ClusterStatus.STOPPED:
+        seeweb_state = 'SHUTOFF'
+    elif state is None:
+        seeweb_state = 'Terminated'  # For termination
+    else:
+        seeweb_state = 'Booted'  # Default fallback
+
+    # Create Seeweb client directly and wait
+    client = _get_seeweb_client()
+    deadline = time.time() + _MAX_BOOT_TIME
+    while time.time() < deadline:
+        cluster_nodes = [
+            s for s in client.fetch_servers()
+            if s.notes and s.notes.startswith(cluster_name_on_cloud)
+        ]
+        if not cluster_nodes:
+            time.sleep(_POLL_INTERVAL)
+            continue
+
+        states = {srv.status for srv in cluster_nodes}
+        if states <= {seeweb_state}:
+            # If all servers are Booted, wait for them to be truly stable
+            if seeweb_state == 'Booted':
+                if _wait_for_all_servers_stable_standalone(cluster_nodes):
+                    return
+                else:
+                    time.sleep(_POLL_INTERVAL)
+                    continue
+            return
+        time.sleep(_POLL_INTERVAL)
+
+    raise TimeoutError(
+        f'Nodes are not all in state {seeweb_state} within timeout')
+
+
+def _wait_for_all_servers_stable_standalone(cluster_nodes,
+                                            max_wait: int = 300) -> bool:
+    """Waits for all cluster servers to be stable (standalone version)."""
+    start_time = time.time()
+    while time.time() - start_time < max_wait:
+        all_stable = True
+
+        for node in cluster_nodes:
+            if node.status == 'Booted':
+                # Check that server is reachable via ping
+                if not _ping_server_standalone(node.ipv4):
+                    all_stable = False
+                    break
+
+                # Do not check SSH here; handled by provisioner.wait_for_ssh().
+
+        if all_stable:
+            # Safety sleep to allow for late reboots
+            time.sleep(1)
+            return True
+
+        time.sleep(1)
+
+    return False
+
+
+def _ping_server_standalone(server_ip: str) -> bool:
+    """Check that server is reachable via ping (standalone version)."""
+    try:
+        result = subprocess.run(['ping', '-c', '1', '-W', '5', server_ip],
+                                capture_output=True,
+                                timeout=10,
+                                check=False)
+        return result.returncode == 0
+    except Exception as e:  # pylint: disable=broad-except
+        logger.error(f'Error pinging {server_ip}: {e}')
+        return False
+
+
+def _check_ssh_ready_standalone(server_ip: str) -> bool:
+    """Check that SSH is available on the server (standalone version)."""
+    try:
+        private_key_path, _ = auth.get_or_generate_keys()
+        private_key_path = os.path.expanduser(private_key_path)
+        ssh_user = 'ecuser'
+        result = subprocess.run([
+            'ssh', '-o', 'ConnectTimeout=10', '-o', 'StrictHostKeyChecking=no',
+            '-o', f'UserKnownHostsFile={os.devnull}', '-o',
+            f'GlobalKnownHostsFile={os.devnull}', '-o', 'IdentitiesOnly=yes',
+            '-i', private_key_path, f'{ssh_user}@{server_ip}',
+            'echo "SSH ready"'
+        ],
+                                capture_output=True,
+                                timeout=15,
+                                check=False)
+        return result.returncode == 0
+    except Exception:  # pylint: disable=broad-except
+        return False
+
+
+def query_instances(
+    cluster_name: str,
+    cluster_name_on_cloud: str,
+    provider_config: Optional[Dict[str, Any]] = None,
+    non_terminated_only: bool = True,
+) -> Dict[str, Tuple[Optional['status_lib.ClusterStatus'], Optional[str]]]:
+    """Query instances status for Seeweb cluster."""
+    del cluster_name  # unused
+    # Use the provided provider_config or default to empty dict
+    if provider_config is None:
+        provider_config = {}
+
+    # Convert Dict to ProvisionConfig for SeewebNodeProvider
+    config = common.ProvisionConfig(
+        provider_config=provider_config,
+        authentication_config={},
+        docker_config={},
+        node_config=provider_config,
+        count=1,  # Not used for query operation
+        tags={},
+        resume_stopped_nodes=False,
+        ports_to_open_on_launch=None,
+    )
+    provider = SeewebNodeProvider(config, cluster_name_on_cloud)
+    seeweb_instances = provider.query_instances()
+
+    # Map Seeweb status to SkyPilot status
+    status_map = {
+        'Booted':
+            status_lib.ClusterStatus.UP,  # Seeweb uses "Booted" for running
+        'RUNNING': status_lib.ClusterStatus.UP,  # All caps version
+        'Booting': status_lib.ClusterStatus.INIT,
+        'PoweringOn': status_lib.ClusterStatus.INIT,
+        'Off': status_lib.ClusterStatus.STOPPED,
+        'Stopped': status_lib.ClusterStatus.STOPPED,
+        'SHUTOFF':
+            status_lib.ClusterStatus.STOPPED,  # Add missing SHUTOFF status
+        'PoweringOff': status_lib.ClusterStatus.
+                       STOPPED,  # Fixed: should be STOPPED, not INIT
+    }
+
+    result: Dict[str, Tuple[Optional[status_lib.ClusterStatus],
+                            Optional[str]]] = {}
+    for name, seeweb_status in seeweb_instances.items():
+        if non_terminated_only and seeweb_status in ('Terminated', 'Deleted'):
+            continue
+        mapped_status = status_map.get(seeweb_status,
+                                       status_lib.ClusterStatus.INIT)
+        # Return tuple of (status, reason) where reason is None for Seeweb
+        result[name] = (mapped_status, None)
+
+    return result
+
+
+# Signature should not include provider_name; router strips it before calling
+def get_cluster_info(
+    region: str,
+    cluster_name_on_cloud: str,
+    provider_config: Optional[Dict[str, Any]] = None,
+) -> 'ClusterInfo':
+    del region  # unused
+    # Use Seeweb client to get cluster instances
+    client = _get_seeweb_client()
+    cluster_nodes = [
+        s for s in client.fetch_servers()
+        if s.notes and s.notes.startswith(cluster_name_on_cloud)
+    ]
+
+    if not cluster_nodes:
+        raise RuntimeError(
+            f'No instances found for cluster {cluster_name_on_cloud}')
+
+    instances = {}
+    head_instance = None
+    for node in cluster_nodes:
+        if getattr(node, 'notes', None) == f'{cluster_name_on_cloud}-head':
+            head_instance = node.name
+            break
+    if head_instance is None:
+        head_instance = cluster_nodes[0].name
+
+    for node in cluster_nodes:
+        # For Seeweb, we take the first node as head
+        if head_instance is None:
+            head_instance = node.name
+
+        # Get server IP (Seeweb uses 'ipv4' attribute)
+        external_ip = node.ipv4
+        internal_ip = external_ip  # For Seeweb, internal IP = external IP
+
+        instances[node.name] = [
+            InstanceInfo(
+                instance_id=node.name,
+                internal_ip=internal_ip,
+                external_ip=external_ip,
+                ssh_port=22,
+                tags={},
+            )
+        ]
+
+    return ClusterInfo(
+        instances=instances,
+        head_instance_id=head_instance,
+        provider_name='Seeweb',
+        provider_config=provider_config,
+    )
+
+
+def open_ports(
+    cluster_name_on_cloud: str,
+    ports: List[str],
+    provider_config: Optional[Dict[str, Any]] = None,
+) -> None:
+    del provider_config  # Unused
+    logger.debug(f'Seeweb: skipping open_ports for {cluster_name_on_cloud}'
+                 f'ports={ports} all ports are open by default')
+    return
+
+
+def cleanup_ports(
+    cluster_name_on_cloud: str,
+    ports: List[str],
+    provider_config: Optional[Dict[str, Any]] = None,
+) -> None:
+    del cluster_name_on_cloud, ports, provider_config  # Unused.
+    return