skypilot-nightly 1.0.0.dev20250718__py3-none-any.whl → 1.0.0.dev20250720__py3-none-any.whl

sky/__init__.py

@@ -5,7 +5,7 @@ from typing import Optional
 import urllib.request
 
 # Replaced with the current commit when building the wheels.
-_SKYPILOT_COMMIT_SHA = '663a28261fc98dfa69214e1d4f1b0bb7b02664e0'
+_SKYPILOT_COMMIT_SHA = '80235386457c47b3c178c1e1842e9b07ed75dc06'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20250718'
+__version__ = '1.0.0.dev20250720'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 
@@ -104,6 +104,7 @@ from sky.client.sdk import job_status
 from sky.client.sdk import launch
 from sky.client.sdk import optimize
 from sky.client.sdk import queue
+from sky.client.sdk import reload_config
 from sky.client.sdk import start
 from sky.client.sdk import status
 from sky.client.sdk import stop
@@ -185,6 +186,7 @@ __all__ = [
     'optimize',
     'launch',
     'exec',
+    'reload_config',
     # core APIs
     'status',
     'start',

sky/backends/backend_utils.py

@@ -104,7 +104,7 @@ WAIT_HEAD_NODE_IP_MAX_ATTEMPTS = 3
 # Fixed IP addresses are used to avoid DNS lookup blocking the check, for
 # machine with no internet connection.
 # Refer to: https://stackoverflow.com/questions/3764291/how-can-i-see-if-theres-an-available-and-active-network-connection-in-python # pylint: disable=line-too-long
-_TEST_IP_LIST = ['https://1.1.1.1', 'https://8.8.8.8']
+_TEST_IP_LIST = ['https://8.8.8.8', 'https://1.1.1.1']
 
 # Allow each CPU thread take 2 tasks.
 # Note: This value cannot be too small, otherwise OOM issue may occur.
@@ -1635,18 +1635,28 @@ def get_node_ips(cluster_yaml: str,
 
 def check_network_connection():
     # Tolerate 3 retries as it is observed that connections can fail.
-    adapter = adapters.HTTPAdapter(max_retries=retry_lib.Retry(total=3))
     http = requests.Session()
-    http.mount('https://', adapter)
-    http.mount('http://', adapter)
-    for i, ip in enumerate(_TEST_IP_LIST):
-        try:
-            http.head(ip, timeout=3)
-            return
-        except (requests.Timeout, requests.exceptions.ConnectionError) as e:
-            if i == len(_TEST_IP_LIST) - 1:
-                raise exceptions.NetworkError('Could not refresh the cluster. '
-                                              'Network seems down.') from e
+    http.mount('https://', adapters.HTTPAdapter())
+    http.mount('http://', adapters.HTTPAdapter())
+
+    # Alternate between IPs on each retry
+    max_retries = 3
+    timeout = 0.5
+
+    for _ in range(max_retries):
+        for ip in _TEST_IP_LIST:
+            try:
+                http.head(ip, timeout=timeout)
+                return
+            except (requests.Timeout, requests.exceptions.ConnectionError):
+                continue
+
+        timeout *= 2  # Double the timeout for next retry
+
+    # If we get here, all IPs failed
+    # Assume network connection is down
+    raise exceptions.NetworkError('Could not refresh the cluster. '
+                                  'Network seems down.')
 
 
 @timeline.event
@@ -2610,7 +2620,7 @@ def is_controller_accessible(
           need_connection_check):
         # Check ssh connection if (1) controller is in INIT state, or (2) we failed to fetch the
         # status, both of which can happen when controller's status lock is held by another `sky jobs launch` or
-        # `sky serve up`. If we have controller's head_ip available and it is ssh-reachable,
+        # `sky serve up`. If we have controller's head_ip available and it is ssh-reachable,
         # we can allow access to the controller.
         ssh_credentials = ssh_credential_from_yaml(handle.cluster_yaml,
                                                    handle.docker_user,

sky/backends/cloud_vm_ray_backend.py

@@ -3988,12 +3988,16 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
         return dict(zip(job_ids, local_log_dirs))
 
     @context_utils.cancellation_guard
-    def tail_logs(self,
-                  handle: CloudVmRayResourceHandle,
-                  job_id: Optional[int],
-                  managed_job_id: Optional[int] = None,
-                  follow: bool = True,
-                  tail: int = 0) -> int:
+    def tail_logs(
+            self,
+            handle: CloudVmRayResourceHandle,
+            job_id: Optional[int],
+            managed_job_id: Optional[int] = None,
+            follow: bool = True,
+            tail: int = 0,
+            require_outputs: bool = False,
+            stream_logs: bool = True,
+            process_stream: bool = False) -> Union[int, Tuple[int, str, str]]:
         """Tail the logs of a job.
 
         Args:
@@ -4003,6 +4007,9 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
             follow: Whether to follow the logs.
             tail: The number of lines to display from the end of the
                 log file. If 0, print all lines.
+            require_outputs: Whether to return the stdout/stderr of the command.
+            stream_logs: Whether to stream the logs to stdout/stderr.
+            process_stream: Whether to process the stream.
 
         Returns:
             The exit code of the tail command. Returns code 100 if the job has
@@ -4022,18 +4029,19 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
             signal.signal(signal.SIGINT, backend_utils.interrupt_handler)
             signal.signal(signal.SIGTSTP, backend_utils.stop_handler)
         try:
-            returncode = self.run_on_head(
+            final = self.run_on_head(
                 handle,
                 code,
-                stream_logs=True,
-                process_stream=False,
+                stream_logs=stream_logs,
+                process_stream=process_stream,
+                require_outputs=require_outputs,
                 # Allocate a pseudo-terminal to disable output buffering.
                 # Otherwise, there may be 5 minutes delay in logging.
                 ssh_mode=command_runner.SshMode.INTERACTIVE,
             )
         except SystemExit as e:
-            returncode = e.code
-        return returncode
+            final = e.code
+        return final
 
     def tail_managed_job_logs(self,
                               handle: CloudVmRayResourceHandle,

sky/catalog/__init__.py

@@ -221,6 +221,8 @@ def get_default_instance_type(cpus: Optional[str] = None,
                               memory: Optional[str] = None,
                               disk_tier: Optional[
                                   resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None,
                               clouds: CloudFilter = None) -> Optional[str]:
     """Returns the cloud's default instance type for given #vCPUs and memory.
 
@@ -234,7 +236,7 @@ def get_default_instance_type(cpus: Optional[str] = None,
     the given CPU and memory requirement.
     """
     return _map_clouds_catalog(clouds, 'get_default_instance_type', cpus,
-                               memory, disk_tier)
+                               memory, disk_tier, region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/aws_catalog.py

@@ -230,10 +230,12 @@ def get_vcpus_mem_from_instance_type(
                                                         instance_type)
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # unused
     if cpus is None and memory is None:
         cpus = f'{_DEFAULT_NUM_VCPUS}+'
@@ -247,7 +249,8 @@ def get_default_instance_type(
     df = _get_df()
     df = df[df['InstanceType'].str.startswith(instance_type_prefix)]
     return common.get_instance_type_for_cpus_mem_impl(df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/azure_catalog.py

@@ -114,10 +114,12 @@ def _get_instance_family(instance_type: str) -> str:
     return instance_family
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     if cpus is None and memory is None:
         cpus = f'{_DEFAULT_NUM_VCPUS}+'
     if memory is None:
@@ -133,7 +135,8 @@ def get_default_instance_type(
 
     df = df.loc[df['InstanceType'].apply(_filter_disk_type)]
     return common.get_instance_type_for_cpus_mem_impl(df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/common.py

@@ -476,8 +476,11 @@ def _filter_region_zone(df: 'pd.DataFrame', region: Optional[str],
 
 
 def get_instance_type_for_cpus_mem_impl(
-        df: 'pd.DataFrame', cpus: Optional[str],
-        memory_gb_or_ratio: Optional[str]) -> Optional[str]:
+        df: 'pd.DataFrame',
+        cpus: Optional[str],
+        memory_gb_or_ratio: Optional[str],
+        region: Optional[str] = None,
+        zone: Optional[str] = None) -> Optional[str]:
     """Returns the cheapest instance type that satisfies the requirements.
 
     Args:
@@ -490,7 +493,10 @@ def get_instance_type_for_cpus_mem_impl(
             returned instance type should have at least the given memory size.
             If the string ends with "x", then the returned instance type should
             have at least the given number of vCPUs times the given ratio.
+        region: The region to filter by.
+        zone: The zone to filter by.
     """
+    df = _filter_region_zone(df, region, zone)
     df = _filter_with_cpus(df, cpus)
     df = _filter_with_mem(df, memory_gb_or_ratio)
     if df.empty:

sky/catalog/cudo_catalog.py

@@ -51,7 +51,9 @@ def get_vcpus_mem_from_instance_type(
 
 def get_default_instance_type(cpus: Optional[str] = None,
                               memory: Optional[str] = None,
-                              disk_tier: Optional[str] = None) -> Optional[str]:
+                              disk_tier: Optional[str] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier
     # NOTE: After expanding catalog to multiple entries, you may
     # want to specify a default instance type or family.
@@ -62,7 +64,8 @@ def get_default_instance_type(cpus: Optional[str] = None,
     if memory is None:
         memory_gb_or_ratio = f'{_DEFAULT_MEMORY_CPU_RATIO}x'
     return common.get_instance_type_for_cpus_mem_impl(_df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/do_catalog.py

@@ -52,11 +52,14 @@ def get_default_instance_type(
     cpus: Optional[str] = None,
     memory: Optional[str] = None,
     disk_tier: Optional[str] = None,
+    region: Optional[str] = None,
+    zone: Optional[str] = None,
 ) -> Optional[str]:
     # NOTE: After expanding catalog to multiple entries, you may
     # want to specify a default instance type or family.
     del disk_tier  # unused
-    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory)
+    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory, region,
+                                                      zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/fluidstack_catalog.py

@@ -52,7 +52,9 @@ def get_vcpus_mem_from_instance_type(
 
 def get_default_instance_type(cpus: Optional[str] = None,
                               memory: Optional[str] = None,
-                              disk_tier: Optional[str] = None) -> Optional[str]:
+                              disk_tier: Optional[str] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # unused
     if cpus is None and memory is None:
         cpus = f'{_DEFAULT_NUM_VCPUS}+'
@@ -61,7 +63,8 @@ def get_default_instance_type(cpus: Optional[str] = None,
     else:
         memory_gb_or_ratio = memory
     return common.get_instance_type_for_cpus_mem_impl(_df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/gcp_catalog.py

@@ -279,10 +279,12 @@ def get_vcpus_mem_from_instance_type(
     return common.get_vcpus_mem_from_instance_type_impl(_df, instance_type)
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     if cpus is None and memory is None:
         cpus = f'{_DEFAULT_NUM_VCPUS}+'
     if memory is None:
@@ -300,7 +302,8 @@ def get_default_instance_type(
 
     df = df.loc[df['InstanceType'].apply(_filter_disk_type)]
     return common.get_instance_type_for_cpus_mem_impl(df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/hyperbolic_catalog.py

@@ -67,9 +67,12 @@ def get_zone_shell_cmd() -> Optional[str]:
 
 def get_default_instance_type(cpus: Optional[str] = None,
                               memory: Optional[str] = None,
-                              disk_tier: Optional[str] = None) -> Optional[str]:
+                              disk_tier: Optional[str] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # Unused
-    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory)
+    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory, region,
+                                                      zone)
 
 
 def get_instance_type_for_accelerator(

sky/catalog/ibm_catalog.py

@@ -92,10 +92,12 @@ def list_accelerators(
                                          case_sensitive, all_regions)
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # unused
     if cpus is None and memory is None:
         cpus = f'{_DEFAULT_NUM_VCPUS}+'
@@ -107,7 +109,8 @@ def get_default_instance_type(
     instance_type_prefix = f'{_DEFAULT_INSTANCE_FAMILY}-'
     df = _df[_df['InstanceType'].str.startswith(instance_type_prefix)]
     return common.get_instance_type_for_cpus_mem_impl(df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def is_image_tag_valid(tag: str, region: Optional[str]) -> bool:

sky/catalog/lambda_catalog.py

@@ -56,10 +56,12 @@ def get_vcpus_mem_from_instance_type(
     return common.get_vcpus_mem_from_instance_type_impl(_df, instance_type)
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # unused
     if cpus is None and memory is None:
         cpus = f'{_DEFAULT_NUM_VCPUS}+'
@@ -68,7 +70,8 @@ def get_default_instance_type(
     else:
         memory_gb_or_ratio = memory
     return common.get_instance_type_for_cpus_mem_impl(_df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/nebius_catalog.py

@@ -51,12 +51,15 @@ def get_vcpus_mem_from_instance_type(
     return common.get_vcpus_mem_from_instance_type_impl(_df, instance_type)
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # unused
-    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory)
+    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory, region,
+                                                      zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/oci_catalog.py

@@ -101,10 +101,12 @@ def get_hourly_cost(instance_type: str,
                                        region, zone)
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     if cpus is None:
         cpus = f'{oci_utils.oci_config.DEFAULT_NUM_VCPUS}+'
 
@@ -127,7 +129,8 @@ def get_default_instance_type(
 
     logger.debug(f'# get_default_instance_type: {df}')
     return common.get_instance_type_for_cpus_mem_impl(df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/paperspace_catalog.py

@@ -52,11 +52,14 @@ def get_default_instance_type(
     cpus: Optional[str] = None,
     memory: Optional[str] = None,
     disk_tier: Optional[str] = None,
+    region: Optional[str] = None,
+    zone: Optional[str] = None,
 ) -> Optional[str]:
     # NOTE: After expanding catalog to multiple entries, you may
     # want to specify a default instance type or family.
     del disk_tier  # unused
-    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory)
+    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory, region,
+                                                      zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/runpod_catalog.py

@@ -41,11 +41,14 @@ def get_vcpus_mem_from_instance_type(
 
 def get_default_instance_type(cpus: Optional[str] = None,
                               memory: Optional[str] = None,
-                              disk_tier: Optional[str] = None) -> Optional[str]:
+                              disk_tier: Optional[str] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # RunPod does not support disk tiers.
     # NOTE: After expanding catalog to multiple entries, you may
     # want to specify a default instance type or family.
-    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory)
+    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory, region,
+                                                      zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/scp_catalog.py

@@ -51,10 +51,12 @@ def get_vcpus_mem_from_instance_type(
     return common.get_vcpus_mem_from_instance_type_impl(_df, instance_type)
 
 
-def get_default_instance_type(
-        cpus: Optional[str] = None,
-        memory: Optional[str] = None,
-        disk_tier: Optional[resources_utils.DiskTier] = None) -> Optional[str]:
+def get_default_instance_type(cpus: Optional[str] = None,
+                              memory: Optional[str] = None,
+                              disk_tier: Optional[
+                                  resources_utils.DiskTier] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier  # unused
     if cpus is None and memory is None:
         cpus = str(_DEFAULT_NUM_VCPUS)
@@ -63,7 +65,8 @@ def get_default_instance_type(
     else:
         memory_gb_or_ratio = memory
     return common.get_instance_type_for_cpus_mem_impl(_df, cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/vast_catalog.py

@@ -48,11 +48,14 @@ def get_vcpus_mem_from_instance_type(
 
 def get_default_instance_type(cpus: Optional[str] = None,
                               memory: Optional[str] = None,
-                              disk_tier: Optional[str] = None) -> Optional[str]:
+                              disk_tier: Optional[str] = None,
+                              region: Optional[str] = None,
+                              zone: Optional[str] = None) -> Optional[str]:
     del disk_tier
     # NOTE: After expanding catalog to multiple entries, you may
     # want to specify a default instance type or family.
-    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory)
+    return common.get_instance_type_for_cpus_mem_impl(_df, cpus, memory, region,
+                                                      zone)
 
 
 def get_accelerators_from_instance_type(

sky/catalog/vsphere_catalog.py

@@ -72,6 +72,8 @@ def get_default_instance_type(
     cpus: Optional[str] = None,
     memory: Optional[str] = None,
     disk_tier: Optional[str] = None,
+    region: Optional[str] = None,
+    zone: Optional[str] = None,
 ) -> Optional[str]:
     del disk_tier  # unused
     if cpus is None and memory is None:
@@ -81,7 +83,8 @@ def get_default_instance_type(
     else:
         memory_gb_or_ratio = memory
     return common.get_instance_type_for_cpus_mem_impl(_get_df(), cpus,
-                                                      memory_gb_or_ratio)
+                                                      memory_gb_or_ratio,
+                                                      region, zone)
 
 
 def get_accelerators_from_instance_type(

sky/client/cli/command.py

@@ -5114,6 +5114,12 @@ def serve_down(
               default=False,
               help='Sync down logs to the local machine. Can be combined with '
               '--controller, --load-balancer, or a replica ID to narrow scope.')
+@click.option(
+    '--tail',
+    default=None,
+    type=int,
+    help='The number of lines to display from the end of the log file. '
+    'Default is None, which means print all lines.')
 @click.argument('service_name', required=True, type=str)
 @click.argument('replica_ids', required=False, type=int, nargs=-1)
 @usage_lib.entrypoint
@@ -5126,6 +5132,7 @@ def serve_logs(
     load_balancer: bool,
     replica_ids: Tuple[int, ...],
     sync_down: bool,
+    tail: Optional[int],
 ):
     """Tail or sync down logs of a service.
 
@@ -5145,12 +5152,26 @@ def serve_logs(
         # Tail the logs of replica 1
         sky serve logs [SERVICE_NAME] 1
         \b
+        # Show the last 100 lines of the controller logs
+        sky serve logs --controller --tail 100 [SERVICE_NAME]
+        \b
         # Sync down all logs of the service (controller, LB, all replicas)
         sky serve logs [SERVICE_NAME] --sync-down
         \b
         # Sync down controller logs and logs for replicas 1 and 3
         sky serve logs [SERVICE_NAME] 1 3 --controller --sync-down
     """
+    if tail is not None:
+        if tail < 0:
+            raise click.UsageError('--tail must be a non-negative integer.')
+        # TODO(arda): We could add ability to tail and follow logs together.
+        if follow:
+            follow = False
+            logger.warning(
+                f'{colorama.Fore.YELLOW}'
+                '--tail and --follow cannot be used together. '
+                f'Changed the mode to --no-follow.{colorama.Style.RESET_ALL}')
+
     chosen_components: Set[serve_lib.ServiceComponent] = set()
     if controller:
         chosen_components.add(serve_lib.ServiceComponent.CONTROLLER)
@@ -5185,7 +5206,8 @@ def serve_logs(
             serve_lib.sync_down_logs(service_name,
                                      local_dir=str(log_dir),
                                      targets=targets_to_sync,
-                                     replica_ids=list(replica_ids))
+                                     replica_ids=list(replica_ids),
+                                     tail=tail)
         style = colorama.Style
         fore = colorama.Fore
         logger.info(f'{fore.CYAN}Service {service_name} logs: '
@@ -5227,7 +5249,8 @@ def serve_logs(
         serve_lib.tail_logs(service_name,
                             target=target_component,
                             replica_id=target_replica_id,
-                            follow=follow)
+                            follow=follow,
+                            tail=tail)
     except exceptions.ClusterNotUpError:
         with ux_utils.print_exception_no_traceback():
             raise