Merge c8a5d4d311 into 6aa8192131

changelogs/fragments/82246-extend-http-retry-codes.yml

@@ -0,0 +1,3 @@
+---
+minor_changes:
+  - Make HTTP status codes to retry during galaxy API calls configurable (https://github.com/ansible/ansible/pull/82246).

lib/ansible/config/base.yml

@@ -1470,6 +1470,18 @@ GALAXY_REQUIRED_VALID_SIGNATURE_COUNT:
   - The number of signatures that must be successful during GPG signature verification while installing or verifying collections.
   - This should be a positive integer or all to indicate all signatures must successfully validate the collection.
   - Prepend + to the value to fail if no valid signatures are found for the collection.
+GALAXY_RETRY_HTTP_ERROR_CODES:
+  type: list
+  default:
+    - "429" # Too many requests
+    - "520" # Galaxy rate limit error code (Cloudflare unknown error)
+    - "502" # Common error from galaxy that may represent any number of transient backend issues
+  env:
+  - name: ANSIBLE_GALAXY_RETRY_HTTP_ERROR_CODES
+  ini:
+  - section: galaxy
+    key: retry_http_error_codes
+  description: List of HTTP Status error codes to retry on.
 HOST_KEY_CHECKING:
   # NOTE: constant not in use by ssh/paramiko plugins anymore, but they do support the same configuration sources
   # TODO: check non ssh connection plugins for use/migration

lib/ansible/galaxy/api.py

@@ -15,7 +15,6 @@ import tarfile
 import time
 import threading
 
-from http import HTTPStatus
 from http.client import BadStatusLine, IncompleteRead
 from urllib.error import HTTPError, URLError
 from urllib.parse import quote as urlquote, urlencode, urlparse, parse_qs, urljoin
@@ -35,11 +34,6 @@ from ansible.utils.path import makedirs_safe
 display = Display()
 _CACHE_LOCK = threading.Lock()
 COLLECTION_PAGE_SIZE = 100
-RETRY_HTTP_ERROR_CODES = {  # TODO: Allow user-configuration
-    HTTPStatus.TOO_MANY_REQUESTS,
-    520,  # Galaxy rate limit error code (Cloudflare unknown error)
-    HTTPStatus.BAD_GATEWAY,  # Common error from galaxy that may represent any number of transient backend issues
-}
 
 
 def cache_lock(func):
@@ -54,7 +48,12 @@ def should_retry_error(exception):
     # Note: cloud.redhat.com masks rate limit errors with 403 (Forbidden) error codes.
     # Since 403 could reflect the actual problem (such as an expired token), we should
     # not retry by default.
-    if isinstance(exception, GalaxyError) and exception.http_code in RETRY_HTTP_ERROR_CODES:
+    try:
+        retry_codes = [int(code) for code in C.GALAXY_RETRY_HTTP_ERROR_CODES]
+    except ValueError as e:
+        raise AnsibleError("Invalid value for HTTP retry code: %s. Only integer values are supported." % e)
+
+    if isinstance(exception, GalaxyError) and exception.http_code in retry_codes:
         return True
 
     if isinstance(exception, AnsibleError) and (orig_exc := getattr(exception, 'orig_exc', None)):

test/units/galaxy/test_api.py

@@ -20,7 +20,7 @@ import ansible.constants as C
 from ansible import context
 from ansible.errors import AnsibleError
 from ansible.galaxy import api as galaxy_api
-from ansible.galaxy.api import CollectionVersionMetadata, GalaxyAPI, GalaxyError
+from ansible.galaxy.api import CollectionVersionMetadata, GalaxyAPI, GalaxyError, should_retry_error
 from ansible.galaxy.token import BasicAuthToken, GalaxyToken, KeycloakToken
 from ansible.module_utils.common.file import S_IRWU_RG_RO
 from ansible.module_utils.common.text.converters import to_native, to_text
@@ -1354,3 +1354,11 @@ def test_clear_cache(cache_dir):
 def test_cache_id(url, expected):
     actual = galaxy_api.get_cache_id(url)
     assert actual == expected
+
+
+def test_should_retry_error_with_invalid_code(monkeypatch):
+    expected = "Invalid value for HTTP retry code"
+    monkeypatch.setattr(C, 'GALAXY_RETRY_HTTP_ERROR_CODES', ["429", "invalid"])
+
+    with pytest.raises(AnsibleError, match=expected):
+        should_retry_error(Exception())