The APM test agent is an application which emulates the APM endpoints of the Datadog agent which can be used for testing Datadog APM client libraries.
See the Features section for the complete list of functionalities provided.
See the HTTP API section for the endpoints available.
See the Development section for how to get the test agent running locally to add additional checks or fix bugs.
The test agent can be installed from PyPI:
pip install ddapm-test-agent
# HTTP on port 8126, OTLP HTTP on port 4318, OTLP GRPC on port 4317, with the web-ui enabled
ddapm-test-agent --port=8126 --otlp-http-port=4318 --otlp-grpc-port=4317 --web-ui-port=8080
or from Docker:
# Run the test agent and mount the snapshot directory
docker run --rm\
-p 8126:8126\
-p 4318:4318\
-p 4317:4317\
-e SNAPSHOT_CI=0\
-v $PWD/tests/snapshots:/snapshots\
ghcr.io/datadog/dd-apm-test-agent/ddapm-test-agent:latest
or from source:
pip install git+https://github.com/Datadog/dd-apm-test-agent
or a specific branch:
pip install git+https://github.com/Datadog/dd-apm-test-agent@{branch}
Many checks are provided by the test agent which will verify trace data. All checks are enabled by default and can be manually disabled.
See the configuration section for the options.
| Check description | Check name |
|---|---|
| Trace count header matches number of traces | trace_count_header |
| Client library version header included in request | meta_tracer_version_header |
| Trace content length header matches payload size | trace_content_length |
All data that is submitted to the test agent can be retrieved.
- Traces can be returned via the
/test/tracesendpoint documented below.
The INFO log level of the test agent outputs useful information about the requests the test agent receives. For traces this includes a visual representation of the traces.
INFO:ddapm_test_agent.agent:received trace payload with 1 trace chunk
INFO:ddapm_test_agent.agent:Chunk 0
[parent]
├─ [child1]
├─ [child2]
└─ [child3]
INFO:ddapm_test_agent.agent:end of payload ----------------------------------------
The test agent provides proxying to the Datadog agent.
This is enabled by passing the agent url to the test agent either via the --agent-url command-line argument or by the DD_TRACE_AGENT_URL or DD_AGENT_URL environment variables.
When proxying is enabled, the response from the Datadog agent will be returned instead of one from the test agent.
At the trace-level, proxying can also be disabled by including the X-Datadog-Agent-Proxy-Disabled header with a value of true. This will disable proxying after a trace
is handled, regardless of whether an agent URL is set.
The test agent provides a mechanism to dual ship LLM Observability events to Datadog, regardless of whether the Datadog agent is running.
If using the Datadog agent, set the DD_AGENT_URL environment variable or --agent-url command-line argument to the URL of the Datadog agent (see Proxy for more details).
If not running a Datadog agent, set the DD_SITE environment variable or --dd-site command-line argument to the site of the Datadog instance to forward events to. Additionally, set the DD_API_KEY environment variable or --dd-api-key command-line argument to the API key to use for the Datadog instance.
To disable LLM Observability event forwarding, set the DISABLE_LLMOBS_DATA_FORWARDING environment variable or --disable-llmobs-data-forwarding command-line argument to true.
The test agent can receive Claude Code hook events and assemble them into LLM Observability traces.
See .claude/setup-hooks.md for full setup instructions (also usable as a Claude Code prompt to automate the setup).
The test agent provides a form of characterization testing which we refer to as snapshotting. This allows library maintainers to ensure that traces don't change unexpectedly when making unrelated changes.
This can be used to write integration tests by having test cases use the tracer to emit traces which are collected by the test agent and compared against reference traces stored previously.
To do snapshot testing with the test agent:
- Ensure traces are associated with a session token (typically the name of the test case) by either:
- Calling the
/test/session/startwith the token endpoint before emitting the traces; or - Attaching an additional query string parameter or header specifying the session token on
/vX.Y/tracerequests (see below for the API specifics). (Required for concurrent test running)
- Calling the
- Emit traces (run the integration test).
- Signal the end of the session and perform the snapshot comparison by calling the
/tests/session/snapshotendpoint with the session token. The endpoint will return a400response code if the snapshot failed along with a plain-text trace of the error which can be forwarded to the test framework to help triage the issue.
The traces are normalized and output in JSON to a file. The following transformations are made to the input:
- Trace ids are overwritten to match the order in which the traces were received.
- Span ids are overwritten to be the DFS order of the spans in the trace tree.
- Parent ids are overwritten using the normalized span ids. However, if the parent is not a span in the trace, the parent id is not overwritten. This is necessary for handling distributed traces where all spans are not sent to the same agent.
- Span attributes are ordered to be more human-readable, with the important attributes being listed first.
- Span attributes are otherwise ordered alphanumerically.
- The span meta and metrics maps if empty are excluded.
The test agent includes an optional and experimental Web UI that provides a dashboard for inspecting agent configuration, viewing received requests, exploring traces and snapshots, and managing tracer-flare and remote configuration.
The UI can be enabled with the --web-ui-port PORT command-line argument or by setting the WEB_UI_PORT environment variable.
Once enabled, the Web UI can be accessed at http://localhost:PORT (default port is 8080).
There is also a maximum number of requests to store in memory to display in the UI, which can be configured with the --max-requests command-line argument or by setting the MAX_REQUESTS environment variable (default is 200 requests).
The test agent can be configured to proxy requests to select provider API endpoints, capturing real requests to the server and recording them to play back for future use. Currently, only OpenAI, Azure OpenAI, and DeepSeek are supported.
These cassettes are recorded by default in the vcr-cassettes directory. However, this can be changed with the --vcr-cassettes-directory command-line option, or VCR_CASSETTES_DIRECTORY environment variable.
The cassettes are matched based on the path, method, and body of the request. To mount a cassette directory when running the test agent in a Docker container, run the container with
docker run --rm\
-p 8126:8126\
-v $PWD/vcr-cassettes:/vcr-cassettes
ghcr.io/datadog/dd-apm-test-agent/ddapm-test-agent:latest
Optionally specifying whatever mounted path is used for the cassettes directory. The test agent comes with a default set of cassettes for OpenAI, Azure OpenAI, DeepSeek, Anthropic, Google GenAI, and AWS Bedrock Runtime.
The test agent can be configured to also register custom 3rd party providers. This is done by setting the VCR_PROVIDER_MAP environment variable or the --vcr-provider-map command-line option to a comma-separated list of provider names and their corresponding base URLs.
VCR_PROVIDER_MAP="provider1=http://provider1.com/,provider2=http://provider2.com/"or
--vcr-provider-map="provider1=http://provider1.com/,provider2=http://provider2.com/"The provider names are used to match the provider name in the request path, and the base URLs are used to proxy the request to the corresponding provider API endpoint.
With this configuration set, you can make the following request to the test agent without error:
curl -X POST 'http://127.0.0.1:9126/vcr/provider1/some/path'To ignore headers in recorded cassettes, you can use the --vcr-ignore-headers flag or VCR_IGNORE_HEADERS environment variable. The list should take the form of header1,header2,header3, and will be omitted from the recorded cassettes.
AWS service proxying, specifically recording cassettes for the first time, requires a AWS_SECRET_ACCESS_KEY environment variable to be set for the container running the test agent. This is used to recalculate the AWS signature for the request, as the one generated client-side likely used {test-agent-host}:{test-agent-port}/vcr/{aws-service} as the host, and the signature will mismatch that on the actual AWS service.
Additionally, the AWS_REGION environment variable can be set, defaulting to us-east-1.
To add a new AWS service to proxy, add an entry in the PROVIDER_BASE_URLS for its provider url, and an entry in the AWS_SERVICES dictionary for the service name, since they are not always a one-to-one mapping with the implied provider url (e.g, https://bedrock-runtime.{AWS_REGION}.amazonaws.com is the provider url, but the service name is bedrock, as bedrock also has multiple sub services, like converse).
To use this feature in your client, you can use the /vcr/{provider} endpoint to proxy requests to the provider API.
from openai import OpenAI
client = OpenAI(base_url="http://127.0.0.1:9126/vcr/openai")The test agent has two endpoints to configure a context around which any VCR cassettes recorded have a suffix of a given test name. To use this, you can hit the
/vcr/test/startwith atest_namebody field to set the test name/vcr/test/stopto clear the test name
This is useful for recording cassettes for a specific test case to easily associate cassettes with that test.
Usage example:
@pytest.mark.fixture
def with_vcr_test_name(request):
with requests.post("http://127.0.0.1:9126/vcr/test/start", json={"test_name": request.node.name}):
yield
requests.post("http://127.0.0.1:9126/vcr/test/stop")
@pytest.mark.fixture
def openai_with_custom_url(with_vcr_test_name):
client = OpenAI(base_url="http://127.0.0.1:9126/vcr/openai")
yield client
def test_openai_with_custom_url(openai_with_custom_url):
"""This test will generate/use a cassette name something similar to `openai_chat_completions_post_abcd1234_test_openai_with_custom_url`"""
...To add a new provider, add a supported provider in the PROVIDER_BASE_URLS dictionary in ddapm_test_agent/vcr_proxy.py, and change your tests or use case to use the new provider in the base url:
base_url = "http://127.0.0.1:9126/vcr/{new_provider}"And pass in a valid API key (if needed) in the way that provider expects.
To redact api keys, modify the filter_headers list in the get_vcr function in ddapm_test_agent/vcr_proxy.py. This can be confirmed by viewing cassettes in the vcr-cassettes directory (or the otherwise specified directory), and verifying that any new cassettes do not contain the api key.
To have the vcr proxy throw a 404 error if a cassette is not found in CI mode to ensure that all cassettes are generated locally and committed, set the VCR_CI_MODE environment variable or the --vcr-ci-mode flag in the cli tool to true (this value defaults to false).
The test agent can be configured via command-line options or via environment variables.
ddapm-test-agent is command used to run a test agent.
Please refer to ddapm-test-agent --help for more information.
ddapm-test-agent-fmt is a command line tool to format or lint snapshot json files.
# Format all snapshot json files
ddapm-test-agent-fmt path/to/snapshots
# Lint snapshot json files
ddapm-test-agent-fmt --check path/to/snapshotsPlease refer to ddapm-test-agent-fmt --help for more information.
-
PORT[8126]: Port to listen on. -
ENABLED_CHECKS[""]: Comma-separated values of checks to enable. Valid values can be found in trace invariant checks -
LOG_LEVEL["INFO"]: Log level to use. DEBUG, INFO, WARNING, ERROR, CRITICAL. -
LOG_SPAN_FMT["[{name}]"]: Format string to use when outputting spans in logs. -
SNAPSHOT_DIR["./snapshots"]: Directory in which snapshots will be stored. Can be overridden by providing thedirquery parameter on/snapshot. -
SNAPSHOT_CI[0]: Toggles CI mode for the snapshot tests. Set to1to enable. CI mode does the following:- When snapshots are unexpectedly generated from a test case a failure will be raised.
-
SNAPSHOT_IGNORED_ATTRS["span_id,trace_id,parent_id,duration,start,metrics.system.pid,metrics.process_id,metrics.system.process_id,meta.runtime-id"]: The attributes to ignore when comparing spans in snapshots. -
DD_AGENT_URL[""]: URL to a Datadog agent. When provided requests will be proxied to the agent. -
DD_APM_RECEIVER_SOCKET[""]: When provided, the test agent will listen for traces on a socket at the path provided (e.g.,/var/run/datadog/apm.socket) -
DD_SUPPRESS_TRACE_PARSE_ERRORS[false]: Set to"true"to disable span parse errors when decoding handled traces. When disabled, errors will not be thrown for metrics incorrectly placed within the meta field, or other type errors related to span tag formatting/types. Can also be set using the--suppress-trace-parse-errors=trueoption. -
SNAPSHOT_REMOVED_ATTRS[""]: The attributes to remove from spans in snapshots. This is useful for removing attributes that are not relevant to the test case. Note that removingspan_idis not permitted to allow span ordering to be maintained. -
SNAPSHOT_REGEX_PLACEHOLDERS[""]: The regex expressions to replace by a placeholder. Expressed as a comma separatedkey:valuelist. Specifyingba[rz]:placeholderwill change any occurrence ofbarorbazto{placeholder}:foobarbazqux->foo{placeholder}{placeholder}qux. This is in particular useful to strip path prefixes or other infrastructure dependent identifiers. -
DD_POOL_TRACE_CHECK_FAILURES[false]: Set to"true"to pool Trace Check failures that occured within Test-Agent memory. These failures can be queried later using the/test/trace_check/failuresendpoint. Can also be set using the--pool-trace-check-failures=trueoption. -
DD_DISABLE_ERROR_RESPONSES[false]: Set to"true"to disable Test-Agent<Response 400>when a Trace Check fails, instead sending a valid<Response 200>. Recommended for use with theDD_POOL_TRACE_CHECK_FAILURESenv variable. Can also be set using the--disable-error-responses=trueoption.
Return traces that have been received by the agent. Traces matching specific trace ids can be requested with the options below.
Specify trace ids as comma separated values (eg. 12345,7890,2468)
Initiate a synchronous session. All subsequent traces received will be associated with the required test token provided.
Sample rates to be returned by the agent in response to trace v0.4 and v0.5 requests.
Example: "{'service:test,env:staging': 0.5, 'service:test2,env:prod': 0.2}" (note the JSON has to be URL-encoded).
Test session token for a test case. Ensure this value is unique to avoid conflicts between sessions.
Perform a snapshot generation or comparison on the data received during the session.
Snapshots are generated when the test agent is not in CI mode and there is no snapshot file present. Otherwise a snapshot comparison will be performed.
To run test cases in parallel this HTTP header must be specified. All test cases sharing a test token will be grouped.
* Required for concurrent tests. Either via query param or HTTP header.
Comma-separated list of keys of which to ignore values for.
The default built-in ignore list is: span_id, trace_id, parent_id,
duration, start, metrics.system.pid, metrics.process_id,
metrics.system.process_id, meta.runtime-id.
default: ./snapshots (relative to where the test agent is run).
Override the directory where the snapshot will be stored and retrieved from. This directory must already exist.
This value will override the environment variable SNAPSHOT_DIR.
Warning: it is an error to specify both dir and file.
An absolute or relative (to the current working directory of the agent) file name where the snap will be stored and retrieved.
Warning: it is an error to specify both file and dir.
Note: the file extension will be appended to the filename.
_tracestats will be appended to the filename for trace stats requests.
Comma-separated list of keys that will be removed from spans in the snapshot.
The default built-in remove list does not remove any keys.
Return all requests that have been received by the agent for the given session token.
Returns the requests in the following json format:
[
{
"headers": {},
"body": "...",
"url": "http...",
"method": "GET"
}
]body is a base64 encoded body of the request.
Return traces that have been received by the agent for the given session token.
Return stats that have been received by the agent for the given session token.
Stats are returned as a JSON list of the stats payloads received.
Return OpenTelemetry logs that have been received by the agent for the given session token.
Logs are returned as a JSON list of the OTLP logs payloads received. The logs are in the standard OpenTelemetry Protocol (OTLP) v1.7.0 format, decoded from protobuf into JSON.
Return OpenTelemetry metrics that have been received by the agent for the given session token.
Metrics are returned as a JSON list of the OTLP metrics payloads received. The metrics are in the standard OpenTelemetry Protocol (OTLP) v1.7.0 format, decoded from protobuf into JSON.
Create a Remote Config payload to retrieve in endpoint /v0.7/config
curl -X POST 'http://0.0.0.0:8126/test/session/responses/config' -d '{"roots": ["eyJ....fX0="], "targets": "ey...19", "target_files": [{"path": "datadog/2/ASM_DATA/blocked_users/config", "raw": "eyJydWxlc19kYXRhIjogW119"}], "client_configs": ["datadog/2/ASM_DATA/blocked_users/config"]}'
Due to Remote Config payload being quite complicated, this endpoint works like /test/session/responses/config (POST)
but you should send a path and a message and this endpoint builds the Remote Config payload.
The keys of the JSON body are path and msg
curl -X POST 'http://0.0.0.0:8126/test/session/responses/config/path' -d '{"path": "datadog/2/ASM_DATA/blocked_users/config", "msg": {"rules_data": []}}'
Get Trace Check failures that occured. If a token is included, trace failures for only that session token are returned unless used in conjuction with return_all, which can be used to return all failures regardless of inputted token. This method returns a <Response 200> if no Trace Check failures are being returned and a <Response 400> if Trace Check failures are being returned. Trace Check failures are returned as a content type of text, with failure messages concatenated in the response body. Optionally, set the use_json query string parameter to true to return Trace Check failures as a JSON response in the following format:
response = {
"<FAILING_CHECK_NAME>" : ["<FAILURE_MESSAGE_1>", "<FAILURE_MESSAGE_2>"]
}
NOTE: To be used in combination with DD_POOL_TRACE_CHECK_FAILURES, or else failures will not be saved within Test-Agent memory and a <Response 200> will always be returned.
curl -X GET 'http://0.0.0.0:8126/test/trace_check/failures'
Clear Trace Check failures that occured. If a token is included, trace failures for only that session token are cleared unless used in conjuction with clear_all. This argument can be used to clear all failures (regardless of inputted session token).
curl -X GET 'http://0.0.0.0:8126/test/trace_check/clear'
Get Trace Check summary results. If a token is included, returns summary results only for Trace Checks run during the session. The return_all optional query string parameter can be used to return all trace check results (regardless of inputted session token). The method returns Trace Check results in the following JSON format:
summary = {
"trace_content_length" : {
"Passed_Checks": 10,
"Failed_Checks": 0,
"Skipped_Checks": 4,
} ...
}
curl -X GET 'http://0.0.0.0:8126/test/trace_check/summary'
Update information about the current tested integration.
curl -X PUT 'http://0.0.0.0:8126/test/session/integrations' -d '{"integration_name": [INTEGRATION_NAME], "integration_version": [INTEGRATION_VERSION],
"dependency_name": [DEPENDENCY_NAME], "tracer_language": [TRACER_LANGUAGE], "tracer_version": [TRACER_VERSION]}'
Return a csv list of all tested integrations received by the agent. The format of returned data will be:
tracer_language,tracer_version,integration_name,integration_version,dependency_name.
curl -X GET 'http://0.0.0.0:8126/test/integrations/tested_versions'
Mimics the pipeline_stats endpoint of the agent, but always returns OK, and logs a line everytime it's called.
Accepts OpenTelemetry Protocol (OTLP) v1.7.0 traces in protobuf format via HTTP. This endpoint validates and decodes OTLP traces payloads for testing OpenTelemetry trace exporters and libraries.
The HTTP endpoint accepts POST requests with Content-Type: application/x-protobuf and Content-Type: application/json and stores the decoded logs for retrieval via the /test/session/traces endpoint.
Accepts OpenTelemetry Protocol (OTLP) v1.7.0 logs in protobuf format via HTTP. This endpoint validates and decodes OTLP logs payloads for testing OpenTelemetry logs exporters and libraries.
The HTTP endpoint accepts POST requests with Content-Type: application/x-protobuf and Content-Type: application/json and stores the decoded logs for retrieval via the /test/session/logs endpoint.
Accepts OpenTelemetry Protocol (OTLP) v1.7.0 metrics in protobuf format via HTTP. This endpoint validates and decodes OTLP metrics payloads for testing OpenTelemetry metrics exporters and libraries.
The HTTP endpoint accepts POST requests with Content-Type: application/x-protobuf and Content-Type: application/json and stores the decoded metrics for retrieval via the /test/session/metrics endpoint.
OTLP logs and metrics can also be sent via GRPC using the OpenTelemetry LogsService.Export, MetricsService.Export, and TraceService.Export methods respectively. The GRPC server implements the standard OTLP service interfaces and forwards all requests to the HTTP server, ensuring consistent processing and session management.
Note: OTLP endpoints are served on separate ports from the main APM endpoints (default: 8126):
- HTTP: Port 4318 (default) - Use
--otlp-http-portto configure - GRPC: Port 4317 (default) - Use
--otlp-grpc-portto configure
Both protocols store decoded data for retrieval via the /test/session/logs, /test/session/metrics, and /test/session/traces HTTP endpoints respectively.
GRPC Client → GRPC Server → HTTP POST → HTTP Server → Agent Storage ↓ ↓ (forwards protobuf) (session management) ↓ ↓ HTTP Retrievable via Response /test/session/{logs,metrics,traces}
Mimics the tracer_flare endpoint of the agent. Returns OK if the flare contains the required form fields, otherwise 400.
Logs a line everytime it's called and stores the tracer flare details in the request under "_tracer_flare".
Return all tracer-flares that have been received by the agent for the given session token.
Returns the tracer-flares in the following json format:
[
{
"source": "...",
"case_id": "...",
"email": "...",
"hostname": "...",
"flare_file": "...",
}
]flare_file is the base64 encoded content of the tracer-flare payload.
If there was an error parsing the tracer-flare form, that will be recorded under error.
Allows to change some settings on the fly. This endpoint takes a POST request with a json content listing the keys and values to apply.
{ 'key': value }Supported keys:
trace_request_delay: sets a delay to apply to trace and telemetry requests
curl -X POST 'http://0.0.0.0:8126/test/settings' -d '{ "trace_request_delay": 5 }'
A Python version of 3.8 or above and riot are required. It is recommended to create
and work out of a virtualenv:
python3.12 -m venv .venv
source .venv/bin/activate
pip install -e '.[testing]'
To run the tests (in Python 3.12):
riot run -p3.12 test
Note: if snapshots need to be (re)generated in the tests set the environment variable GENERATE_SNAPSHOTS=1.
GENERATE_SNAPSHOTS=1 riot run --pass-env -p3.12 test -k test_trace_missing_received
To lint, format and type-check the code:
riot run -s flake8
riot run -s fmt
riot run -s mypy
To build (and tag) the dockerfile:
docker build --tag testagent .Run the tagged image:
docker run --rm -v ${PWD}/snaps:/snapshots --publish 8126:8126 agentThis project follows semver and so bug fixes, breaking
changes, new features, etc must be accompanied by a release note. To generate a
release note:
riot run reno new <short-description-of-change>document the changes in the generated file, remove the irrelevant sections and commit the release note with the change.
- Checkout the
masterbranch and make sure it's up to date.
git checkout master && git pull- Generate the release notes and use
pandocto format them for Github:
riot run -s reno report --no-show-source | pandoc -f rst -t gfm --wrap=noneCopy the output into a new release: https://github.com/DataDog/dd-apm-test-agent/releases/new.
- Enter a tag for the release (following
semver) (eg.v1.1.3,v1.0.3,v1.2.0). - Use the tag without the
vas the title. - Save the release as a draft and pass the link to someone else to give a quick review.
- If all looks good hit publish