ob-metaflow-extensions 1.1.147__tar.gz → 1.1.149__tar.gz

{ob-metaflow-extensions-1.1.147 → ob-metaflow-extensions-1.1.149}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ob-metaflow-extensions
-Version: 1.1.147
+Version: 1.1.149
 Summary: Outerbounds Platform Extensions for Metaflow
 Author: Outerbounds, Inc.
 License: Commercial

{ob-metaflow-extensions-1.1.147 → ob-metaflow-extensions-1.1.149}/metaflow_extensions/outerbounds/plugins/secrets/secrets.py

@@ -4,13 +4,48 @@ from typing import Dict
 import base64
 import json
 import requests
+import random
 import time
+import sys
 
 
 class OuterboundsSecretsException(Exception):
     pass
 
 
+def _api_server_get(*args, conn_error_retries=2, **kwargs):
+    """
+    There are two categories of errors that we need to handle when dealing with any API server.
+    1. HTTP errors. These are are errors that are returned from the API server.
+        - How to handle retries for this case will be application specific.
+    2. Errors when the API server may not be reachable (DNS resolution / network issues)
+        - In this scenario, we know that something external to the API server is going wrong causing the issue.
+        - Failing pre-maturely in the case might not be the best course of action since critical user jobs might crash on intermittent issues.
+        - So in this case, we can just planely retry the request.
+
+    This function handles the second case. It's a simple wrapper to handle the retry logic for connection errors.
+    If this function is provided a `conn_error_retries` of 5, then the last retry will have waited 32 seconds.
+    Generally this is a safe enough number of retries after which we can assume that something is really broken. Until then,
+    there can be intermittent issues that would resolve themselves if we retry gracefully.
+    """
+    _num_retries = 0
+    noise = random.uniform(-0.5, 0.5)
+    while _num_retries < conn_error_retries:
+        try:
+            return requests.get(*args, **kwargs)
+        except requests.exceptions.ConnectionError:
+            if _num_retries <= conn_error_retries - 1:
+                # Exponential backoff with 2^(_num_retries+1) seconds
+                time.sleep((2 ** (_num_retries + 1)) + noise)
+                _num_retries += 1
+            else:
+                print(
+                    "[@secrets] Failed to connect to the API server. ",
+                    file=sys.stderr,
+                )
+                raise
+
+
 class OuterboundsSecretsApiResponse:
     def __init__(self, response):
         self.response = response
@@ -133,7 +168,9 @@ class OuterboundsSecretsProvider(SecretsProvider):
         retryable_status_codes = [409]
         json_payload = json.dumps(payload)
         for attempt in range(2):  # 0 = initial attempt, 1-2 = retries
-            response = requests.get(url, data=json_payload, headers=request_headers)
+            response = _api_server_get(
+                url, data=json_payload, headers=request_headers, conn_error_retries=5
+            )
             if response.status_code not in retryable_status_codes:
                 break
 
@@ -141,7 +178,6 @@ class OuterboundsSecretsProvider(SecretsProvider):
                 sleep_time = 0.5 * (attempt + 1)
                 time.sleep(sleep_time)
 
-        response = requests.get(url, data=json_payload, headers=request_headers)
         self._handle_error_response(response)
         return OuterboundsSecretsApiResponse(response.json())
 

{ob-metaflow-extensions-1.1.147 → ob-metaflow-extensions-1.1.149}/metaflow_extensions/outerbounds/plugins/snowflake/snowflake.py

@@ -7,9 +7,10 @@ from metaflow.metaflow_config import SERVICE_URL
 from metaflow.metaflow_config_funcs import init_config
 from typing import Dict
 from os import environ
-
+import sys
 import json
 import requests
+import random
 import time
 
 
@@ -75,7 +76,9 @@ def get_snowflake_token(user: str = "", role: str = "", integration: str = "") -
     }
     json_payload = json.dumps(payload)
     headers = provisioner.get_service_auth_header()
-    response = requests.get(snowflake_token_url, data=json_payload, headers=headers)
+    response = _api_server_get(
+        snowflake_token_url, data=json_payload, headers=headers, conn_error_retries=5
+    )
     response.raise_for_status()
     return response.json()["token"]
 
@@ -150,6 +153,39 @@ def connect(user: str = "", role: str = "", integration: str = "", **kwargs):
         )
 
 
+def _api_server_get(*args, conn_error_retries=2, **kwargs):
+    """
+    There are two categories of errors that we need to handle when dealing with any API server.
+    1. HTTP errors. These are are errors that are returned from the API server.
+        - How to handle retries for this case will be application specific.
+    2. Errors when the API server may not be reachable (DNS resolution / network issues)
+        - In this scenario, we know that something external to the API server is going wrong causing the issue.
+        - Failing pre-maturely in the case might not be the best course of action since critical user jobs might crash on intermittent issues.
+        - So in this case, we can just planely retry the request.
+
+    This function handles the second case. It's a simple wrapper to handle the retry logic for connection errors.
+    If this function is provided a `conn_error_retries` of 5, then the last retry will have waited 32 seconds.
+    Generally this is a safe enough number of retries after which we can assume that something is really broken. Until then,
+    there can be intermittent issues that would resolve themselves if we retry gracefully.
+    """
+    _num_retries = 0
+    noise = random.uniform(-0.5, 0.5)
+    while _num_retries < conn_error_retries:
+        try:
+            return requests.get(*args, **kwargs)
+        except requests.exceptions.ConnectionError:
+            if _num_retries <= conn_error_retries - 1:
+                # Exponential backoff with 2^(_num_retries+1) seconds
+                time.sleep((2 ** (_num_retries + 1)) + noise)
+                _num_retries += 1
+            else:
+                print(
+                    "[@snowflake] Failed to connect to the API server. ",
+                    file=sys.stderr,
+                )
+                raise
+
+
 class Snowflake:
     def __init__(
         self, user: str = "", role: str = "", integration: str = "", **kwargs
@@ -273,7 +309,9 @@ class SnowflakeIntegrationProvisioner:
         retryable_status_codes = [409]
         json_payload = json.dumps(payload)
         for attempt in range(2):  # 0 = initial attempt, 1-2 = retries
-            response = requests.get(url, data=json_payload, headers=request_headers)
+            response = _api_server_get(
+                url, data=json_payload, headers=request_headers, conn_error_retries=5
+            )
             if response.status_code not in retryable_status_codes:
                 break
 
@@ -281,7 +319,9 @@ class SnowflakeIntegrationProvisioner:
                 sleep_time = 0.5 * (attempt + 1)
                 time.sleep(sleep_time)
 
-        response = requests.get(url, data=json_payload, headers=request_headers)
+        response = _api_server_get(
+            url, data=json_payload, headers=request_headers, conn_error_retries=5
+        )
         self._handle_error_response(response)
         return response.json()
 

{ob-metaflow-extensions-1.1.147 → ob-metaflow-extensions-1.1.149}/ob_metaflow_extensions.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ob-metaflow-extensions
-Version: 1.1.147
+Version: 1.1.149
 Summary: Outerbounds Platform Extensions for Metaflow
 Author: Outerbounds, Inc.
 License: Commercial

ob-metaflow-extensions-1.1.149/ob_metaflow_extensions.egg-info/requires.txt

@@ -0,0 +1,3 @@
+boto3
+kubernetes
+ob-metaflow==2.15.10.1

{ob-metaflow-extensions-1.1.147 → ob-metaflow-extensions-1.1.149}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_namespace_packages
 from pathlib import Path
 
 
-version = "1.1.147"
+version = "1.1.149"
 this_directory = Path(__file__).parent
 long_description = (this_directory / "README.md").read_text()
 
@@ -18,6 +18,6 @@ setup(
     install_requires=[
         "boto3",
         "kubernetes",
-        "ob-metaflow == 2.15.7.2",
+        "ob-metaflow == 2.15.10.1",
     ],
 )

ob-metaflow-extensions-1.1.147/ob_metaflow_extensions.egg-info/requires.txt

@@ -1,3 +0,0 @@
-boto3
-kubernetes
-ob-metaflow==2.15.7.2