Happy to help dig in. A few things can cause task-level logs and the graph to disappear. Let’s quickly narrow it down. First, a few questions: - Which Prefect version are you running (2.x or 3.x)? If unsure, run:

prefect version

- Prefect Cloud or self-hosted server? - Did this start after an upgrade or code change? - Do you see flow logs but no task logs, or are all logs missing? - In your flow code, are you calling tasks with

my_task()

(or

.submit()

), or using

my_task.fn()

? Quick checks you can do now - In the UI (Flow Run page → Graph tab), ensure the filters for “Tasks” are enabled. Sometimes the Graph filter hides tasks. - Verify the client-side logging settings being used at runtime:

prefect config view --show-sources --show-defaults

Confirm: -

PREFECT_LOGGING_TO_API_ENABLED

is True (otherwise logs won’t reach the UI) - If you rely on

print

, either set

@task(log_prints=True)

or set

PREFECT_LOGGING_LOG_PRINTS=True

- Ensure your code is creating task runs: - Calling

my_task.fn(...)

bypasses orchestration and will not create task runs/logs or appear on the graph. Use

my_task(...)

or

my_task.submit(...)

inside a

@flow

. Self-hosted server checks (Prefect 3.x) - Make sure the task run recorder service is enabled on your API: -

PREFECT_SERVER_SERVICES_TASK_RUN_RECORDER_ENABLED=true

- If this is disabled, task runs/logs/graph data won’t be persisted. - If your flows have very large numbers of tasks, the server may cap what’s returned for the Graph: -

PREFECT_SERVER_FLOW_RUN_GRAPH_MAX_NODES

and

PREFECT_SERVER_FLOW_RUN_GRAPH_MAX_ARTIFACTS

control limits (defaults are high, e.g., 10,000) Minimal code sanity check Try running this to confirm tasks are being recorded/logged:

from prefect import flow, task, get_run_logger

@task(log_prints=True, task_run_name="demo-task")
def t(x):
    print(f"print from task: {x}")
    logger = get_run_logger()
    <http://logger.info|logger.info>(f"log from task: {x}")
    return x * 2

@flow(log_prints=True)
def f():
    a = t(1)            # creates a task run
    b = t.submit(2)     # also creates a task run
    # c = t.fn(3)       # would NOT create a task run (bypasses orchestration)
    return a, b.result()

if __name__ == "__main__":
    f()

After running, open the flow run in the UI and check: - Graph tab shows two task nodes - Task-level logs appear under each task run Common causes we see - Using

task.fn(...)

instead of `task(...)`/`.submit(...)` (no task runs/logs/graph) -

PREFECT_LOGGING_TO_API_ENABLED

is False in the runtime environment (no logs in UI) - Relying on

print

without

log_prints=True

(prints won’t appear as logs) - Self-hosted API with task run recorder service disabled - UI Graph filters hiding tasks Docs and references - Tasks concept: Prefect 3.x Tasks - Settings reference: Prefect 3.x Settings - Logging customization: Logging customization If you can share: - Prefect version - Cloud vs self-hosted - A short snippet of how you call your tasks - Output of

prefect config view --show-sources --show-defaults

for the logging-related keys …I can pinpoint the cause quickly and suggest the exact fix.

Thanks — version skew between the API (server) and the runtime (worker/flow pods) is the most common cause of “no task logs / no task graph” in self‑hosted 3.x. Why this breaks in 3.x - Task runs and the Graph in 3.x are built from events the client sends to the API. If the API can’t parse those events (older server) or the event recorder service is off, you won’t see task runs, task logs, or the graph. - Major versions must match (2.x vs 3.x will not work). Even within 3.x, running a newer client against an older server can degrade features like task events/graph. Let’s verify and fix 1) Confirm exact versions on both sides - Server/API (from inside the API pod):

prefect version
curl -sSL $PREFECT_API_URL/admin/version
prefect config view --show-sources --show-defaults

- Worker runtime (the pod that runs your flow):

prefect version
prefect config view --show-sources --show-defaults

Please share the two Prefect versions (e.g., 3.4.2 on server, 3.6.0 on worker), and whether any are on 2.x. 2) Align versions - Pin server and worker images to the same Prefect 3.x minor/patch (e.g., 3.6.0 on both). Avoid floating tags like 3-latest in production. - In Kubernetes manifests/Helm, set the same image tag (e.g., prefecthq/prefect:3.6.0) for both the server API and for the worker/flow job image. 3) Check required services and connectivity on the server - Ensure the task run recorder is enabled on the API: -

PREFECT_SERVER_SERVICES_TASK_RUN_RECORDER_ENABLED=true

- Verify the API can accept WebSocket connections from your workers (load balancer/proxy must allow WS). Blocked WS = no task events created. - For very large flows, verify graph limits aren’t filtering nodes: -

PREFECT_SERVER_FLOW_RUN_GRAPH_MAX_NODES

(default is large, 10,000) 4) Check logging settings on the worker/runtime - Confirm the worker/flow pods are actually sending logs to the API: -

PREFECT_LOGGING_TO_API_ENABLED=true

(default is true) - If you rely on print(), enable: - Per task or flow:

@task(log_prints=True)

/

@flow(log_prints=True)

- Or global:

PREFECT_LOGGING_LOG_PRINTS=true

5) Quick sanity test Run a tiny flow to verify task runs and task logs appear:

from prefect import flow, task, get_run_logger

@task(log_prints=True, task_run_name="demo-task")
def t(x):
    print(f"print from task: {x}")
    get_run_logger().info(f"log from task: {x}")
    return x * 2

@flow(log_prints=True)
def f():
    a = t(1)            # creates a task run
    b = t.submit(2)     # creates a task run
    # t.fn(3) would bypass orchestration (no task run/logs/graph)
    return a, b.result()

if __name__ == "__main__":
    f()

Open the flow run in the UI → Graph tab. You should see two task nodes and their logs. If it’s still blank after aligning versions: - Check API pod logs for TaskRunRecorder messages or WS errors. - Check the flow pod logs for version check warnings or WS failures. - Verify UI filters in the Graph tab aren’t hiding tasks. References - Tasks concept: Prefect 3.x Tasks - Settings reference (logging, services, graph limits): Prefect 3.x Settings - Logging customization: Logging customization If you can share: - The exact Prefect versions on server and worker - Whether any side is on 2.x - Output of

prefect config view --show-sources --show-defaults

for logging/service keys …I can give you the precise steps to align and restore task logs/graph.

Got it. Best path is to upgrade your AKS worker and its flow-run job image(s) to match the server at 3.6.20. Here’s how to align everything: 1) Update the worker Deployment image (the poller) - Change the container image for your Kubernetes worker Deployment to Prefect 3.6.20:

kubectl -n <namespace> set image deploy/<worker-deployment-name> <container-name>=prefecthq/prefect:3.6.20
kubectl -n <namespace> rollout status deploy/<worker-deployment-name>
kubectl -n <namespace> exec deploy/<worker-deployment-name> -- prefect version

Tip: If you use a custom image, ensure it installs

prefect==3.6.20

. 2) Ensure flow-run jobs also use Prefect 3.6.20 The Kubernetes worker launches a new pod per flow run. That pod image must have the same Prefect version. There are two common ways this is set: - Via the work pool base job template (most common) - Export the default template, edit, and update:

prefect work-pool get-default-base-job-template --type kubernetes --file k8s.json
  # edit k8s.json → set "image": "prefecthq/prefect:3.6.20"
  prefect work-pool update "<your-pool-name>" --base-job-template k8s.json

- Via your deployment’s infrastructure overrides (if you set a custom image per deployment) - Update the image there to

prefecthq/prefect:3.6.20

or your custom image that has

prefect==3.6.20

. - If you build your own image, pin Prefect inside it:

FROM python:3.11-slim
  RUN pip install --no-cache-dir 'prefect==3.6.20'
  # plus any extras you need, e.g., 'prefect[kubernetes]'

3) Verify versions at runtime - Worker pod:

kubectl -n <namespace> exec deploy/<worker-deployment-name> -- prefect version

- Flow-run pod (after triggering a run), check logs for Prefect version or run a quick diagnostic:

from prefect import flow, task, get_run_logger, __version__ as PREFECT_VERSION

@task
def show():
    get_run_logger().info(f"Prefect version in task: {PREFECT_VERSION}")

@flow
def f():
    show()

if __name__ == "__main__":
    f()

4) After aligning, if task logs/graph are still missing, double-check: - Server setting:

PREFECT_SERVER_SERVICES_TASK_RUN_RECORDER_ENABLED=true

- Worker logs show no WebSocket/connectivity errors to the API - Logging to API is on in flow pods:

PREFECT_LOGGING_TO_API_ENABLED=true

- If you rely on prints, enable

log_prints=True

on flows/tasks or set

PREFECT_LOGGING_LOG_PRINTS=true

If you want, share: - The worker Deployment name/namespace and whether you use a custom image - Whether your flow-run image is set via the work pool base job template or per-deployment I can give you an exact patch command or template diff.