I'm attempting to create an AWS step function with a Map state whose input (that is, the array to iterate over) comes from the Execution Input. A reduce example JSON step function looks like:
{
"StartAt": "pass",
"States": {
"pass": {
"Type": "Pass",
"Next": "map-sleep"
},
"map-sleep": {
"MaxConcurrency": 5,
"InputPath": "$$.Execution.Input['data']",
"Iterator": {
"StartAt": "wait",
"States": {
"wait": {
"SecondsPath": "$['length']",
"Type": "Wait",
"End": true
}
}
},
"Type": "Map",
"Next": "final-wait"
},
"final-wait": {
"Seconds": 10,
"Type": "Wait",
"End": true
}
}
}
However, when I attempt to create this, I'm greeted by the error:
An error occurred (InvalidDefinition) when calling the CreateStateMachine operation: Invalid State Machine Definition: 'SCHEMA_VALIDATION_FAILED: Value must be a valid JSONPath. at /States/map-sleep/InputPath'
I infer from this that the InputPath is wrong, but I don't quite understand why, or what the correct way to express what I'm trying to do is. (This code was generated using the Python Step Functions SDK, and if it's helpful I can share this code, but I figured reducing it to the JSON would make it easier to consider).
Well, I strongly suspect this is not the optimal answer, but it looks like by combining a Pass node using Parameters with a Map node, you can get the desired outcome:
"map-sleep-pass": {
"Parameters": {
"items.$": "$$.Execution.Input['data']"
},
"Type": "Pass",
"Next": "map-sleep"
},
"map-sleep": {
"MaxConcurrency": 5,
"InputPath": "$.items",
"Iterator": {
"StartAt": "wait",
"States": {
"wait": {
"SecondsPath": "$['length']",
"Type": "Wait",
"End": true
}
}
},
"Type": "Map",
"Next": "final-wait"
},
Related
I am using AWS step function to iterate over a list in an input document where for each iteration, I need to invoke an external service. So I want to iterate over each item and run a step using lambda:invoke.waitForTaskToken and pass the TaskToken into the execution of each iteration.
The problem I'm running into is how to use both an ItemSelector at the Map state level but also inject the TaskToken during the internal step. I need to use an ItemSelector because I want each item to also contain information from the input to Map state. The AWS Docs state:
The ItemSelector field replaces the Parameters field within the Map state. If you use the Parameters field in your Map state definitions to create custom input, we highly recommend that you replace them with ItemSelector.
But they also say:
During an execution, the context object is populated with relevant data for the Parameters field from where it is accessed. The value for a Task field is null if the Parameters field is outside of a task state.
These two statements seem to imply that what I'm trying to do is impossible.
So, what I want is something like:
{
"StartAt": "ExampleMapState",
"States": {
"ExampleMapState": {
"Type": "Map",
"ItemsPath": "$.items",
"ItemSelector": {
"dynamic.$": "$.dynamic",
"ContextIndex.$": "$$.Map.Item.Index",
"ContextValue.$": "$$.Map.Item.Value"
},
"ItemProcessor": {
"ProcessorConfig": {
"Mode": "INLINE"
},
"StartAt": "TestPass",
"States": {
"TestPass": {
"Type": "Task",
"Parameters": {
"FunctionName": "arn:aws:lambda:us-west-2:123456789012:function:echo-lambda",
"Payload": {
"item.$": "$",
"token.$": "$$.Task.Token"
}
},
"Resource": "arn:aws:states:::lambda:invoke.waitForTaskToken",
"End": true
}
}
},
"End": true
}
}
}
But this doesn't work because the ItemSelector overrides the Payload of the internal TestPass state. Is there a way to get this to work?
ETA: I figured I would try putting $$.Task.Token in ItemSelector just in case it would magically work but it ended up throwing an error because $$.Task does not exist in the context object at that level.
Example with this (invalid) configuration:
{
"StartAt": "ExampleMapState",
"States": {
"ExampleMapState": {
"Type": "Map",
"ItemsPath": "$.items",
"ItemSelector": {
"dynamic.$": "$.dynamic",
"ContextIndex.$": "$$.Map.Item.Index",
"ContextValue.$": "$$.Map.Item.Value",
"token.$": "$$.Task.Token"
},
"ItemProcessor": {
"ProcessorConfig": {
"Mode": "INLINE"
},
"StartAt": "TestPass",
"States": {
"TestPass": {
"Type": "Task",
"Parameters": {
"FunctionName": "arn:aws:lambda:us-west-2:123456789012:function:echo-lambda"
},
"Resource": "arn:aws:states:::lambda:invoke.waitForTaskToken",
"End": true
}
}
},
"End": true
}
}
}
Based on my research I don't think what I'm trying to do is possible. What I ended up implementing is a workaround. I modified the function providing input to this step function to put the dynamic info that I needed into every item in the list I am iterating over. So my step function definition now looks something like this
{
"StartAt": "ExampleMapState",
"States": {
"ExampleMapState": {
"Type": "Map",
"ItemsPath": "$.items",
"ItemProcessor": {
"ProcessorConfig": {
"Mode": "INLINE"
},
"StartAt": "TestPass",
"States": {
"TestPass": {
"Type": "Task",
"Parameters": {
"FunctionName": "arn:aws:lambda:us-west-2:123456789012:function:echo-lambda",
"Payload": {
"item.$": "$",
"token.$": "$$.Task.Token"
}
},
"Resource": "arn:aws:states:::lambda:invoke.waitForTaskToken",
"End": true
}
}
},
"End": true
}
}
}
And an example input to this step function looks like:
{
"dynamic": "info",
"items": [
{
"dynamic": "info",
"resize": "true",
"format": "jpg"
},
{
"dynamic": "info",
"resize": "false",
"format": "png"
},
{
"dynamic": "info",
"resize": "true",
"format": "jpg"
}
]
}
It's not great because I have to repeat info into every item ahead of time but it works.
I have been working on step functions for a couple of weeks now. I am using map state in my step functions to iterate over an array. The array has an additional inner array as well, thus I would like to employ an addition map step in the "outer" map state. AWS documentation does not go into this level of details (as of now), therefore, I wanted to share that I managed to make it work.
This is how I managed to nest map steps:
"OuterMapState": {
"Type": "Map",
"ItemsPath": "$.shipped",
"MaxConcurrency": 0,
"Iterator": {
"StartAt": "Validate",
"States": {
"Validate": {
"Type": "Task",
"Resource": "arn:aws:lambda:us-east-1:123456789012:function:ship-val",
"Next": "InnerMapState"
},
"InnerMapState": {
"Type": "Map",
"ItemsPath": "$.shipped.innerArray",
"MaxConcurrency": 0,
"Iterator": {
"StartAt": "doSomething",
"States": {
"doSomething": {
"Type": "Pass",
"End": true
}
}
}
},
"End": true
}
},
"End": true
}
Good luck on using AWS Step functions.
I'm sure someone will point me to an immediate solution, but I've been at this for hours, so I'm just going to ask.
I cannot get a State Machine to accept an initial input. The intent is to set up an EventBridge trigger pointed at the State Machine with a static JSON passed to the SM to initiate with the proper parameters. In development, I'm just using Step Functions option to pass a JSON as the initial input when you select "New Execution".
This is the input:
{"event":{
"country": "countryA",
"landing_bucket": "aws-glue-countryA-inputs",
"landing_key": "countryA-Bucket/prefix/filename.csv",
"forecast_bucket": "aws-forecast-countryA",
"forecast_key": "inputs/",
"date_start": "2018-01-01",
"validation": "False",
"validation_size": 90
}
}
When looking at what is passed at the ExecutionStarted log entry:
{
"input": {
"country": "countryA",
"landing_bucket": "aws-glue-countryA-inputs",
"landing_key": "countryA-Bucket/prefix/filename.csv",
"forecast_bucket": "aws-forecast-countryA",
"forecast_key": "inputs/",
"date_start": "2018-01-01",
"validation": "False",
"validation_size": 90
}
,
"inputDetails": {
"truncated": false
},
"roleArn": "arn:aws:iam::a-valid-service-role"
}
This is the State Machine:
"Comment": "A pipeline!",
"StartAt": "Invoke Preprocessor",
"States": {
"Invoke Preprocessor": {
"Type": "Task",
"Resource": "arn:aws:states:::lambda:invoke",
"InputPath": "$.input",
"Parameters": {
"FunctionName": "arn:aws:lambda:my-lambda-arn:$LATEST"
},
"Next": "EndSM"
},
"EndSM": {
"Type": "Pass",
"Result": "Ended",
"End": true
}
}
}
I've tried nearly anything I can think of from changing the InputPath to assigning the "input" dictionary directly to a variable:
"$.event":"$.input"
To drilling down to the individual variables and assigning those directly like:
"$.country:"$.country". I've also used the new Step Functions Data Flow Simulator and can't get anywhere. If anyone has any thoughts, I'd really appreciate it.
Thanks!
Edited for correct solution:
You need to set the Payload.$ parameter to $. That will pass in the entire input object to the lambda.
{
"Comment": "A pipeline!",
"StartAt": "Invoke Preprocessor",
"States": {
"Invoke Preprocessor": {
"Type": "Task",
"Resource": "arn:aws:states:::lambda:invoke",
"Parameters": {
"FunctionName": "arn:aws:lambda:my-lambda-arn:$LATEST",
"Payload.$": "$"
},
"Next": "EndSM"
},
"EndSM": {
"Type": "Pass",
"Result": "Ended",
"End": true
}
}
}
Another thing you could do is specify the input in the parameters, this will allow you to specify only all/certain parts of the json to pass in.
{
"Comment": "A pipeline!",
"StartAt": "Invoke Preprocessor",
"States": {
"Invoke Preprocessor": {
"Type": "Task",
"Resource": "arn:aws:states:::lambda:invoke",
"InputPath": "$",
"Parameters": {
"FunctionName": "arn:aws:lambda:my-lambda-arn:$LATEST",
"input_event.$": "$.event"
},
"Next": "EndSM"
},
"EndSM": {
"Type": "Pass",
"Result": "Ended",
"End": true
}
}
}
From the code perspective you could just reference it like so (python):
input = event['input_event']
I am writing an AWS step function, and for one of the steps, I wish to call a lambda that accepts an array as one of the inputs. However, if I try to pass in a JsonPath into the array, I get
The value for the field 'arrayField.$' must be a STRING that contains a JSONPath but was an ARRAY
My step function definition:
{
"StartAt": "First",
"States": {
"First": {
"Type": "Pass",
"Parameters": {
"type": "person"
},
"ResultPath": "$.output",
"Next": "Second"
},
"Second": {
"Type": "Task",
"Resource": "arn:aws:lambda:us-east-1:<aws_id>:function:MyFunction",
"Parameters": {
"regularParameter": "some string",
"arrayParameter.$": ["$.output.type"]
},
"Next": "Succeed"
},
"Succeed": {
"Type": "Succeed"
}
}
}
How can I use jsonPath inside the array?
Since a new release you could use the intrinsic function States.Array:
"arrayParameter.$": "States.Array($.output.type)"
https://docs.aws.amazon.com/step-functions/latest/dg/amazon-states-language-intrinsic-functions.html
As #Seth Miller mentioned JsonPath resolution within arrays doesn't work unfortunately. If the amount of values to replace in the array is small and known there's a simple workaround (in my case I needed an array of size 1).
The steps are:
Initialise the array with the number of values you need;
Replace each value using "ResultPath": "$.path.to.array[n]";
Use "$.path.to.array" in your task.
Simple, working example:
{
"StartAt": "First",
"States": {
"First": {
"Type": "Pass",
"Parameters": {
"type": "person"
},
"ResultPath": "$.output",
"Next": "Initialise Array"
},
"Initialise Array": {
"Comment": "Add an entry for each value you intend to have in the final array, the values here don't matter.",
"Type": "Pass",
"Parameters": [
0
],
"ResultPath": "$.arrayParameter",
"Next": "Fill Array"
},
"Fill Array": {
"Comment": "Replace the first entry of array with parameter",
"Type": "Pass",
"InputPath": "$.output.type",
"ResultPath": "$.arrayParameter[0]",
"End": true
}
}
}
And to use the resulting array in your task example:
"Second": {
"Type": "Task",
"Resource": "arn:aws:lambda:us-east-1:<aws_id>:function:MyFunction",
"Parameters": {
"regularParameter": "some string",
"arrayParameter.$": "$.arrayParameter"
},
"Next": "Succeed"
},
Another way to approach this is by using Parallel state that outputs an array of objects and then use jsonPath to convert it to a simple array:
{
"StartAt": "Parallel",
"States": {
"Parallel": {
"Type": "Parallel",
"Next": "Use Array",
"ResultPath": "$.items",
"Branches": [
{
"StartAt": "CreateArray",
"States": {
"CreateArray": {
"Type": "Pass",
"Parameters": {
"value": "your value"
},
"End": true
}
}
}
]
},
"Use Array": {
"Type": "Pass",
"Parameters": {
"items.$": "$.items[*].value"
},
"End": true
}
}
}
In this example, Parallel state outputs the following json:
{
"items": [
{
"value": "your value"
}
]
}
And "Use Array" state produces:
{
"items": [
"your value"
]
}
JSONPath inside parameters field need to be a string. So if you want to pass to lambda function a parameter called arrayParameter, you´ll need to make a jsonPath query that extract that array.
For example, if inside the key output is a key called outputArray with the array as its value.
Input JSON:
{
"pre": "sdigf",
"output": {
"result": 1,
"outputArray": ["test1","test2","test.."]
}
}
The parameter sintax:
"arrayParameter.$": "$.output.outputArray"
Reasonable advice
I ran into a use case for JsonPath resolution within arrays today and found (like you have) that the functionality does not exist today. I ended up deciding that doing the data massaging in code was simpler and cleaner. For example, you could create a small Lambda that takes in the object emitted by First and massages it to a format acceptable to Second and adds it to the output (WaterKnight mentions this solution in a comment to another question).
This assumes that you are, for some reason, unable to change the format of the input to that Lambda in Second (which would be the absolute shortest path here).
Unreasonable advice
That said, if you want a way to do this completely within Step Functions that is fairly gross, you can use the result of a Map state that executes Pass states. The output of the Map state is an array that aggregate the output of each constituent Pass state. These Pass states simply emit the value(s) you want in the final array using the Parameters attribute. An example Step Function definition follows. I did warn that it is gross and that I went a different way to solve the problem.
{
"StartAt": "First",
"Comment": "Please don't actually do this",
"States": {
"First": {
"Type": "Pass",
"Parameters": {
"type": "person"
},
"ResultPath": "$.output",
"Next": "Add Array"
},
"Add Array": {
"Comment": "A Map state needs some array to loop over in order to work. We will give it a dummy array. Add an entry for each value you intend to have in the final array. The values here don't matter.",
"Type": "Pass",
"Result": [
0
],
"ResultPath": "$.dummy",
"Next": "Mapper"
},
"Mapper": {
"Comment": "Add a Pass state with the appropriate Parameters for each field you want to map into the output array",
"Type": "Map",
"InputPath": "$",
"ItemsPath": "$.dummy",
"Parameters": {
"output.$": "$.output"
},
"Iterator": {
"StartAt": "Massage",
"States": {
"Massage": {
"Type": "Pass",
"Parameters": {
"type.$": "$.output.type"
},
"OutputPath": "$.type",
"End": true
}
}
},
"ResultPath": "$.output.typeArray",
"Next": "Second"
},
"Second": {
"Comment": "The Lambda in your example is replaced with Pass so that I could test this",
"Type": "Pass",
"Parameters": {
"regularParameter": "some string",
"arrayParameter.$": "$.output.typeArray"
},
"Next": "Succeed"
},
"Succeed": {
"Type": "Succeed"
}
}
}
As many answers correctly pointed out, it's not possible to do it exactly the way you need. But I would suggest another solution: an array of dictionaries. It's not exactly what you need, but is native and not hacky.
"Second": {
"Type": "Task",
"Resource": "arn:aws:lambda:us-east-1:<aws_id>:function:MyFunction",
"Parameters": {
"regularParameter": "some string",
"arrayParameter": [{"type.$": "$.output.type"}]
},
"Next": "Succeed"
},
The result would be
{
"regularParameter": "some string",
"arrayParameter": [{"type": "SingleItemWrappedToAnArray"}]
}
It can not be described with Parallel State in AWS Step Function.
B and C should be in parallel.
C sends messages to both D and E.
D and E should be in parallel.
{
"StartAt": "A",
"States": {
"A": {
"Type": "Pass",
"Next": "Parallel State 1"
},
"Parallel State 1": {
"Type": "Parallel",
"Branches": [{
"StartAt": "B",
"States": {
"B": {
"Type": "Pass",
"End": true
}
}
},
{
"StartAt": "C",
"States": {
"C": {
"Type": "Pass",
"End": true
}
}
}
],
"Next": "Parallel State 2"
},
"Parallel State 2": {
"Type": "Parallel",
"Branches": [{
"StartAt": "D",
"States": {
"D": {
"Type": "Pass",
"End": true
}
}
},
{
"StartAt": "E",
"States": {
"E": {
"Type": "Pass",
"End": true
}
}
}
],
"Next": "F"
},
"F": {
"Type": "Pass",
"End": true
}
}
}
Answer is No , inside step function no state can set multiple states (invokes both successors)to its Next task. As per AWS step function cannot start State Machine as StartAt by providing multiple State names.
You can tweak your logic and use The Parallel state and achive same ,If you share your usecase may be help to solve problems.
How to specify multiple result path values in AWS Step Functions
A Parallel state provides each branch with a copy of its own input
data (subject to modification by the InputPath field). It generates
output that is an array with one element for each branch, containing
the output from that branch.
https://aws.amazon.com/blogs/aws/new-step-functions-support-for-dynamic-parallelism/
Example of state function
{
"Comment": "An example of the Amazon States Language using a choice state.",
"StartAt": "FirstState",
"States": {
"FirstState": {
"Type": "Task",
"Resource": "arn:aws:lambda:REGION:ACCOUNT_ID:function:FUNCTION_NAME",
"Next": "ChoiceState"
},
"ChoiceState": {
"Type" : "Choice",
"Choices": [
{
"Variable": "$.foo",
"NumericEquals": 1,
"Next": "FirstMatchState"
},
{
"Variable": "$.foo",
"NumericEquals": 2,
"Next": "SecondMatchState"
}
],
"Default": "DefaultState"
},
"FirstMatchState": {
"Type" : "Task",
"Resource": "arn:aws:lambda:REGION:ACCOUNT_ID:function:OnFirstMatch",
"Next": "NextState"
},
"SecondMatchState": {
"Type" : "Task",
"Resource": "arn:aws:lambda:REGION:ACCOUNT_ID:function:OnSecondMatch",
"Next": "NextState"
},
"DefaultState": {
"Type": "Fail",
"Error": "DefaultStateError",
"Cause": "No Matches!"
},
"NextState": {
"Type": "Task",
"Resource": "arn:aws:lambda:REGION:ACCOUNT_ID:function:FUNCTION_NAME",
"End": true
}
}
}
https://docs.aws.amazon.com/step-functions/latest/dg/connect-to-resource.html#connect-wait-example
https://sachabarbs.wordpress.com/2018/10/30/aws-step-functions/
As I answered in How to simplify complex parallel branch interdependencies for Step Functions, what you asked is better to be modeled as DAG but not state machine.
Depends on your use case, you might be able to workaround it (just as #horatiu-jeflea 's answer), but it's a workaround (not the straightforward way) anyway.