# Inconsistent/Elevated SDK Metrics (Python SDK)

**URL:** <https://community.temporal.io/t/inconsistent-elevated-sdk-metrics-python-sdk/18654>\
**Category:** Community Support\
**Tags:** python-sdk\
**Created:** [November 12, 2025, 6:47pm UTC](https://community.temporal.io/t/inconsistent-elevated-sdk-metrics-python-sdk/18654 "2025-11-12T18:47:32Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![emcc](https://avatars.discourse-cdn.com/v4/letter/e/9fc348/32.png) [@emcc](https://community.temporal.io/u/emcc)\
**Post date:** [November 12, 2025, 6:47pm UTC](https://community.temporal.io/t/inconsistent-elevated-sdk-metrics-python-sdk/18654/1 "2025-11-12T18:47:32Z")

</div>

Hi there,

My team is working with Temporal via the Python SDK and is piping SDK metrics into Datadog. We’re trying to set up monitoring for workflow failures via the `temporal_workflow_failed` metric, but are seeing strange/unexpected behavior. For ~70 failed workflows in the past 12 hours, we are seeing the metric register a count in the neighborhood of 7000.

The workflow itself is fairly simple - 1 activity that retries once, no retries of the workflow itself. The metric is being filtered to this exact workflow\_type and operating environment. The [SDK documentation](https://docs.temporal.io/references/sdk-metrics#workflow_failed)reads to me like it should only be registering 1 increment for a workflow reaching failed state. Any ideas that might help us chase down the source of this issue or similar experiences are appreciated!

---

<div class="post-metadata">

**Author:** ![antonio.perez](https://sea2.discourse-cdn.com/flex016/user_avatar/community.temporal.io/antonio.perez/32/2417_2.png) [@antonio.perez](https://community.temporal.io/u/antonio.perez)\
**Post date:** [November 13, 2025, 12:54pm UTC](https://community.temporal.io/t/inconsistent-elevated-sdk-metrics-python-sdk/18654/2 "2025-11-13T12:54:58Z")

</div>

Hi,

not sure if this is the issue, but with Datadog you need to configure the OpenTelemetry to use **DELTA temporality** instead of CUMULATIVE

```auto
metrics=OpenTelemetryConfig(....  
     metric_temporality=OpenTelemetryMetricTemporality.DELTA
),

```

> **[Producing Delta Temporality Metrics with OpenTelemetry](https://docs.datadoghq.com/opentelemetry/guide/otlp_delta_temporality/?tab=python)**
>
> Datadog, the leading service for cloud-scale monitoring.

---

<div class="post-metadata">

**Author:** ![emcc](https://avatars.discourse-cdn.com/v4/letter/e/9fc348/32.png) [@emcc](https://community.temporal.io/u/emcc)\
**Post date:** [November 13, 2025, 4:10pm UTC](https://community.temporal.io/t/inconsistent-elevated-sdk-metrics-python-sdk/18654/3 "2025-11-13T16:10:35Z")

</div>

Thanks for this! I’ll try this out and see if it helps!

---

<div class="post-metadata">

**Author:** ![emcc](https://avatars.discourse-cdn.com/v4/letter/e/9fc348/32.png) [@emcc](https://community.temporal.io/u/emcc)\
**Post date:** [December 2, 2025, 7:42pm UTC](https://community.temporal.io/t/inconsistent-elevated-sdk-metrics-python-sdk/18654/4 "2025-12-02T19:42:48Z")

</div>

Just following up to confirm that this shift to DELTA temporality cleared up our metrics issues here! Thanks so much for the assistance!