DataSourceOracle: retry twice (and document why we retry at all) (#536)

2 files changed, 9 insertions, 4 deletions

diff --git a/cloudinit/sources/DataSourceOracle.py b/cloudinit/sources/DataSourceOracle.py
index 2354648b..20d6487d 100644
--- a/cloudinit/sources/DataSourceOracle.py
+++ b/cloudinit/sources/DataSourceOracle.py
@@ -297,7 +297,10 @@ def read_opc_metadata(*, fetch_vnics_data: bool = False):
             `fetch_vnics_data` is True, else None
 
     """
-    retries = 1
+    # Per Oracle, there are short windows (measured in milliseconds) throughout
+    # an instance's lifetime where the IMDS is being updated and may 404 as a
+    # result.  To work around these windows, we retry a couple of times.
+    retries = 2
 
     def _fetch(metadata_version: int, path: str) -> dict:
         headers = {
diff --git a/cloudinit/sources/tests/test_oracle.py b/cloudinit/sources/tests/test_oracle.py
index 2fb9a20a..7bd23813 100644
--- a/cloudinit/sources/tests/test_oracle.py
+++ b/cloudinit/sources/tests/test_oracle.py
@@ -481,9 +481,11 @@ class TestReadOpcMetadata:
         "v2_failure_count,v1_failure_count,expected_body,expectation",
         [
             (1, 0, json.loads(OPC_V2_METADATA), does_not_raise()),
-            (2, 0, json.loads(OPC_V1_METADATA), does_not_raise()),
-            (2, 1, json.loads(OPC_V1_METADATA), does_not_raise()),
-            (2, 2, None, pytest.raises(UrlError)),
+            (2, 0, json.loads(OPC_V2_METADATA), does_not_raise()),
+            (3, 0, json.loads(OPC_V1_METADATA), does_not_raise()),
+            (3, 1, json.loads(OPC_V1_METADATA), does_not_raise()),
+            (3, 2, json.loads(OPC_V1_METADATA), does_not_raise()),
+            (3, 3, None, pytest.raises(UrlError)),
         ]
     )
     def test_retries(self, v2_failure_count, v1_failure_count,