flowmesh-cli-stack 0.1.1__tar.gz → 0.1.3__tar.gz

{flowmesh_cli_stack-0.1.1/src/flowmesh_cli_stack.egg-info → flowmesh_cli_stack-0.1.3}/PKG-INFO

@@ -1,13 +1,13 @@
 Metadata-Version: 2.4
 Name: flowmesh-cli-stack
-Version: 0.1.1
+Version: 0.1.3
 Summary: FlowMesh CLI stack commands
 License-Expression: Apache-2.0
 Requires-Python: >=3.12
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: flowmesh-cli==0.1.1
-Requires-Dist: flowmesh-sdk-stack==0.1.1
+Requires-Dist: flowmesh-cli==0.1.3
+Requires-Dist: flowmesh-sdk-stack==0.1.3
 Requires-Dist: docker>=7.1.0
 Requires-Dist: packaging>=24
 Requires-Dist: pyyaml>=6.0.2

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3}/pyproject.toml

@@ -4,15 +4,15 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "flowmesh-cli-stack"
-version = "0.1.1"
+version = "0.1.3"
 description = "FlowMesh CLI stack commands"
 readme = "README.md"
 requires-python = ">=3.12"
 license = "Apache-2.0"
 license-files = ["LICENSE"]
 dependencies = [
-    "flowmesh-cli==0.1.1",
-    "flowmesh-sdk-stack==0.1.1",
+    "flowmesh-cli==0.1.3",
+    "flowmesh-sdk-stack==0.1.3",
     "docker>=7.1.0",
     "packaging>=24",
     "pyyaml>=6.0.2",

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3}/src/flowmesh_cli_stack/assets/.env.example

@@ -93,6 +93,13 @@ SSH_DEFAULT_IDLE_SEC=
 SSH_MAX_TTL_SEC=
 SSH_POLL_INTERVAL_SEC=
 SSH_STOP_TIMEOUT_SEC=
+SSH_MAX_CPU=
+SSH_MAX_MEMORY=
+SSH_MAX_PIDS=
+# Whether to apply requested GPU limits to SSH tasks.
+# If false, SSH tasks are allocated all available GPUs
+# regardless of their resource requests.
+ENABLE_SSH_GPU_LIMIT=false
 
 # ==== General Settings ====
 TZ=Asia/Singapore
@@ -110,9 +117,11 @@ TASK_MERGE_MAX_BATCH_SIZE=4
 ENABLE_CONTEXT_REUSE=true
 WORKER_CACHE_TTL_SEC=3600
 ENABLE_STAGE_WEIGHT_STICKINESS=false
+TASK_NO_WORKER_GRACE_SEC=60
 ENABLE_WORKER_WATCHDOG=true
 WORKER_DEATH_CHECK_INTERVAL=30
 WORKER_DEATH_GRACE_SEC=60
+WORKER_REHYDRATION_GRACE_SEC=120
 
 # ==== Server Heartbeat ====
 SERVER_HEARTBEAT_INTERVAL=30
@@ -166,11 +175,15 @@ NEBULA_API_TOKEN=
 
 # ==== External Plugins ====
 # Plugins are Python packages dropped under FLOWMESH_PLUGIN_DIR 
-# (host-mounted to /app/plugins on the server) and selected by 
-# FLOWMESH_PLUGINS as a comma-separated list of top-level module 
-# names. Each named module must expose `install()` returning a 
-# `HookBindings`. Leave both empty unless you ship a plugin.
+# (read-only at /app/plugins) and selected by FLOWMESH_PLUGINS as 
+# a comma-separated list of top-level module names. Each must 
+# expose `install()` returning a `HookBindings`. 
+# FLOWMESH_PLUGIN_DATA_DIR is writable at /app/plugin-data for 
+# plugin state. Leave all empty unless you ship a plugin.
 FLOWMESH_PLUGIN_DIR=./plugins
+# A path (`./x`, `/abs/x`) -> host bind-mount (auto-created).
+# A bare name -> external Docker volume of that name.
+FLOWMESH_PLUGIN_DATA_DIR=./plugin-data
 FLOWMESH_PLUGINS=
 
 # ==== Agent Executor (youtu-agent / utu) ====

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3}/src/flowmesh_cli_stack/assets/compose.yml

@@ -166,6 +166,7 @@ services:
       - /var/run/docker.sock:/var/run/docker.sock
       - ${SERVER_WORKER_CONFIG:-./configs/worker_config.yaml}:/etc/flowmesh/worker_config.yaml:ro
       - ${FLOWMESH_PLUGIN_DIR:-./plugins}:/app/plugins:ro
+      - ${FLOWMESH_PLUGIN_DATA_DIR:-./plugin-data}:/app/plugin-data
       - ${REDIS_TLS_DIR:-./secrets/tls/redis}:/etc/ssl/redis:ro
       - ${SERVER_TLS_DIR:-./secrets/tls/server}:/etc/ssl/server:ro
     restart: unless-stopped
@@ -199,3 +200,6 @@ volumes:
     name: ${FLOWMESH_STACK_SLUG:-flowmesh_node}_metrics
   flowmesh_server_logs:
     name: ${FLOWMESH_STACK_SLUG:-flowmesh_node}_server_logs
+  flowmesh_plugin_data:
+    external: true
+    name: ${FLOWMESH_PLUGIN_DATA_VOLUME:-${FLOWMESH_STACK_SLUG:-flowmesh_node}_plugin_data}

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3}/src/flowmesh_cli_stack/env_schema.py

@@ -301,6 +301,24 @@ STACK_ENV_SCHEMA = EnvSchema(
                 EnvVar("SSH_MAX_TTL_SEC", var_type=EnvVarType.FLOAT, min_value=0),
                 EnvVar("SSH_POLL_INTERVAL_SEC", var_type=EnvVarType.FLOAT, min_value=0),
                 EnvVar("SSH_STOP_TIMEOUT_SEC", var_type=EnvVarType.FLOAT, min_value=0),
+                EnvVar(
+                    "SSH_MAX_CPU",
+                    var_type=EnvVarType.FLOAT,
+                    min_value=0,
+                    min_inclusive=False,
+                ),
+                EnvVar("SSH_MAX_MEMORY"),
+                EnvVar("SSH_MAX_PIDS", var_type=EnvVarType.INT, min_value=1),
+                EnvVar(
+                    "ENABLE_SSH_GPU_LIMIT",
+                    "false",
+                    var_type=EnvVarType.BOOL,
+                    description=[
+                        "Whether to apply requested GPU limits to SSH tasks.",
+                        "If false, SSH tasks are allocated all available GPUs",
+                        "regardless of their resource requests.",
+                    ],
+                ),
             ],
         ),
         EnvSection(
@@ -370,6 +388,12 @@ STACK_ENV_SCHEMA = EnvSchema(
                     "false",
                     var_type=EnvVarType.BOOL,
                 ),
+                EnvVar(
+                    "TASK_NO_WORKER_GRACE_SEC",
+                    "60",
+                    var_type=EnvVarType.INT,
+                    min_value=0,
+                ),
                 EnvVar("ENABLE_WORKER_WATCHDOG", "true", var_type=EnvVarType.BOOL),
                 EnvVar(
                     "WORKER_DEATH_CHECK_INTERVAL",
@@ -383,6 +407,12 @@ STACK_ENV_SCHEMA = EnvSchema(
                     var_type=EnvVarType.INT,
                     min_value=0,
                 ),
+                EnvVar(
+                    "WORKER_REHYDRATION_GRACE_SEC",
+                    "120",
+                    var_type=EnvVarType.INT,
+                    min_value=0,
+                ),
             ],
         ),
         EnvSection(
@@ -512,10 +542,11 @@ STACK_ENV_SCHEMA = EnvSchema(
             title="External Plugins",
             description=[
                 "Plugins are Python packages dropped under FLOWMESH_PLUGIN_DIR ",
-                "(host-mounted to /app/plugins on the server) and selected by ",
-                "FLOWMESH_PLUGINS as a comma-separated list of top-level module ",
-                "names. Each named module must expose `install()` returning a ",
-                "`HookBindings`. Leave both empty unless you ship a plugin.",
+                "(read-only at /app/plugins) and selected by FLOWMESH_PLUGINS as ",
+                "a comma-separated list of top-level module names. Each must ",
+                "expose `install()` returning a `HookBindings`. ",
+                "FLOWMESH_PLUGIN_DATA_DIR is writable at /app/plugin-data for ",
+                "plugin state. Leave all empty unless you ship a plugin.",
             ],
             vars=[
                 EnvVar(
@@ -525,6 +556,15 @@ STACK_ENV_SCHEMA = EnvSchema(
                     use_default=True,
                     ensure_path="create",
                 ),
+                EnvVar(
+                    "FLOWMESH_PLUGIN_DATA_DIR",
+                    "./plugin-data",
+                    use_default=True,
+                    description=[
+                        "A path (`./x`, `/abs/x`) -> host bind-mount (auto-created).",
+                        "A bare name -> external Docker volume of that name.",
+                    ],
+                ),
                 EnvVar("FLOWMESH_PLUGINS", ""),
             ],
         ),

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3}/src/flowmesh_cli_stack/stack.py

@@ -35,6 +35,7 @@ from .env_schema import STACK_ENV_SCHEMA, deploy_overrides, role_overrides
 from .utils import (
     DEFAULT_ENV_FILE,
     STACK_PATH_KEYS,
+    apply_plugin_data_env,
     apply_stack_resource_env,
     ensure_deploy_paths,
     parse_node_role,
@@ -58,6 +59,7 @@ def _stack() -> DockerComposeStack:
         except ValueError as exc:
             logging.error(str(exc))
             raise typer.Exit(code=1)
+        apply_plugin_data_env(Path.cwd())
 
     return DockerComposeStack(
         compose_file=stack_compose_file(),
@@ -529,33 +531,87 @@ def down(
     logging.success("FlowMesh stack stopped.")
 
 
+STACK_SERVICES = ("server", "redis_control", "redis_telemetry")
+"""Compose services that can be restarted individually."""
+
+WORKER_MANAGING_SERVICES = ("server",)
+"""Services whose restart tears down the supervisor; drain workers first."""
+
+
 @app.command()
 def restart(
+    services: list[str] | None = typer.Argument(
+        None,
+        help=(
+            "Restart only these compose services in place (any of: "
+            f"{', '.join(STACK_SERVICES)}), leaving the rest of the stack "
+            "running. Omit to restart the whole stack."
+        ),
+    ),
     env_file: Path = typer.Option(
         DEFAULT_ENV_FILE, "--env-file", help="Env file for compose"
     ),
     image_tag: str | None = typer.Option(
         None, "--image-tag", help="Override FLOWMESH_VERSION"
     ),
+    pull: bool = typer.Option(
+        True, "--pull/--no-pull", help="Pull the target image before recreating."
+    ),
 ) -> None:
-    """Drain workers and restart the stack."""
-    logging.info("Draining workers...")
-    _drain_workers(env_file)
-    _compose(
-        ["down"],
-        env_file=env_file,
-        env=image_env_overrides(image_tag),
-        profile="root",
-    )
+    """Drain workers and restart the stack, or recreate specific services in place.
+
+    With one or more SERVICE arguments the stack is left running and only those services
+    are recreated; when any of them manages workers (the server / supervisor) its
+    workers are drained first so their in-flight tasks requeue onto other nodes.
+    """
+    if not services:
+        logging.info("Draining workers...")
+        _drain_workers(env_file)
+        _compose(
+            ["down"],
+            env_file=env_file,
+            env=image_env_overrides(image_tag),
+            profile="root",
+        )
+        profile = "root" if _node_role(env_file) == NodeRole.ROOT else None
+        _compose(
+            ["up", "-d", "--wait"],
+            env_file=env_file,
+            env=image_env_overrides(image_tag),
+            to_deploy=True,
+            profile=profile,
+        )
+        logging.success("FlowMesh stack is up.")
+        return
+
+    requested = list(dict.fromkeys(services))
+    unknown = [svc for svc in requested if svc not in STACK_SERVICES]
+    if unknown:
+        logging.error(
+            f"Unknown service(s): {', '.join(unknown)}. "
+            f"Known services: {', '.join(STACK_SERVICES)}."
+        )
+        raise typer.Exit(code=1)
+
+    if any(svc in WORKER_MANAGING_SERVICES for svc in requested):
+        logging.info("Draining workers...")
+        _drain_workers(env_file)
+
     profile = "root" if _node_role(env_file) == NodeRole.ROOT else None
+    up_args = ["up", "-d", "--no-deps", "--force-recreate", "--wait"]
+    if pull:
+        up_args += ["--pull", "always"]
+    up_args += requested
+    joined = ", ".join(requested)
+    logging.info(f"Recreating service(s): {joined}...")
     _compose(
-        ["up", "-d", "--wait"],
+        up_args,
         env_file=env_file,
         env=image_env_overrides(image_tag),
         to_deploy=True,
         profile=profile,
     )
-    logging.success("FlowMesh stack is up.")
+    logging.success(f"Service(s) restarted: {joined}.")
 
 
 @app.command()

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3}/src/flowmesh_cli_stack/utils.py

@@ -57,6 +57,22 @@ def apply_stack_resource_env() -> None:
         os.environ[WORKER_RESULTS_DIR_ENV] = results_volume
 
 
+_PLUGIN_DATA_PATH_PREFIXES = ("/", "./", "../", "~")
+_PLUGIN_DATA_ALIAS = "flowmesh_plugin_data"
+_PLUGIN_DATA_DEFAULT = "./plugin-data"
+_PLUGIN_DATA_VOLUME_ENV = "FLOWMESH_PLUGIN_DATA_VOLUME"
+
+
+def apply_plugin_data_env(base_dir: Path) -> None:
+    raw = os.environ.get("FLOWMESH_PLUGIN_DATA_DIR", "").strip()
+    if not raw or raw.startswith(_PLUGIN_DATA_PATH_PREFIXES):
+        resolved = resolve_path(raw, default=_PLUGIN_DATA_DEFAULT, base_dir=base_dir)
+        os.environ["FLOWMESH_PLUGIN_DATA_DIR"] = resolved.as_posix()
+    else:
+        os.environ[_PLUGIN_DATA_VOLUME_ENV] = raw
+        os.environ["FLOWMESH_PLUGIN_DATA_DIR"] = _PLUGIN_DATA_ALIAS
+
+
 def stack_compose_file() -> Path:
     return asset_path("flowmesh_cli_stack.assets", "compose.yml")
 
@@ -118,6 +134,14 @@ def ensure_deploy_paths(base_dir: Path) -> None:
             base_dir=base_dir,
         )
     )
+    if not os.environ.get(_PLUGIN_DATA_VOLUME_ENV):
+        ensure_dir(
+            resolve_path(
+                os.getenv("FLOWMESH_PLUGIN_DATA_DIR", ""),
+                default=_PLUGIN_DATA_DEFAULT,
+                base_dir=base_dir,
+            )
+        )
 
 
 def parse_node_role(raw: str) -> NodeRole:

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3}/src/flowmesh_cli_stack/worker.py

@@ -18,7 +18,13 @@ from flowmesh_stack.workers import (
     select_worker_images,
 )
 
-from .utils import DEFAULT_ENV_FILE, STACK_PATH_KEYS, stack_node_client
+from .utils import (
+    DEFAULT_ENV_FILE,
+    STACK_PATH_KEYS,
+    STACK_SLUG_ENV,
+    stack_node_client,
+    stack_resource_env_overrides,
+)
 
 app = get_typer(help="Create and manage workers on the local node.")
 
@@ -74,6 +80,15 @@ def worker_up(
             "Repeat --config-raw to provide multiple configs."
         ),
     ),
+    name_template: str | None = typer.Option(
+        None,
+        "--name-template",
+        help=(
+            "Worker name template for cpu/gpu presets. Placeholders: "
+            "{slug}, {kind}, {idx}, {gpu}. "
+            "Default: '{slug}_worker_{kind}_{idx|gpu}'."
+        ),
+    ),
     env_file: Path = typer.Option(
         DEFAULT_ENV_FILE, "--env-file", help="Env file to load defaults"
     ),
@@ -84,6 +99,11 @@ def worker_up(
 ) -> None:
     """Create and start one or more workers from presets or a custom config file."""
     client = stack_node_client(env_file, base_url, token or None)
+    try:
+        slug = stack_resource_env_overrides(os.environ)[STACK_SLUG_ENV]
+    except ValueError as exc:
+        logging.error(str(exc))
+        raise typer.Exit(code=1)
     logging.info("Creating workers...")
     try:
         created = create_workers(
@@ -93,6 +113,8 @@ def worker_up(
             targets=targets,
             config_paths=config,
             config_raw=config_raw,
+            name_template=name_template,
+            slug=slug,
         )
     except FlowMeshError as exc:
         logging.error(str(exc))

{flowmesh_cli_stack-0.1.1 → flowmesh_cli_stack-0.1.3/src/flowmesh_cli_stack.egg-info}/PKG-INFO

@@ -1,13 +1,13 @@
 Metadata-Version: 2.4
 Name: flowmesh-cli-stack
-Version: 0.1.1
+Version: 0.1.3
 Summary: FlowMesh CLI stack commands
 License-Expression: Apache-2.0
 Requires-Python: >=3.12
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: flowmesh-cli==0.1.1
-Requires-Dist: flowmesh-sdk-stack==0.1.1
+Requires-Dist: flowmesh-cli==0.1.3
+Requires-Dist: flowmesh-sdk-stack==0.1.3
 Requires-Dist: docker>=7.1.0
 Requires-Dist: packaging>=24
 Requires-Dist: pyyaml>=6.0.2

flowmesh_cli_stack-0.1.3/src/flowmesh_cli_stack.egg-info/requires.txt

@@ -0,0 +1,5 @@
+flowmesh-cli==0.1.3
+flowmesh-sdk-stack==0.1.3
+docker>=7.1.0
+packaging>=24
+pyyaml>=6.0.2

flowmesh_cli_stack-0.1.1/src/flowmesh_cli_stack.egg-info/requires.txt

@@ -1,5 +0,0 @@
-flowmesh-cli==0.1.1
-flowmesh-sdk-stack==0.1.1
-docker>=7.1.0
-packaging>=24
-pyyaml>=6.0.2