ob-metaflow-extensions 1.4.10__tar.gz → 1.4.12__tar.gz

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ob_metaflow_extensions
-Version: 1.4.10
+Version: 1.4.12
 Summary: Outerbounds Platform Extensions for Metaflow
 Author: Outerbounds, Inc.
 License: Commercial

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/metaflow_extensions/outerbounds/plugins/apps/core/app_cli.py

@@ -339,7 +339,7 @@ def deployment_instance_options(func):
         "--readiness-wait-time",
         type=int,
         help="The time (in seconds) to monitor the deployment for readiness after the readiness condition is met.",
-        default=4,
+        default=15,
     )
     @click.option(
         "--deployment-timeout",
@@ -377,8 +377,8 @@ def _package_necessary_things(app_config: AppConfig, logger):
     #      from the command line and that will alleviate any need to package any other directories for
     #
 
-    package_dir = app_config.get_state("packaging_directory")
-    if package_dir is None:
+    package_dirs = app_config.get_state("packaging_directories")
+    if package_dirs is None:
         app_config.set_state("code_package_url", None)
         app_config.set_state("code_package_key", None)
         return
@@ -391,13 +391,26 @@ def _package_necessary_things(app_config: AppConfig, logger):
         datastore_type=DEFAULT_DATASTORE, code_package_prefix=CODE_PACKAGE_PREFIX
     )
     package_url, package_key = packager.store(
-        paths_to_include=[package_dir], file_suffixes=suffixes
+        paths_to_include=package_dirs, file_suffixes=suffixes
     )
     app_config.set_state("code_package_url", package_url)
     app_config.set_state("code_package_key", package_key)
     logger("💾 Code package saved to : %s" % app_config.get_state("code_package_url"))
 
 
+def _sniff_pyproject_and_requirements(packaging_directories: List[str]):
+    pyproject_path = None
+    requirements_path = None
+    for directory in packaging_directories:
+        pyproject_toml = os.path.join(directory, "pyproject.toml")
+        requirements_txt = os.path.join(directory, "requirements.txt")
+        if os.path.exists(pyproject_toml):
+            pyproject_path = pyproject_toml
+        elif os.path.exists(requirements_txt):
+            requirements_path = requirements_txt
+    return pyproject_path, requirements_path
+
+
 @app.command(help="Deploy an app to the Outerbounds Platform.")
 @click.option(
     "--config-file",
@@ -449,23 +462,19 @@ def deploy(
             system_msg=True,
         )
 
-        packaging_directory = None
-        package_src_path = app_config.get("package", {}).get("src_path", None)
-        if package_src_path:
-            if os.path.isfile(package_src_path):
-                raise AppConfigError("src_path must be a directory, not a file")
-            elif os.path.isdir(package_src_path):
-                packaging_directory = os.path.abspath(package_src_path)
-            else:
-                raise AppConfigError(f"src_path '{package_src_path}' does not exist")
-        else:
-            # If src_path is None then we assume then we can assume for the moment
+        package_src_paths = app_config.get("package", {}).get("src_paths", [])
+        if package_src_paths is None:
+            package_src_paths = []
+
+        if len(package_src_paths) == 0:
+            # If src_paths is None then we assume then we can assume for the moment
             # that we can package the current working directory.
-            packaging_directory = os.getcwd()
+            package_src_paths = [os.getcwd()]
 
-        app_config.set_state("packaging_directory", packaging_directory)
+        app_config.set_state("packaging_directories", package_src_paths)
         logger(
-            "📦 Packaging directory : %s" % app_config.get_state("packaging_directory"),
+            "📦 Packaging directories : %s"
+            % ", ".join(app_config.get_state("packaging_directories")),
         )
 
         if app_config.get("no_deps", False):
@@ -489,11 +498,10 @@ def deploy(
                 )  # python gets a default value so it's always set.
                 # The user has not set any dependencies, so we can sniff the packaging directory
                 # for a dependencies file.
-                requirements_file = os.path.join(
-                    packaging_directory, "requirements.txt"
+                pyproject_toml, requirements_file = _sniff_pyproject_and_requirements(
+                    package_src_paths
                 )
-                pyproject_toml = os.path.join(packaging_directory, "pyproject.toml")
-                if os.path.exists(pyproject_toml):
+                if pyproject_toml:
                     app_config.set_state(
                         "dependencies",
                         {
@@ -504,7 +512,7 @@ def deploy(
                     logger(
                         "📦 Using dependencies from pyproject.toml: %s" % pyproject_toml
                     )
-                elif os.path.exists(requirements_file):
+                elif requirements_file:
                     app_config.set_state(
                         "dependencies",
                         {

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/metaflow_extensions/outerbounds/plugins/apps/core/capsule.py

@@ -7,7 +7,7 @@ import sys
 import time
 from functools import partial
 import shlex
-from typing import Optional, List, Dict, Any, Tuple, Union
+from typing import Optional, List, Dict, Any, Tuple, Union, Callable
 from .utils import TODOException, safe_requests_wrapper, MaximumRetriesExceeded
 from .app_config import AppConfig, CAPSULE_DEBUG, AuthType
 from . import experimental
@@ -75,7 +75,6 @@ class CapsuleStateMachine:
         return self._status_trail
 
     def add_status(self, status: CapsuleStatus):
-        assert type(status) == dict, "TODO: Make this check somewhere else"
         self._status_trail.append({"timestamp": time.time(), "status": status})
 
     @property
@@ -116,7 +115,9 @@ class CapsuleStateMachine:
         pass
 
     def save_debug_info(self, state_dir: str):
-        debug_path = os.path.join(state_dir, f"debug_capsule_{self._capsule_id}.json")
+        debug_path = os.path.join(
+            state_dir, f"debug_capsule_sm_{self._capsule_id}.json"
+        )
         with open(debug_path, "w") as f:
             json.dump(self._status_trail, f, indent=4)
 
@@ -427,7 +428,7 @@ class CapsuleApi:
                 message="Capsule JSON decode failed",
             )
 
-    def get(self, capsule_id: str):
+    def get(self, capsule_id: str) -> Dict[str, Any]:
         _url = os.path.join(self._base_url, capsule_id)
         response = self._wrapped_api_caller(
             requests.get,
@@ -446,6 +447,7 @@ class CapsuleApi:
                 message="Capsule JSON decode failed",
             )
 
+    # TODO: refactor me since name *currently(9/8/25)* is unique across capsules.
     def get_by_name(self, name: str, most_recent_only: bool = True):
         _url = os.path.join(self._base_url, f"?displayName={name}")
         response = self._wrapped_api_caller(
@@ -726,17 +728,51 @@ class CapsuleDeployer:
                 f"A capsule upgrade was triggered outside current deployment instance. Current deployment version was discarded. Current deployment version: {current_deployment_instance_version} and new version: {capsule_response.get('version', None)}",
             )
 
+    def _update_capsule_and_worker_sm(
+        self,
+        capsule_sm: "CapsuleStateMachine",
+        workers_sm: "CapsuleWorkersStateMachine",
+        logger: Callable[[str], None],
+    ) -> Tuple[Dict[str, Any], List[Dict[str, Any]]]:
+        capsule_response = self.get()
+        capsule_sm.add_status(capsule_response.get("status", {}))  # type: ignore
+
+        # We need to check if someone has not upgraded the capsule under the hood and
+        # the current deployment instance is invalid.
+        self._backend_version_mismatch_check(
+            capsule_response, self.current_deployment_instance_version  # type: ignore
+        )
+        workers_response = self.get_workers()
+        capsule_sm.report_current_status(logger)
+        workers_sm.add_status(workers_response)
+        workers_sm.report_current_status(logger)
+        return capsule_response, workers_response
+
+    def _publish_capsule_debug_info(
+        self,
+        capsule_sm: "CapsuleStateMachine",
+        workers_sm: "CapsuleWorkersStateMachine",
+        capsule_response: Dict[str, Any],
+    ):
+        if CAPSULE_DEBUG and self._debug_dir:
+            capsule_sm.save_debug_info(self._debug_dir)
+            workers_sm.save_debug_info(self._debug_dir)
+            debug_path = os.path.join(
+                self._debug_dir, f"debug_capsule_{self.identifier}.json"
+            )
+            with open(debug_path, "w") as f:
+                f.write(json.dumps(capsule_response, indent=4))
+
     def _monitor_worker_readiness(
         self,
         workers_sm: "CapsuleWorkersStateMachine",
+        capsule_sm: "CapsuleStateMachine",
     ):
         """returns True if the worker is crashlooping, False otherwise"""
         logger = self._logger_fn or partial(print, file=sys.stderr)
         for i in range(self._readiness_wait_time):
             time.sleep(1)
-            workers_response = self.get_workers()
-            workers_sm.add_status(workers_response)
-            workers_sm.report_current_status(logger)
+            self._update_capsule_and_worker_sm(capsule_sm, workers_sm, logger)
             if workers_sm.is_crashlooping:
                 return True
         return False
@@ -782,21 +818,19 @@ class CapsuleDeployer:
             minimum_replicas=min_replicas,
         )
         self.status = state_machine
+
+        # This loop will check all the conditions that help verify the terminal state.
+        # How it works is by extracting the statuses of the capsule and workers and
+        # then adding them as a part of a state-machine that helps track transitions and
+        # helps derive terminal states.
+        # We will first keep checking for terminal conditions or outright failure conditions
+        # If we reach a teminal condition like described in `DEPLOYMENT_READY_CONDITIONS`, then
+        # we will further check for readiness conditions.
         for i in range(self._create_timeout):
             time.sleep(1)
-            capsule_response = self.get()
-            workers_response = self.get_workers()
-
-            # We first need to check if someone has not upgraded the capsule under the hood and
-            # the current deployment instance is invalid.
-            self._backend_version_mismatch_check(
-                capsule_response, self.current_deployment_instance_version  # type: ignore
+            capsule_response, _ = self._update_capsule_and_worker_sm(
+                state_machine, workers_state_machine, logger
             )
-            state_machine.add_status(capsule_response.get("status", {}))  # type: ignore
-            workers_state_machine.add_status(workers_response)
-            state_machine.report_current_status(logger)
-
-            workers_state_machine.report_current_status(logger)
             # Deployment readiness checks will determine what is the terminal state
             # of the workerstate machine. If we detect a terminal state in the workers,
             # then even if the capsule upgrade is still in progress we will end up crashing
@@ -839,7 +873,8 @@ class CapsuleDeployer:
                         % self.identifier
                     )
                     _further_readiness_check_failed = self._monitor_worker_readiness(
-                        workers_state_machine
+                        workers_state_machine,
+                        state_machine,
                     )
 
                 if CAPSULE_DEBUG:
@@ -883,13 +918,18 @@ class CapsuleDeployer:
 
                 break
 
-            if CAPSULE_DEBUG and self._debug_dir:
-                state_machine.save_debug_info(self._debug_dir)
-                workers_state_machine.save_debug_info(self._debug_dir)
-                if i % 3 == 0:  # Every 3 seconds report the status
-                    logger(
-                        f"[debug] 💊 {self.capsule_type} {self.identifier} deployment status: {state_machine.current_status} | worker states: {workers_state_machine.current_status} | capsule_ready : {capsule_ready} | further_check_worker_readiness {further_check_worker_readiness}"
-                    )
+            self._publish_capsule_debug_info(
+                state_machine, workers_state_machine, capsule_response
+            )
+
+            if CAPSULE_DEBUG and i % 3 == 0:  # Every 3 seconds report the status
+                logger(
+                    f"[debug] 💊 {self.capsule_type} {self.identifier} deployment status: {state_machine.current_status} | worker states: {workers_state_machine.current_status} | capsule_ready : {capsule_ready} | further_check_worker_readiness {further_check_worker_readiness}"
+                )
+
+        self._publish_capsule_debug_info(
+            state_machine, workers_state_machine, capsule_response
+        )
 
         # We will only check ready_to_serve_traffic under the following conditions:
         # If the readiness condition is not Async and min_replicas in this deployment
@@ -907,13 +947,6 @@ class CapsuleDeployer:
                 f"Capsule {self.identifier} failed to be ready to serve traffic",
             )
 
-        if CAPSULE_DEBUG and self._debug_dir:
-            state_machine.save_debug_info(self._debug_dir)
-            workers_state_machine.save_debug_info(self._debug_dir)
-            logger(
-                f"[debug] 💊 {self.capsule_type} {self.identifier} deployment status [on return]: {state_machine.current_status} | worker states: {workers_state_machine.current_status}"
-            )
-
         return dict(
             id=self.identifier,
             auth_type=self.capsule_type,

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/metaflow_extensions/outerbounds/plugins/apps/core/code_package/code_packager.py

@@ -441,6 +441,7 @@ class CodePackager:
         root,
         exclude_hidden=True,
         suffixes=None,
+        normalized_rel_path=False,
     ) -> List[Tuple[str, str]]:
         """
         Walk a directory and yield tuples of (file_path, relative_arcname) for files
@@ -456,6 +457,8 @@ class CodePackager:
             Whether to exclude hidden files and directories (those starting with '.')
         suffixes : List[str], optional
             List of file suffixes to include (e.g. ['.py', '.txt'])
+        normalized_rel_path : bool, default False
+            Whether to normalize the relative from the root. ie if the root is /a/b/c and the file is /a/b/c/d/e.py then the relative path will be d/e.py
 
         Returns
         -------
@@ -468,6 +471,8 @@ class CodePackager:
         for file_path, rel_path in symlink_friendly_walk(
             root, exclude_hidden, suffixes
         ):
+            if normalized_rel_path:
+                rel_path = file_path.replace(root, "")
             files.append((file_path, rel_path))
         return files
 
@@ -520,6 +525,7 @@ class CodePackager:
                         path,
                         exclude_hidden=True,
                         suffixes=suffixes,
+                        normalized_rel_path=True,
                     ):
                         tar.add(
                             file_path,

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/metaflow_extensions/outerbounds/plugins/apps/core/config/typed_configs.py

@@ -40,6 +40,7 @@ class ResourceConfigDict(TypedDict, total=False):
     memory: Optional[str]
     gpu: Optional[str]
     disk: Optional[str]
+    shared_memory: Optional[str]
 
 
 class AuthConfigDict(TypedDict, total=False):
@@ -67,7 +68,7 @@ class DependencyConfigDict(TypedDict, total=False):
 
 
 class PackageConfigDict(TypedDict, total=False):
-    src_path: Optional[str]
+    src_paths: Optional[list]
     suffixes: Optional[list]
 
 

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/metaflow_extensions/outerbounds/plugins/apps/core/config/unified_config.py

@@ -222,6 +222,18 @@ class ResourceConfig(metaclass=ConfigMeta):
         parsing_fn=UnitParser("disk").parse,
     )
 
+    shared_memory = ConfigField(
+        cli_meta=CLIOption(
+            name="shared_memory",
+            cli_option_str="--shared-memory",
+            help="Shared memory resource request and limit.",
+        ),
+        field_type=str,
+        example="1Gi",
+        validation_fn=UnitParser.validation_wrapper_fn("memory"),
+        parsing_fn=UnitParser("memory").parse,
+    )
+
 
 class HealthCheckConfig(metaclass=ConfigMeta):
     """Health check configuration."""
@@ -541,14 +553,16 @@ class DependencyConfig(metaclass=ConfigMeta):
 class PackageConfig(metaclass=ConfigMeta):
     """Package configuration."""
 
-    src_path = ConfigField(
+    src_paths = ConfigField(
         cli_meta=CLIOption(
             name="package_src_path",
             cli_option_str="--package-src-path",
+            multiple=True,
             help="The path to the source code to deploy with the App.",
+            click_type=str,
         ),
-        field_type=str,
-        example="./",
+        field_type=list,
+        example=["./"],
     )
     suffixes = ConfigField(
         cli_meta=CLIOption(
@@ -560,6 +574,28 @@ class PackageConfig(metaclass=ConfigMeta):
         example=[".py", ".ipynb"],
     )
 
+    @staticmethod
+    def validate(package_config: "PackageConfig"):
+        if package_config.src_paths is None:
+            return True
+        if package_config.src_paths:
+            for path in package_config.src_paths:
+                if not os.path.exists(path):
+                    raise ConfigValidationFailedException(
+                        field_name="src_paths",
+                        field_info=package_config._get_field("src_paths"),  # type: ignore
+                        current_value=package_config.src_paths,
+                        message=f"Path does not exist : `{path}`",
+                    )
+                if not os.path.isdir(path):
+                    raise ConfigValidationFailedException(
+                        field_name="src_paths",
+                        field_info=package_config._get_field("src_paths"),  # type: ignore
+                        current_value=package_config.src_paths,
+                        message=f"Path is not a directory : `{path}`",
+                    )
+        return True
+
 
 def everything_is_string(*args):
     return all(isinstance(arg, str) for arg in args)
@@ -852,6 +888,7 @@ How to read this schema:
         cli_meta=None,  # No top-level CLI option, only nested fields have CLI options
         field_type=PackageConfig,
         help="Configurations associated with packaging the app.",
+        validation_fn=PackageConfig.validate,
     )
 
     no_deps = ConfigField(

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/metaflow_extensions/outerbounds/plugins/apps/core/config_schema.yaml

@@ -129,6 +129,12 @@ properties:
         example: 1Gi
         mutation_behavior: union
         cli_option: --disk
+      shared_memory:
+        description: Shared memory resource request and limit. (validation applied)
+        type: string
+        example: 1Gi
+        mutation_behavior: union
+        cli_option: --shared-memory
     mutation_behavior: union
   auth:
     title: AuthConfig
@@ -218,7 +224,7 @@ properties:
         description: The Python version to use for the app.
         type: string
         example: '3.10'
-        mutation_behavior: not_allowed
+        mutation_behavior: union
         cli_option: --python
       pypi:
         description: |-
@@ -247,10 +253,13 @@ properties:
     type: object
     required: []
     properties:
-      src_path:
+      src_paths:
         description: The path to the source code to deploy with the App.
-        type: string
-        example: ./
+        type: array
+        items:
+          type: string
+        example:
+        - ./
         mutation_behavior: union
         cli_option: --package-src-path
       suffixes:

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/metaflow_extensions/outerbounds/plugins/apps/core/experimental/__init__.py

@@ -55,24 +55,31 @@ def _code_info(app_config: "AppConfig"):
         def _call_git(args, path=None):
             return "", 1, True
 
-    repo_info = get_repository_info(app_config.get_state("packaging_directory", None))
-    if len(repo_info) == 0:
+    package_dirs = app_config.get_state("packaging_directories")
+    if package_dirs is None:
         return None
+    for directory in package_dirs:
+        repo_info = get_repository_info(directory)
+        if len(repo_info) == 0:
+            continue
+        git_log_info, returncode, failed = _call_git(
+            ["log", "-1", "--pretty=%B"],
+            path=directory,
+        )
+        repo_url = repo_info["repo_url"]
+        if isinstance(repo_url, str):
+            _url = (
+                repo_url if not repo_url.endswith(".git") else repo_url.rstrip(".git")
+            )
+        else:
+            _url = str(repo_url)
+        _code_info = {
+            "commitId": repo_info["commit_sha"],
+            "commitLink": os.path.join(_url, "commit", str(repo_info["commit_sha"])),
+        }
+        if not failed and returncode == 0 and isinstance(git_log_info, str):
+            _code_info["commitMessage"] = git_log_info.strip()
 
-    git_log_info, returncode, failed = _call_git(
-        ["log", "-1", "--pretty=%B"],
-        path=app_config.get_state("packaging_directory", None),
-    )
-    repo_url = repo_info["repo_url"]
-    if isinstance(repo_url, str):
-        _url = repo_url if not repo_url.endswith(".git") else repo_url.rstrip(".git")
-    else:
-        _url = str(repo_url)
-    _code_info = {
-        "commitId": repo_info["commit_sha"],
-        "commitLink": os.path.join(_url, "commit", str(repo_info["commit_sha"])),
-    }
-    if not failed and returncode == 0 and isinstance(git_log_info, str):
-        _code_info["commitMessage"] = git_log_info.strip()
+        return _code_info
 
-    return _code_info
+    return None

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/ob_metaflow_extensions.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ob-metaflow-extensions
-Version: 1.4.10
+Version: 1.4.12
 Summary: Outerbounds Platform Extensions for Metaflow
 Author: Outerbounds, Inc.
 License: Commercial

ob_metaflow_extensions-1.4.12/ob_metaflow_extensions.egg-info/requires.txt

@@ -0,0 +1,3 @@
+boto3
+kubernetes
+ob-metaflow==2.18.2.1

{ob_metaflow_extensions-1.4.10 → ob_metaflow_extensions-1.4.12}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_namespace_packages
 from pathlib import Path
 
 
-version = "1.4.10"
+version = "1.4.12"
 this_directory = Path(__file__).parent
 long_description = (this_directory / "README.md").read_text()
 
@@ -19,6 +19,6 @@ setup(
     install_requires=[
         "boto3",
         "kubernetes",
-        "ob-metaflow == 2.18.1.1",
+        "ob-metaflow == 2.18.2.1",
     ],
 )

ob_metaflow_extensions-1.4.10/ob_metaflow_extensions.egg-info/requires.txt

@@ -1,3 +0,0 @@
-boto3
-kubernetes
-ob-metaflow==2.18.1.1