Skip to content

Latest commit

 

History

History
101 lines (81 loc) · 5.38 KB

README.md

File metadata and controls

101 lines (81 loc) · 5.38 KB

sfncli

Utility to create AWS Step Function (SFN) activities out of command line programs.

Usage

$ sfncli -h
Usage of sfncli:
  -activityname string
    	The activity name to register with AWS Step Functions. $VAR and ${VAR} env variables are expanded.
  -cmd string
    	The command to run to process activity tasks.
  -region string
    	The AWS region to send Step Function API calls. Defaults to AWS_REGION.
  -cloudwatchregion string
      The AWS region to send metric data. Defaults to the value of region.
  -version
    	Print the version and exit.
  -workername string
    	The worker name to send to AWS Step Functions when processing a task. Environment variables are expanded. The magic string MAGIC_ECS_TASK_ARN will be expanded to the ECS task ARN via the metadata service.
  -workdirectory string
    	A directory path that is passed to the `cmd` using an env var `WORK_DIR`. For each activity task a new directory is created in `workdirectory` and it is cleaned up after the activity task exits. Defaults to "", does not create directory or set `WORK_DIR`

Example:

sfncli -activityname sleep-100 -region us-west-2 --cloudwatchregion us-west-1 -workername sleep-worker -cmd sleep 100

High-level logic

  • On startup, call CreateActivity to register an Activity with Step Functions.
  • Begin polling GetActivityTask for tasks.
  • Get a task. Take the JSON input for the task and
    • if it's a JSON object, use this as the last arg to the cmd passed to sfncli.
    • if it's anything else (e.g. JSON array), an error is thrown.
    • if _EXECUTION_NAME is missing from the payload, an error is thrown
    • the _EXECUTION_NAME payload attribute value is added to the environment of the cmd as _EXECUTION_NAME.
    • if workdirectory is set, create a sub-directory and add it to the environment of the cmd as WORK_DIR.
  • Start SendTaskHeartbeat loop.
  • When the command exits:
    • Call SendTaskFailure if it exited nonzero, was killed, or sfncli received SIGTERM.
    • Call SendTaskSuccess otherwise. Parse the last line of the stdout of the command as the output for the task (it must be JSON).
    • If workdirectory was set then cleanup WORK_DIR/sub-directory-for-task

Errors

Error names in SFN state machines are useful for debugging and setting up branching/retry logic in state machine definitions. sfncli will report the following error names if it encounters errors it can identify:

  • sfncli.TaskInputNotJSON: input to the task was not JSON
  • sfncli.TaskFailureTaskInputMissingExecutionName: input is missing _EXECUTION_NAME attribute
  • sfncli.CommandNotFound: the command passed to sfncli was not found
  • sfncli.CommandKilled: the command process received SIGKILL
  • sfncli.CommandExitedNonzero: the command process exited with a nonzero exit code
  • sfncli.TaskOutputNotJSON: the task output (last line of command's stdout) was not JSON
  • sfncli.CommandTerminated: sfncli or the command received SIGTERM
  • sfncli.Unknown: unexpected / unclassified errors

The command should signal an error by exiting with a nonzero status code. In this case, the behavior is:

  1. If the last line of stdout was a JSON-formatted string with an error field, report an error to Step Functions with that field as the name and the value of the cause field in the output line as the cause.
  2. Otherwise, report an error with name sfncli.CommandExitedNonzero with the last line of stderr as the cause.

Local testing

Start up a test activity that runs echo on the work it receives.

go run ./cmd/sfncli -region us-west-2 -activityname test-activity -workername sfncli-test -cmd echo

Create a new state machine that uses this activity for one of its states (this requires you to create a role for use with Step Functions):

aws --region us-west-2 stepfunctions create-state-machine --name test-state-machine --role-arn arn:aws:iam::589690932525:role/raf-test-step-functions --definition '{
    "Comment": "Testing out step functions",
    "StartAt": "foo",
    "Version": "1.0",
    "TimeoutSeconds": 60,
    "States": {
        "foo": {
            "Resource": "arn:aws:states:us-west-2:589690932525:activity:test-activity",
            "Type": "Task",
            "End": true
        }
    }
}'

Note that you will need to replace the Resource above to reflect the correct ARN with your AWS account ID.

Start an execution of the state machine (again replacing the ARN below with the correct account ID):

aws --region us-west-2 stepfunctions start-execution --state-machine-arn arn:aws:states:us-west-2:589690932525:stateMachine:test-state-machine  --input '{"_EXECUTION_NAME":"en", "hello": "world"}'

You should see echo run with the argument {"hello": "world"}.