mlrun 1.7.0rc4__py3-none-any.whl → 1.7.0rc20__py3-none-any.whl

mlrun/runtimes/nuclio/function.py

@@ -22,9 +22,11 @@ from time import sleep
 import nuclio
 import nuclio.utils
 import requests
-import semver
 from aiohttp.client import ClientSession
 from kubernetes import client
+from mlrun_pipelines.common.mounts import VolumeMount
+from mlrun_pipelines.common.ops import deploy_op
+from mlrun_pipelines.mounts import mount_v3io, v3io_cred
 from nuclio.deploy import find_dashboard_url, get_deploy_status
 from nuclio.triggers import V3IOStreamTrigger
 
@@ -36,15 +38,11 @@ import mlrun.utils.helpers
 from mlrun.common.schemas import AuthInfo
 from mlrun.config import config as mlconf
 from mlrun.errors import err_to_str
-from mlrun.kfpops import deploy_op
 from mlrun.lists import RunList
 from mlrun.model import RunObject
 from mlrun.platforms.iguazio import (
-    VolumeMount,
-    mount_v3io,
     parse_path,
     split_path,
-    v3io_cred,
 )
 from mlrun.runtimes.base import FunctionStatus, RunError
 from mlrun.runtimes.pod import KubeResource, KubeResourceSpec
@@ -56,33 +54,9 @@ def validate_nuclio_version_compatibility(*min_versions):
     """
     :param min_versions: Valid minimum version(s) required, assuming no 2 versions has equal major and minor.
     """
-    parsed_min_versions = [
-        semver.VersionInfo.parse(min_version) for min_version in min_versions
-    ]
-    try:
-        parsed_current_version = semver.VersionInfo.parse(mlconf.nuclio_version)
-    except ValueError:
-        # only log when version is set but invalid
-        if mlconf.nuclio_version:
-            logger.warning(
-                "Unable to parse nuclio version, assuming compatibility",
-                nuclio_version=mlconf.nuclio_version,
-                min_versions=min_versions,
-            )
-        return True
-
-    parsed_min_versions.sort(reverse=True)
-    for parsed_min_version in parsed_min_versions:
-        if (
-            parsed_current_version.major == parsed_min_version.major
-            and parsed_current_version.minor == parsed_min_version.minor
-            and parsed_current_version.patch < parsed_min_version.patch
-        ):
-            return False
-
-        if parsed_current_version >= parsed_min_version:
-            return True
-    return False
+    return mlrun.utils.helpers.validate_component_version_compatibility(
+        "nuclio", *min_versions
+    )
 
 
 def min_nuclio_versions(*versions):
@@ -91,10 +65,7 @@ def min_nuclio_versions(*versions):
             if validate_nuclio_version_compatibility(*versions):
                 return function(*args, **kwargs)
 
-            message = (
-                f"{function.__name__} is supported since nuclio {' or '.join(versions)}, currently using "
-                f"nuclio {mlconf.nuclio_version}, please upgrade."
-            )
+            message = f"'{function.__qualname__}' function requires Nuclio v{' or v'.join(versions)} or higher"
             raise mlrun.errors.MLRunIncompatibleVersionError(message)
 
         return wrapper
@@ -291,6 +262,9 @@ class RemoteRuntime(KubeResource):
     def status(self, status):
         self._status = self._verify_dict(status, "status", NuclioStatus)
 
+    def pre_deploy_validation(self):
+        pass
+
     def set_config(self, key, value):
         self.spec.config[key] = value
         return self
@@ -342,17 +316,21 @@ class RemoteRuntime(KubeResource):
 
             git::
 
-                fn.with_source_archive("git://github.com/org/repo#my-branch",
-                        handler="main:handler",
-                        workdir="path/inside/repo")
+                fn.with_source_archive(
+                    "git://github.com/org/repo#my-branch",
+                    handler="main:handler",
+                    workdir="path/inside/repo",
+                )
 
             s3::
 
                 fn.spec.nuclio_runtime = "golang"
-                fn.with_source_archive("s3://my-bucket/path/in/bucket/my-functions-archive",
+                fn.with_source_archive(
+                    "s3://my-bucket/path/in/bucket/my-functions-archive",
                     handler="my_func:Handler",
                     workdir="path/inside/functions/archive",
-                    runtime="golang")
+                    runtime="golang",
+                )
         """
         self.spec.build.source = source
         # update handler in function_handler
@@ -540,11 +518,16 @@ class RemoteRuntime(KubeResource):
         :param project:    project name
         :param tag:        function tag
         :param verbose:    set True for verbose logging
-        :param auth_info:  service AuthInfo
+        :param auth_info:  service AuthInfo (deprecated and ignored)
         :param builder_env: env vars dict for source archive config/credentials e.g. builder_env={"GIT_TOKEN": token}
         :param force_build: set True for force building the image
         """
-        # todo: verify that the function name is normalized
+        if auth_info:
+            # TODO: remove in 1.9.0
+            warnings.warn(
+                "'auth_info' is deprecated for nuclio runtimes in 1.7.0 and will be removed in 1.9.0",
+                FutureWarning,
+            )
 
         old_http_session = getattr(self, "_http_session", None)
         if old_http_session:
@@ -561,15 +544,12 @@ class RemoteRuntime(KubeResource):
         if tag:
             self.metadata.tag = tag
 
-        save_record = False
         # Attempt auto-mounting, before sending to remote build
         self.try_auto_mount_based_on_config()
         self._fill_credentials()
         db = self._get_db()
         logger.info("Starting remote function deploy")
-        data = db.remote_builder(
-            self, False, builder_env=builder_env, force_build=force_build
-        )
+        data = db.deploy_nuclio_function(func=self, builder_env=builder_env)
         self.status = data["data"].get("status")
         self._update_credentials_from_remote_build(data["data"])
 
@@ -581,15 +561,18 @@ class RemoteRuntime(KubeResource):
         #       now, functions can be not exposed (using service type ClusterIP) and hence
         #       for BC we first try to populate the external invocation url, and then
         #       if not exists, take the internal invocation url
-        if self.status.external_invocation_urls:
+        if (
+            self.status.external_invocation_urls
+            and self.status.external_invocation_urls[0] != ""
+        ):
             self.spec.command = f"http://{self.status.external_invocation_urls[0]}"
-            save_record = True
-        elif self.status.internal_invocation_urls:
+        elif (
+            self.status.internal_invocation_urls
+            and self.status.internal_invocation_urls[0] != ""
+        ):
             self.spec.command = f"http://{self.status.internal_invocation_urls[0]}"
-            save_record = True
-        elif self.status.address:
+        elif self.status.address and self.status.address != "":
             self.spec.command = f"http://{self.status.address}"
-            save_record = True
 
         logger.info(
             "Successfully deployed function",
@@ -597,13 +580,11 @@ class RemoteRuntime(KubeResource):
             external_invocation_urls=self.status.external_invocation_urls,
         )
 
-        if save_record:
-            self.save(versioned=False)
+        self.save(versioned=False)
 
         return self.spec.command
 
     def _wait_for_function_deployment(self, db, verbose=False):
-        text = ""
         state = ""
         last_log_timestamp = 1
         while state not in ["ready", "error", "unhealthy"]:
@@ -611,7 +592,7 @@ class RemoteRuntime(KubeResource):
                 int(mlrun.mlconf.httpdb.logs.nuclio.pull_deploy_status_default_interval)
             )
             try:
-                text, last_log_timestamp = db.get_builder_status(
+                text, last_log_timestamp = db.get_nuclio_deploy_status(
                     self, last_log_timestamp=last_log_timestamp, verbose=verbose
                 )
             except mlrun.db.RunDBError:
@@ -769,10 +750,13 @@ class RemoteRuntime(KubeResource):
             runtime_env["MLRUN_NAMESPACE"] = mlconf.namespace
         if self.metadata.credentials.access_key:
             runtime_env[
-                mlrun.runtimes.constants.FunctionEnvironmentVariables.auth_session
+                mlrun.common.runtimes.constants.FunctionEnvironmentVariables.auth_session
             ] = self.metadata.credentials.access_key
         return runtime_env
 
+    def _get_serving_spec(self):
+        return None
+
     def _get_nuclio_config_spec_env(self):
         env_dict = {}
         external_source_env_dict = {}
@@ -958,6 +942,62 @@ class RemoteRuntime(KubeResource):
             data = json.loads(data)
         return data
 
+    def with_sidecar(
+        self,
+        name: str = None,
+        image: str = None,
+        ports: typing.Optional[typing.Union[int, list[int]]] = None,
+        command: typing.Optional[str] = None,
+        args: typing.Optional[list[str]] = None,
+    ):
+        """
+        Add a sidecar container to the function pod
+        :param name:    Sidecar container name.
+        :param image:   Sidecar container image.
+        :param ports:   Sidecar container ports to expose. Can be a single port or a list of ports.
+        :param command: Sidecar container command instead of the image entrypoint.
+        :param args:    Sidecar container command args (requires command to be set).
+        """
+        name = name or f"{self.metadata.name}-sidecar"
+        sidecar = self._set_sidecar(name)
+        if image:
+            sidecar["image"] = image
+
+        ports = mlrun.utils.helpers.as_list(ports)
+        # according to RFC-6335, port name should be less than 15 characters,
+        # so we truncate it if needed and leave room for the index
+        port_name = name[:13].rstrip("-_") if len(name) > 13 else name
+        sidecar["ports"] = [
+            {
+                "name": f"{port_name}-{i}",
+                "containerPort": port,
+                "protocol": "TCP",
+            }
+            for i, port in enumerate(ports)
+        ]
+
+        # if it is a redeploy, 'command' might be set with the previous invocation url.
+        # in this case, we don't want to use it as the sidecar command
+        if command and not command.startswith("http"):
+            sidecar["command"] = mlrun.utils.helpers.as_list(command)
+
+        if args and sidecar["command"]:
+            sidecar["args"] = mlrun.utils.helpers.as_list(args)
+
+        # populate the sidecar resources from the function spec
+        if self.spec.resources:
+            sidecar["resources"] = self.spec.resources
+
+    def _set_sidecar(self, name: str) -> dict:
+        self.spec.config.setdefault("spec.sidecars", [])
+        sidecars = self.spec.config["spec.sidecars"]
+        for sidecar in sidecars:
+            if sidecar["name"] == name:
+                return sidecar
+
+        sidecars.append({"name": name})
+        return sidecars[-1]
+
     def _trigger_of_kind_exists(self, kind: str) -> bool:
         if not self.spec.config:
             return False

mlrun/runtimes/nuclio/serving.py

@@ -14,17 +14,17 @@
 
 import json
 import os
+import warnings
 from copy import deepcopy
-from typing import Union
+from typing import TYPE_CHECKING, Optional, Union
 
 import nuclio
 from nuclio import KafkaTrigger
 
 import mlrun
 import mlrun.common.schemas
-from mlrun.datastore import parse_kafka_url
+from mlrun.datastore import get_kafka_brokers_from_dict, parse_kafka_url
 from mlrun.model import ObjectList
-from mlrun.model_monitoring.tracking_policy import TrackingPolicy
 from mlrun.runtimes.function_reference import FunctionReference
 from mlrun.secrets import SecretsStore
 from mlrun.serving.server import GraphServer, create_graph_server
@@ -43,6 +43,10 @@ from .function import NuclioSpec, RemoteRuntime
 
 serving_subkind = "serving_v2"
 
+if TYPE_CHECKING:
+    # remove this block in 1.9.0
+    from mlrun.model_monitoring import TrackingPolicy
+
 
 def new_v2_model_server(
     name,
@@ -303,12 +307,12 @@ class ServingRuntime(RemoteRuntime):
 
     def set_tracking(
         self,
-        stream_path: str = None,
-        batch: int = None,
-        sample: int = None,
-        stream_args: dict = None,
-        tracking_policy: Union[TrackingPolicy, dict] = None,
-    ):
+        stream_path: Optional[str] = None,
+        batch: Optional[int] = None,
+        sample: Optional[int] = None,
+        stream_args: Optional[dict] = None,
+        tracking_policy: Optional[Union["TrackingPolicy", dict]] = None,
+    ) -> None:
         """apply on your serving function to monitor a deployed model, including real-time dashboards to detect drift
            and analyze performance.
 
@@ -317,31 +321,17 @@ class ServingRuntime(RemoteRuntime):
         :param batch:           Micro batch size (send micro batches of N records at a time).
         :param sample:          Sample size (send only one of N records).
         :param stream_args:     Stream initialization parameters, e.g. shards, retention_in_hours, ..
-        :param tracking_policy: Tracking policy object or a dictionary that will be converted into a tracking policy
-                                object. By using TrackingPolicy, the user can apply his model monitoring requirements,
-                                such as setting the scheduling policy of the model monitoring batch job or changing
-                                the image of the model monitoring stream.
 
                                 example::
 
                                     # initialize a new serving function
                                     serving_fn = mlrun.import_function("hub://v2-model-server", new_name="serving")
-                                    # apply model monitoring and set monitoring batch job to run every 3 hours
-                                    tracking_policy = {'default_batch_intervals':"0 */3 * * *"}
-                                    serving_fn.set_tracking(tracking_policy=tracking_policy)
+                                    # apply model monitoring
+                                    serving_fn.set_tracking()
 
         """
-
         # Applying model monitoring configurations
         self.spec.track_models = True
-        self.spec.tracking_policy = None
-        if tracking_policy:
-            if isinstance(tracking_policy, dict):
-                # Convert tracking policy dictionary into `model_monitoring.TrackingPolicy` object
-                self.spec.tracking_policy = TrackingPolicy.from_dict(tracking_policy)
-            else:
-                # Tracking_policy is already a `model_monitoring.TrackingPolicy` object
-                self.spec.tracking_policy = tracking_policy
 
         if stream_path:
             self.spec.parameters["log_stream"] = stream_path
@@ -351,6 +341,14 @@ class ServingRuntime(RemoteRuntime):
             self.spec.parameters["log_stream_sample"] = sample
         if stream_args:
             self.spec.parameters["stream_args"] = stream_args
+        if tracking_policy is not None:
+            warnings.warn(
+                "The `tracking_policy` argument is deprecated from version 1.7.0 "
+                "and has no effect. It will be removed in 1.9.0.\n"
+                "To set the desired model monitoring time window and schedule, use "
+                "the `base_period` argument in `project.enable_model_monitoring()`.",
+                FutureWarning,
+            )
 
     def add_model(
         self,
@@ -367,8 +365,8 @@ class ServingRuntime(RemoteRuntime):
 
         Example, create a function (from the notebook), add a model class, and deploy::
 
-            fn = code_to_function(kind='serving')
-            fn.add_model('boost', model_path, model_class='MyClass', my_arg=5)
+            fn = code_to_function(kind="serving")
+            fn.add_model("boost", model_path, model_class="MyClass", my_arg=5)
             fn.deploy()
 
         only works with router topology, for nested topologies (model under router under flow)
@@ -450,7 +448,7 @@ class ServingRuntime(RemoteRuntime):
 
         example::
 
-            fn.add_child_function('enrich', './enrich.ipynb', 'mlrun/mlrun')
+            fn.add_child_function("enrich", "./enrich.ipynb", "mlrun/mlrun")
 
         :param name:   child function name
         :param url:    function/code url, support .py, .ipynb, .yaml extensions
@@ -489,11 +487,8 @@ class ServingRuntime(RemoteRuntime):
                         "worker_allocation_mode", "static"
                     )
 
-                if (
-                    stream.path.startswith("kafka://")
-                    or "kafka_bootstrap_servers" in stream.options
-                ):
-                    brokers = stream.options.get("kafka_bootstrap_servers")
+                brokers = get_kafka_brokers_from_dict(stream.options)
+                if stream.path.startswith("kafka://") or brokers:
                     if brokers:
                         brokers = brokers.split(",")
                     topic, brokers = parse_kafka_url(stream.path, brokers)
@@ -644,8 +639,7 @@ class ServingRuntime(RemoteRuntime):
             force_build=force_build,
         )
 
-    def _get_runtime_env(self):
-        env = super()._get_runtime_env()
+    def _get_serving_spec(self):
         function_name_uri_map = {f.name: f.uri(self) for f in self.spec.function_refs}
 
         serving_spec = {
@@ -658,9 +652,7 @@ class ServingRuntime(RemoteRuntime):
             "graph_initializer": self.spec.graph_initializer,
             "error_stream": self.spec.error_stream,
             "track_models": self.spec.track_models,
-            "tracking_policy": self.spec.tracking_policy.to_dict()
-            if self.spec.tracking_policy
-            else None,
+            "tracking_policy": None,
             "default_content_type": self.spec.default_content_type,
         }
 
@@ -668,8 +660,7 @@ class ServingRuntime(RemoteRuntime):
             self._secrets = SecretsStore.from_list(self.spec.secret_sources)
             serving_spec["secret_sources"] = self._secrets.to_serial()
 
-        env["SERVING_SPEC_ENV"] = json.dumps(serving_spec)
-        return env
+        return json.dumps(serving_spec)
 
     def to_mock_server(
         self,
@@ -735,8 +726,11 @@ class ServingRuntime(RemoteRuntime):
         example::
 
             serving_fn = mlrun.new_function("serving", image="mlrun/mlrun", kind="serving")
-            serving_fn.add_model('my-classifier',model_path=model_path,
-                                  class_name='mlrun.frameworks.sklearn.SklearnModelServer')
+            serving_fn.add_model(
+                "my-classifier",
+                model_path=model_path,
+                class_name="mlrun.frameworks.sklearn.SklearnModelServer",
+            )
             serving_fn.plot(rankdir="LR")
 
         :param filename:  target filepath for the image (None for the notebook)