Troubleshooting Serverless Monitoring for AWS Step Functions
このページは日本語には対応しておりません。随時翻訳に取り組んでいます。
翻訳に関してご質問やご意見ございましたら、
お気軽にご連絡ください。
I cannot see any traces
- Ensure that the
DD_TRACE_ENABLED
tag is set to true
on the Step Function in your AWS console. - In your AWS console, open your Step Function’s logging tab. Ensure that Log level is set to
ALL
, and that Include execution data is selected. - Ensure that the CloudWatch log group (also found on the logging tab) has a subscription filter to the Datadog Lambda Forwarder in the same region.
Verify that logs are forwarded successfully to Datadog
- Check the Datadog Lambda Forwarder for error messages. Ensure that you have correctly set your API key and Datadog site.
- Enable
DEBUG
logs on the Datadog Lambda Forwarder by setting the environment variable DD_LOG_LEVEL
to debug
.
Verify that logs are searchable on Live Search and have DD_TRACE_ENABLED tag
In Datadog, go to Logs > Log Stream. Search for source:stepfunction
. You may need to trigger the state machine a few times. If you need to upgrade Datadog Lambda Forwarder from an older version, check that after the upgrade, the Forwarder has the DD_FETCH_STEP_FUNCTIONS_TAGS
tag set to true
. If the upgraded Forwarder does not have the DD_FETCH_STEP_FUNCTIONS_TAGS
tag, your Forwarder may not be upgraded correctly.
If the Forwarder and state machine tags are set up correctly with the previous steps, the logs are tagged with DD_TRACE_ENABLED:true
.
Search historic logs
To enable searching historic logs, add a temporary index to the forwarded logs. In Datadog, open the Logs Indexes tab. Click the New Index button in the upper right.
Choose a name, set the index filter to Source:stepfunction
, leave everything else with default values, and save.
If your organization has an existing all-encompassing index with a low limit, place your new index at the top.
Note: Indexing logs is not a requirement for getting traces and may incur additional cost. If you are troubleshooting a specific issue, you may wish to temporarily send logs to an index, debug, and delete the index afterwards. See Indexes for more information.
Verify that your Step Function is using the latest version
- AWS may release updates to the Step Function API or introduce newer versions of the Step Function definitions. Older versions may result in unexpected log formatting or behavior.
- It’s also recommended that you are using the latest version of the Datadog Lambda Forwarder to avoid discrepancies in how logs are forwarded.
Caution when using custom log pipelines
- Custom log pipelines can offer flexibility in processing logs, but altering the log format too much can lead to issues downstream, such as logs not being parsed or recognized.
- Avoid making significant changes to the Step Function log structure that change the JSON format.
Lambda traces are not merging with Step Function traces
- Verify that you can see both Lambda traces and Step Function traces in Datadog.
- Verify that you are using Python layer v95+ or Node.js layer v112+.
- In your AWS console, open your Step Function and ensure that your state machine has
"Payload.$": "States.JsonMerge($$, $, false)"
on the Lambda steps. - Execute your Step Function once and verify that the
TaskScheduled
event log of the Lambda step has the payload containing data from the Step Function context object. If you do not have a TaskScheduled
event and only have a LambdaFunctionScheduled
event, update the task in Step Functions definition to use the recommended Lambda integration. See the AWS documentation for instructions on how to do this.
I can see the aws.stepfunctions
root span but I cannot see any step spans
Please enable the Include execution data
option on the state machine’s logging. After enabling this option, log execution input, data passed between states, and execution output is logged. The Datadog backend uses the logs to construct these step spans for you.
Traces are missing intermittently
When searching traces, select the Live Search option in the upper right corner. If Live Search shows your trace, add “@trace_type:stepfunctions” to the retention filter and set the desired retention rate. For debugging, Datadog recommends setting the retention rate to 100%. The filter can be disabled after debugging is done.
Some step spans are missing in the traces
- Actions from Lambda, DynamoDB, StepFunction, and most of the other AWS services are supported.
Wait
, Choice
, Success
, Fail
, Pass
, Inline MapState
, and Parallel
are supported, while Distributed MapState
is not supported.
Customized way to deploy Datadog Lambda Forwarder
If you are using your customized way to deploy Datadog Lambda Forwarder, here are some tips that can help you debug enabling Step Functions tracing:
- On the forwarder, set the environment variable
DD_FETCH_STEP_FUNCTIONS_TAGS
to true
. - To enable Step Functions trace generation on the Datadog backend, the Datadog-Forwarder layer version must be greater than 31. This version is able to fetch state machine tags, including the required
DD_TRACE_ENABLED
tag. - You can also set the
DD_STEP_FUNCTIONS_TRACE_ENABLED
tag at the Forwarder-level to enable tracing for all Step Functions using that Forwarder on v3.121.0+. - The IAM role for the forwarder should have
tags:getResources
permission. - Set up a subscription filter on your state machine CloudWatch log group to the Datadog forwarder.
- To verify if logs are reaching the Datadog backend, open the Log Explorer page and search
source:stepfunction
with the Live
search timeframe (which shows all logs going into Datadog’s logs intake). If you cannot see any logs, check if there are any error logs on the Datadog Forwarder such as wrong/invalid API key. Adding the environment variable DD_LOG_LEVEL
of DEBUG
helps you debug the Forwarder issue. If you see Step Functions logs, verify that the logs have the dd_trace_enable:true
tag (all tags are normalized) and you should see Step Function traces associated with the log in a few minutes.
Notes
If your Lambda has the DD_TRACE_EXTRACTOR
environment variable set, its traces cannot be merged.