Since I wasn't able to wait for this bug to be resolved. I worked around the problem by scheduling the dbt job item via the API and then polling until complete in a notebook.

The full notebook is too large to share here, but here is a basic example of how it works. Note that some of this leverages the API in a way that is not documented yet, but it appears to work fine.

import time
from urllib.parse import unquote, urlparse

import requests

FABRIC_API_BASE_URL = "https://api.fabric.microsoft.com/v1"
WORKSPACE_ID = "<workspace-id>"
DBT_ITEM_ID = "<dbt-project-item-id>"
ACCESS_TOKEN = "<bearer-token>"  # acquire via your identity provider (e.g. MSAL / service principal)

# The command to execute: dbt operation + arguments (mirrors `dbt run --select ...`)
execution_data = {
    "operation": "run",
    "arguments": {
        "select": "tag:daily",
        "fullRefresh": False,
        "threads": 4,
    },
}

headers = {
    "Authorization": f"Bearer {ACCESS_TOKEN}",
    "Content-Type": "application/json",
}

# 1. Submit the job
submit_url = f"{FABRIC_API_BASE_URL}/workspaces/{WORKSPACE_ID}/items/{DBT_ITEM_ID}/jobs/Execute/instances"
response = requests.post(submit_url, headers=headers, json={"executionData": execution_data})
response.raise_for_status()

# Fabric returns the job instance location (may be relative or absolute)
location = response.headers["Location"]
job_instance_id = unquote(urlparse(location).path.rstrip("/").split("/")[-1])

# 2. Poll the job instance until it reaches a terminal state
instance_url = f"{FABRIC_API_BASE_URL}/workspaces/{WORKSPACE_ID}/items/{DBT_ITEM_ID}/jobs/instances/{job_instance_id}"

while True:
    status_response = requests.get(instance_url, headers=headers)
    status_response.raise_for_status()
    status = status_response.json().get("status", "").upper()

    if status in {"COMPLETED", "SUCCEEDED"}:
        print("dbt job succeeded")
        break
    if status in {"FAILED", "CANCELLED"}:
        raise RuntimeError(f"dbt job failed with status: {status}")

    time.sleep(20)  # poll interval

Hi,

Hi Michael,

This sounds like a product limitation or serialization issue with the current Fabric Git integration rather than something you're doing incorrectly. If the Operation setting (Run, Build, Compile) is configured in the UI but is not persisted in the repository artifacts, deployments through Fabric CI/CD can result in an invalid activity definition in the target workspace.

The fact that manually adding "operation": "build" to the pipeline JSON resolves the runtime error, but Fabric subsequently removes it during the next commit, suggests the Git representation of the dbt activity is not fully round-tripping all properties. In that case, the source of truth in the UI and the source-controlled artifact are becoming inconsistent.

A few things I'd recommend checking:

• Verify all environments are running the same Fabric feature set and Git integration experience.
• Confirm whether the issue occurs with newly created dbt activities or only existing ones.
• Compare the JSON generated before and after changing the Operation value to see whether any related properties are being persisted.
• Test deploying the pipeline directly between workspaces without Git to determine whether the issue is specifically tied to source control serialization.

If the behavior is reproducible, it is likely a bug and worth raising with Microsoft Support. Be sure to include:

• The pipeline JSON stored in Git
• Screenshots of the Operation setting in the UI
• Deployment logs showing the missing Operation error
• The Fabric workspace and Git integration experience being used

In the meantime, if Fabric continues to remove the property during commits, any manual JSON edits are unlikely to be a sustainable workaround because the platform will overwrite them during synchronization. Hopefully others in the community can confirm whether they're seeing the same behavior with dbt activities and Git-managed deployments.

Thanks,

Manoj Annavajjala

Thank you Prince0011  , ryan-schofield  and sannavajjala   for the immediate response.

Hi michaelfabric ,

Thanks for reaching out to fabric community.

Could you please let us know whether the responses provided above helped you in resolving the issue and let us know if you have any additional questions, we are happy to address. 

Thanks!!