Skip to content

Workshop Resources#

Quick reference links and additional learning materials for Dynatrace AI Observability.

Official Documentation#

Dynatrace#

Resource Link
Dynatrace Documentation docs.dynatrace.com
OpenTelemetry Integration OTLP Ingest
Grail & DQL DQL Documentation
Dynatrace MCP MCP Server Documentation

Dynatrace Intelligence#

Resource Link
Blog post Announcement
Learn more Documentation

OpenLLMetry / Traceloop#

Resource Link
OpenLLMetry GitHub traceloop/openllmetry
Traceloop SDK Docs docs.traceloop.com
Supported Libraries Integrations List

OpenTelemetry#

Resource Link
OpenTelemetry Python opentelemetry.io/docs/instrumentation/python
OTLP Specification OTLP Spec
Semantic Conventions GenAI Conventions

Workshop Code Reference#

Dynatrace Instrumentation Template#

import os
from traceloop.sdk import Traceloop

# REQUIRED: Set Delta temporality for Dynatrace
os.environ["OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE"] = "delta"

# Initialize Traceloop for Dynatrace
headers = {"Authorization": f"Api-Token {os.getenv('DT_API_TOKEN')}"}
Traceloop.init(
    app_name="your-service-name",
    api_endpoint=os.getenv("DT_ENDPOINT"),  # https://xxx.live.dynatrace.com/api/v2/otlp
    headers=headers
)

Required Environment Variables#

# Dynatrace Configuration
DT_ENDPOINT=https://YOUR_ENV.live.dynatrace.com/api/v2/otlp
DT_API_TOKEN=dt0c01.XXXXXXXXXX.YYYYYYYYYYYY

# Application Configuration
ATTENDEE_ID=your-unique-id

Python Dependencies#

traceloop-sdk>=0.15.0
opentelemetry-exporter-otlp>=1.22.0

Useful DQL Queries#

LLM Token Usage#

fetch spans
| filter service.name == "ai-chat-service-*"
| filter isNotNull(llm.usage.total_tokens)
| summarize 
    total_tokens = sum(llm.usage.total_tokens),
    prompt_tokens = sum(llm.usage.prompt_tokens),
    completion_tokens = sum(llm.usage.completion_tokens)
  by bin(timestamp, 1h)

LLM Latency Analysis#

fetch spans
| filter span.name matches "azure_openai.*"
| summarize 
    avg_latency = avg(duration),
    p50_latency = percentile(duration, 50),
    p95_latency = percentile(duration, 95),
    p99_latency = percentile(duration, 99)
  by span.name

Request Distribution by Model#

fetch spans
| filter isNotNull(llm.model)
| summarize count = count() by llm.model
| sort count desc

Error Rate for AI Requests#

fetch spans
| filter service.name == "ai-chat-service-*"
| summarize 
    total = count(),
    errors = countIf(otel.status_code == "ERROR")
| fieldsAdd error_rate = errors * 100.0 / total

API Token Permissions#

For this workshop, your Dynatrace API token needs:

Permission Purpose
openTelemetryTrace.ingest Ingest trace data via OTLP
metrics.ingest Ingest metrics via OTLP
entities.read Read entity information (for MCP)
problems.read Read problem data (for MCP)
logs.read Read log data (for MCP)
DataExport Export data via DQL (for MCP)

Further Learning#

Dynatrace University#

AI/ML Observability#

GitHub Resources#

Getting Help#

During the Workshop: Raise your hand for instructor assistance.

After the Workshop:

← Lab 4: Automation | Back to Home →