fix(gcp): use redis backend for celery autoscaler when broker is redis

[arniechops]

Summary

• When the celery broker is redis, the autoscaler calls celery_app() without specifying backend_protocol, which defaults to "s3"
• The S3 backend path tries to create a boto3 AWS session, which fails on GCP with ProfileNotFound: The config profile (default) could not be found
• This causes celery-autoscaler-redis to CrashLoopBackOff on GCP clusters
• Fix: pass backend_protocol="redis" when broker is redis (matching how the servicebus path already passes backend_protocol="abs")

Test plan

• Deploy to GCP cluster and verify celery-autoscaler-redis-0 starts without crash-looping
• Verify celery task autoscaling still works on AWS (no regression — AWS uses SQS broker, not redis)

Greptile Summary

This PR fixes a GCP-specific crash in the celery-autoscaler-redis pod by passing backend_protocol="redis" when constructing the Celery inspect app for the Redis broker path. Previously, the code fell back to the default backend_protocol="s3", which triggered a boto3 AWS session initialization (via _get_backend_url_and_conf) that fails on GCP with ProfileNotFound.

Key changes:

• In the broker_type == "redis" branch of main(), celery_app() is now called with backend_protocol="redis", consistent with how the servicebus branch already passes backend_protocol="abs".
• The fix is minimal and targeted — no other code paths are affected. AWS deployments using SQS are unaffected.
• The _get_backend_url_and_conf function shows that "redis" backend resolves to get_redis_endpoint(1) (db index 1), which is a valid, always-available endpoint in a Redis-based deployment.

Confidence Score: 5/5

Safe to merge — targeted one-line fix that correctly mirrors the existing servicebus pattern and avoids a well-understood crash on GCP.

The change is small, well-motivated, and aligns with the existing convention (servicebus uses backend_protocol='abs'). The backend_protocol='redis' path in _get_backend_url_and_conf is straightforward and does not touch AWS credentials. The only remaining feedback is a P2 style suggestion to add a clarifying comment. No logic or correctness issues found.

No files require special attention — the single changed file is straightforward.

Important Files Changed

Filename	Overview
model-engine/model_engine_server/core/celery/celery_autoscaler.py	Adds backend_protocol='redis' when constructing the Celery inspect app for the Redis broker path, mirroring the existing backend_protocol='abs' pattern used for the Service Bus path. This prevents a boto3 AWS profile lookup from crashing the autoscaler on GCP.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[main: autoscaler_broker] --> B{broker_type?}
    B -- redis --> C["celery_app with backend_protocol='redis'"]
    B -- sqs --> D["inspect = empty, reads directly from SQS"]
    B -- servicebus --> E["celery_app with backend_protocol='abs'"]

    C --> F[_get_backend_url_and_conf]
    E --> F

    F -- redis --> G["get_redis_endpoint db=1, no AWS session needed"]
    F -- s3 old default --> H{cloud_provider?}
    H -- aws --> I["session + boto3, works on AWS"]
    H -- gcp --> J["boto3 default profile, ProfileNotFound crash on GCP"]
    F -- abs --> K["azureblockblob URL, no AWS session needed"]

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: model-engine/model_engine_server/core/celery/celery_autoscaler.py
Line: 606-614

Comment:
**Add comment explaining the workaround**

The `backend_protocol="redis"` is a non-obvious workaround — without a comment, future readers won't know why it's needed here but not defaulted elsewhere. The `servicebus` block (line 622) has the same gap. A short inline comment (similar to the PR description) would make the intent clear and prevent the fix from being accidentally reverted.

```suggestion
                app=celery_app(
                    None,
                    broker_type=broker_type,
                    task_visibility=db_index,
                    aws_role=aws_profile,
                    # Use redis backend to avoid S3/boto3 session initialization,
                    # which fails on GCP with ProfileNotFound when no AWS credentials exist.
                    backend_protocol="redis",
                )
```

**Rule Used:** Add comments to explain complex or non-obvious log... ([source](https://app.greptile.com/review/custom-context?memory=928586f9-9432-435e-a385-026fa49318a2))

**Learnt From**
[scaleapi/scaleapi#126958](https://github.com/scaleapi/scaleapi/pull/126958)

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (2): Last reviewed commit: "fix" | Re-trigger Greptile}