Very slow iteration over jobs with MLClient.jobs.list()

[dstoe]

• Package Name: azure.ai.ml.MLClient -> azure.ai.ml.operations._job_operations._append_tid_to_studio_url
• Package Version: 1.34.1
• Operating System: linux
• Python Version: 3.12

Describe the bug
Iterating over azure.ai.ml.MLClient.jobs.list() is unnecessarily slow, since there's a full authentication flow initiated on every job item just to find out the tenant-id. The iteration repeatedly calls azure.ai.ml.operations._job_operations._append_tid_to_studio_url which has in it's implementation:

                        # Extract the tenant id from the credential using PyJWT
                        decode = jwt.decode(
                            self._credential.get_token(*default_scopes).token,
                            options={"verify_signature": False, "verify_aud": False},
                        )
                        tid = decode["tid"]
                        formatted_tid = TID_FMT.format(tid)
                        studio_endpoint.endpoint = studio_url + formatted_tid

As explicitly noted in the get_token function, this function does not cache any data.

To Reproduce

from azure.ai.ml import MLClient
from azure.identity import AzureCliCredential


ml_client = MLClient(
    AzureCliCredential(),
    subscription_id="foo",
    workspace_name="bar",
    resource_group_name="foobar",
)

# Load 100 jobs and exit
for index, job_run in enumerate(ml_client.jobs.list()):
    if index > 100:
        break
    del job_run

Timing of the above with the vanilla-implementation in azure.ai.ml.operations._job_operations._append_tid_to_studio_url:

$ time python test.py 

real	1m17.016s
user	0m55.264s
sys	0m7.458s

With the code-block in azure.ai.ml.operations._job_operations._append_tid_to_studio_url replaced with hard-coding my tenant-id:

time python test.py 

real	0m13.239s
user	0m3.261s
sys	0m0.346s

Expected behavior
Looping over many jobs shouldn't be unnecessarily slow.

[github-actions]

Thank you for your feedback. Tagging and routing to the team member(s) best able to assist.

//cc: Amit Chauhan (@achauhan-scc) Jayesh Tanna (@jayesh-tanna) kingernupur

[github-actions]

🎯 Agentic Issue Triage

Summary: Iterating over MLClient.jobs.list() is extremely slow because _append_tid_to_studio_url calls credential.get_token() on every job item to extract the tenant ID, triggering a full authentication flow each time. Hardcoding the tenant ID reduces iteration time from ~77s to ~13s for 100 jobs.

📋 Issue Details

Package: azure-ai-ml v1.34.1
Affected API: azure.ai.ml.operations._job_operations._append_tid_to_studio_url
Scenarios:

• Iterating over a large list of jobs via MLClient.jobs.list() is ~6x slower than expected due to repeated auth token fetches
• Each iteration calls credential.get_token() which is documented to NOT cache results

Root ask: The tenant ID lookup should be cached so the authentication flow isn't repeated for every item in the jobs list.

🔎 Debugging / Reproduction Notes

The issue is in _append_tid_to_studio_url: it decodes a JWT token to extract the tid (tenant ID) on every call. Since get_token() does not cache, this triggers a full auth round-trip each time.

Suggested investigation steps:

1. Cache the tenant ID after the first successful extraction (e.g., as an instance variable on the JobOperations class)
2. Alternatively, extract the tenant ID once during MLClient initialization and pass it through
3. Consider using credential.get_token(...).token only once and storing the decoded tid for reuse across the operation's lifetime

🏷️ Label Confidence

• Category: Client — azure-ai-ml is a data-plane client library (not a management-plane azure-mgmt-* package)
• Service: Machine Learning — matches /sdk/ml/azure-ai-ml/ and the CODEOWNERS %Machine Learning entry
• Confidence: High — package name and service path are unambiguous

👥 Owner Routing

• Matched CODEOWNERS entry: # ServiceLabel: %Machine Learning (line 308) — bottom-to-top scan, first match on service label "Machine Learning"
• AzureSdkOwners: none listed
• ServiceOwners: @achauhan-scc, @jayesh-tanna, @kingernupur
• Routing action: Added "Service Attention" and "needs-team-attention" labels; @mentioned ServiceOwners via routing comment
• Scan notes: An earlier entry at line 305 also matched %Machine Learning with ServiceOwners @Azure/azure-ml-sdk @azureml-github``, but the bottom-to-top scan stops at the first match (line 308)

Generated by Agentic Triage for issue #48415 · 26.8 AIC · ⌖ 6.73 AIC · ⊞ 9.6K · ◷