Skip to content

Databricks Apps telemetry tables stay empty after following observability docs #171

@uniquejava

Description

@uniquejava

A minimal reproduction repo is available here: https://github.com/uniquejava/mcp-server-hello-world

Problem

Databricks Apps telemetry target tables are created successfully, but they remain empty even after:

  • enabling App telemetry
  • deploying the app successfully
  • configuring the app per the observability docs
  • generating real traffic to the app

I reproduced this both with a custom app and with the official mcp-server-hello-world sample from this repo.

What I changed

For the official sample app, I only made the minimal observability changes from the docs:

  • start command uses opentelemetry-instrument
  • set OTEL_TRACES_SAMPLER=always_on
  • add required OTel dependencies:
  • opentelemetry-distro
  • opentelemetry-exporter-otlp-proto-grpc
  • opentelemetry-instrumentation-fastapi

What works

  • app deploys successfully
  • app is running
  • app is reachable
  • MCP endpoint works
  • requests succeed (for example health)
  • OTel packages are present in the app runtime

What does not work

These tables remain empty:

  • workspace.demo.otel_logs
  • workspace.demo.otel_metrics
  • workspace.demo.otel_spans
  • workspace.demo.otel_annotations
SELECT COUNT(*) FROM workspace.demo.otel_logs;
SELECT COUNT(*) FROM workspace.demo.otel_metrics;
SELECT COUNT(*) FROM workspace.demo.otel_spans;
SELECT COUNT(*) FROM workspace.demo.otel_annotations;

All return 0.

Additional checks

  • target tables were created successfully
  • target tables are Delta tables
  • DESCRIBE DETAIL workspace.demo.otel_logs shows otel.schemaVersion = v2
  • app Service Principal has USE CATALOG, USE SCHEMA, and table-level SELECT / MODIFY
  • still no data appears

Why I think this is not app-specific

I first saw this in a custom MCP server app. To rule out app-specific causes, I repeated the experiment with the official sample app from this repo and
got the same result.

Questions

  • Is this a known issue?
  • Are there additional requirements not covered in the observability docs?
  • Are there workspace / metastore / edition limitations for Databricks Apps telemetry?
  • How can I verify whether the Databricks-managed telemetry ingestion pipeline is actually active for my app?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions