azure/errors: introduce reportable errors for imds (#3647)

Always report failure to host, but report failure to fabric only
outside of _check_if_nic_is_primary() which is expected to fail if
nic is not primary.

Add two types of reportable errors for IMDS metadata:

- add ReportableErrorImdsUrlError() for url errors.

- add ReportableErrorImdsMetadataParsingException() for parsing errors.

Tweak ReportableError repr to be a bit friendlier.

Signed-off-by: Chris Patterson <cpatterson@microsoft.com>

2 files changed, 65 insertions, 11 deletions

diff --git a/cloudinit/sources/DataSourceAzure.py b/cloudinit/sources/DataSourceAzure.py
index b8087406..dc8b2a21 100644
--- a/cloudinit/sources/DataSourceAzure.py
+++ b/cloudinit/sources/DataSourceAzure.py
@@ -543,7 +543,7 @@ class DataSourceAzure(sources.DataSource):
 
         imds_md = {}
         if self._is_ephemeral_networking_up():
-            imds_md = self.get_metadata_from_imds()
+            imds_md = self.get_metadata_from_imds(report_failure=True)
 
         if not imds_md and ovf_source is None:
             msg = "No OVF or IMDS available"
@@ -575,7 +575,7 @@ class DataSourceAzure(sources.DataSource):
 
             md, userdata_raw, cfg, files = self._reprovision()
             # fetch metadata again as it has changed after reprovisioning
-            imds_md = self.get_metadata_from_imds()
+            imds_md = self.get_metadata_from_imds(report_failure=True)
 
         # Report errors if IMDS network configuration is missing data.
         self.validate_imds_network_metadata(imds_md=imds_md)
@@ -667,18 +667,33 @@ class DataSourceAzure(sources.DataSource):
         return crawled_data
 
     @azure_ds_telemetry_reporter
-    def get_metadata_from_imds(self) -> Dict:
-        retry_deadline = time() + 300
+    def get_metadata_from_imds(self, report_failure: bool) -> Dict:
+        start_time = time()
+        retry_deadline = start_time + 300
+        error_string: Optional[str] = None
+        error_report: Optional[errors.ReportableError] = None
         try:
             return imds.fetch_metadata_with_api_fallback(
                 retry_deadline=retry_deadline
             )
-        except (UrlError, ValueError) as error:
-            report_diagnostic_event(
-                "Ignoring IMDS metadata due to: %s" % error,
-                logger_func=LOG.warning,
+        except UrlError as error:
+            error_string = str(error)
+            duration = time() - start_time
+            error_report = errors.ReportableErrorImdsUrlError(
+                exception=error, duration=duration
             )
-            return {}
+        except ValueError as error:
+            error_string = str(error)
+            error_report = errors.ReportableErrorImdsMetadataParsingException(
+                exception=error
+            )
+
+        self._report_failure(error_report, host_only=not report_failure)
+        report_diagnostic_event(
+            "Ignoring IMDS metadata due to: %s" % error_string,
+            logger_func=LOG.warning,
+        )
+        return {}
 
     def clear_cached_attrs(self, attr_defaults=()):
         """Reset any cached class attributes to defaults."""
@@ -976,7 +991,7 @@ class DataSourceAzure(sources.DataSource):
         # Primary nic detection will be optimized in the future. The fact that
         # primary nic is being attached first helps here. Otherwise each nic
         # could add several seconds of delay.
-        imds_md = self.get_metadata_from_imds()
+        imds_md = self.get_metadata_from_imds(report_failure=False)
         if imds_md:
             # Only primary NIC will get a response from IMDS.
             LOG.info("%s is the primary nic", ifname)
diff --git a/cloudinit/sources/azure/errors.py b/cloudinit/sources/azure/errors.py
index ca902a03..966725b0 100644
--- a/cloudinit/sources/azure/errors.py
+++ b/cloudinit/sources/azure/errors.py
@@ -10,8 +10,11 @@ from datetime import datetime
 from io import StringIO
 from typing import Any, Dict, List, Optional
 
+import requests
+
 from cloudinit import version
 from cloudinit.sources.azure import identity
+from cloudinit.url_helper import UrlError
 
 LOG = logging.getLogger(__name__)
 
@@ -81,7 +84,12 @@ class ReportableError(Exception):
         )
 
     def __repr__(self) -> str:
-        return self.as_encoded_report()
+        return (
+            f"{self.__class__.__name__}("
+            f"reason={self.reason}, "
+            f"timestamp={self.timestamp}, "
+            f"supporting_data={self.supporting_data})"
+        )
 
 
 class ReportableErrorDhcpInterfaceNotFound(ReportableError):
@@ -99,6 +107,37 @@ class ReportableErrorDhcpLease(ReportableError):
         self.supporting_data["interface"] = interface
 
 
+class ReportableErrorImdsUrlError(ReportableError):
+    def __init__(self, *, exception: UrlError, duration: float) -> None:
+        # ConnectTimeout sub-classes ConnectError so order is important.
+        if isinstance(exception.cause, requests.ConnectTimeout):
+            reason = "connection timeout querying IMDS"
+        elif isinstance(exception.cause, requests.ConnectionError):
+            reason = "connection error querying IMDS"
+        elif isinstance(exception.cause, requests.ReadTimeout):
+            reason = "read timeout querying IMDS"
+        elif exception.code:
+            reason = "http error querying IMDS"
+        else:
+            reason = "unexpected error querying IMDS"
+
+        super().__init__(reason)
+
+        if exception.code:
+            self.supporting_data["http_code"] = exception.code
+
+        self.supporting_data["duration"] = duration
+        self.supporting_data["exception"] = repr(exception)
+        self.supporting_data["url"] = exception.url
+
+
+class ReportableErrorImdsMetadataParsingException(ReportableError):
+    def __init__(self, *, exception: ValueError) -> None:
+        super().__init__("error parsing IMDS metadata")
+
+        self.supporting_data["exception"] = repr(exception)
+
+
 class ReportableErrorUnhandledException(ReportableError):
     def __init__(self, exception: Exception) -> None:
         super().__init__("unhandled exception")