feat(azure): report failure if missing customdata by cadejacobson · Pull Request #6779 · canonical/cloud-init

cadejacobson · 2026-03-02T20:57:52Z

Proposed Commit Message

feat(azure): report failure if missing customdata

Add experimental Azure datasource config key
experimental_fail_on_missing_customdata with a default of
false and use it to control reporting when OVF custom data is
missing despite IMDS indicating that hasCustomData=true.

Additional Context

On Azure, IMDS contains a field indicating whether the Custom Data field is expected to be found in the provisioning media. In some rare cases, this field is set to True but there is no custom data found in the provisioning media. In the future, this will be a failure reported to Azure by removing the experimental flag.

Merge type

Squash merge using "Proposed Commit Message"

When ovf-env.xml is present but does not contain custom data, yet IMDS indicates that custom data was provided to the VM (via hasCustomData), report a provisioning failure. This helps surface cases where custom data is silently lost during provisioning. The behavior is gated behind a new feature flag, EXPERIMENTAL_FAIL_ON_MISSING_CUSTOMDATA, which is disabled by default while undergoing scale testing. Once validated, it will be renamed and enabled for new distro releases. Changes: - Add EXPERIMENTAL_FAIL_ON_MISSING_CUSTOMDATA feature flag - Add _hascustomdata_from_imds() helper to query IMDS metadata - Report ReportableErrorImdsInvalidMetadata when custom data is expected but missing from OVF provisioning media

github-actions · 2026-03-17T00:35:31Z

Hello! Thank you for this proposed change to cloud-init. This pull request is now marked as stale as it has not seen any activity in 14 days. If no activity occurs within the next 7 days, this pull request will automatically close.

If you are waiting for code review and you are seeing this message, apologies! Please reply, tagging blackboxsw, and he will ensure that someone takes a look soon.

(If the pull request is closed and you would like to continue working on it, please do tag blackboxsw to reopen it.)

peytonr18

LGTM!

cjp256

LGTM, github won't let me resolve last comments.

Thanks!!

cadejacobson · 2026-04-16T15:33:20Z

@blackboxsw @holmanb
This PR should be ready for review. Thanks!

blackboxsw

Thank you @cadejacobson for this proposal. It looks good. I have a few minor suggestions (tests, comments and typing).

I also have a couple of questions about the hasCustomData expected values and whether we really need an ovf_source local variable.

Otherwise this looks good to me.

blackboxsw · 2026-04-27T18:22:28Z

        return None


+def _hascustomdata_from_imds(imds_data) -> Optional[bool]:


Please type hint params when we add new functions/methods.

Suggested change

def _hascustomdata_from_imds(imds_data) -> Optional[bool]:

def _hascustomdata_from_imds(imds_data: Dict) -> Optional[bool]:

Updated to include the type hint!

blackboxsw · 2026-04-27T18:26:55Z

        return None


+def _hascustomdata_from_imds(imds_data) -> Optional[bool]:


Let's make this a strict boolean and avoid Optional[bool] truthiness behavior.
We don't need None, let's use return False instead.

Suggested change

def _hascustomdata_from_imds(imds_data) -> Optional[bool]:

def _hascustomdata_from_imds(imds_data) -> bool:

blackboxsw · 2026-04-27T18:34:06Z


+def _hascustomdata_from_imds(imds_data) -> Optional[bool]:
+    try:
+        return imds_data["extended"]["compute"]["hasCustomData"]


Is it possible for this key to be present but with a value not in (True, False)? I can't seem to find Azure docs on this field, but I can state that our instance data represents what appears to be a strict bool value ```
"extended": {
"compute": {
"hasCustomData": false,

You are correct. If present, this field should only be in the values of (True, False). There is a very very small chance that IMDS returns malformed JSON, at which point the field could be either incorrectly populated or not present, but this is an unexpected case.

We have been using None to indicate the difference between False and not-present. Programmatically None and False are treated the same with the if statement, the but reported error has more clarity in case there's a situation where IMDS unexpectedly is missing data.

The alternative I suppose is to specifically raise ReportableErrorImdsMissingHasCustomData or something. The thing is that if IMDS metadata was fetched using the fallback api-version, we might not see it and force an error where there is none.

blackboxsw · 2026-04-27T18:34:19Z

+    try:
+        return imds_data["extended"]["compute"]["hasCustomData"]
+    except KeyError:
+        return None


Suggested change

return None

return False

blackboxsw · 2026-04-27T18:36:10Z

+            (True, None),
+            (True, "myCustomData"),
+            (True, ""),
+            (False, None),


It appears that hasCustomData in IMDS compute are boolean values. Why is this test setting None, non-empty and empty strings for this value?are we testing.

I believe this test should probably only be checking hasCustomData in (True, False). Do I have an incorrect read on this?

In this test, we specify the hasCustomData on the left param and the customData field on the right. The hasCustomData does only stay in the group (True, False), but for the CustomData, we do specify None as well as empty and non-empty strings to verify that we catch various styles of empty CustomData that could be found in the OVF source.

yeah the right is the actual customData found on provisioning media (ovf-env.xml), this is just enumerating the cases the possible values seen there (expected or not)

blackboxsw · 2026-04-27T18:40:08Z


-        # only use userdata from imds if OVF did not provide custom data
-        # userdata provided by IMDS is always base64 encoded
+        # Only use userdata from imds if OVF did not provide custom data.


Please retain IMDS capitalization convention through comments where possible.

Suggested change

# Only use userdata from imds if OVF did not provide custom data.

# Only use userdata from IMDS if OVF did not provide custom data.

Updated to improve this capitalization!

blackboxsw · 2026-04-27T18:47:11Z

            report_diagnostic_event(msg)
            raise sources.InvalidMetaDataException(msg)

+        self.seed = ovf_source or "IMDS"


I'm not certain why we need a separate self.seed and ovf_source given the various call-sites throughout DataSourceAzure. They appear to be synonymous and we may be able to avoid using ovf_source in general right. What do Azure devs think?

i agree it'd be nice to standardize on self.seed

Replaced the references to ovf_source with self.seed in the DataSourceAzure.py file. All Azure tests continue to pass as expected.

blackboxsw · 2026-04-27T18:51:26Z

            if cfg.get("ProvisionGuestProxyAgent"):
                self._check_azure_proxy_agent_status()

+            ovf_source = "IMDS"


It feels like setting this ovf_source to IMDS may be more conspicuously placed at the open of the if pps_type != PPSType.NONE: conditional to indicate the context of this section more clearly. Down on line 768 it feels more 'hidden'. Depending on what folks decide related to the question about with self.seed instance variable versus ovf_source local var this specific assignment may become self.seed = "IMDS".

on a related note, I've never really looked closely at how it was used until now: https://github.com/cadejacobson/cloud-init/blob/26eb69fb1b2bb73797716c3e19d65aeabd7361ac/cloudinit/sources/DataSourceAzure.py#L361

Moved this assignment to the top of the if branch, and replaced the ovf_source with self.seed.

cadejacobson · 2026-05-01T18:05:39Z

Thank you @cadejacobson for this proposal. It looks good. I have a few minor suggestions (tests, comments and typing).

I also have a couple of questions about the hasCustomData expected values and whether we really need an ovf_source local variable.

Otherwise this looks good to me.

@blackboxsw Thanks for the review! I will work with Chris Patterson when he is back on Monday to ensure I give you the most accurate information I can on the cases of malformed IMDS data. After that, I will get the rest of the comments addressed. Thanks!

cadejacobson added 5 commits March 2, 2026 12:56

azure: test missing customdata path

b5a68cb

Move location of experimental flag

044f1eb

Remove unused feature flag

6c3d1f3

Linting

9418021

cadejacobson changed the title ~~Azure missing customdata~~ feat(azure): report failure if missing customdata Mar 2, 2026

github-actions Bot added the stale-pr Pull request is stale; will be auto-closed soon label Mar 17, 2026

cadejacobson added 2 commits March 17, 2026 07:48

Merge main into branch azure-missing-customdata

a042b66

Linting

a598ca1

github-actions Bot removed the stale-pr Pull request is stale; will be auto-closed soon label Mar 18, 2026