Errors when you create or update
Creating or updating a function
When you save a function, Elemental Inference checks the syntax of both expressions and runs the request expression against sample media. For a passthrough function, it also runs the response expression on the result. For a model target, the response expression is checked for syntax only, so an error in it appears only when the function runs on a feed.
| Cause | Error | What to do |
|---|---|---|
| An expression isn't valid JSONata. | ValidationException |
The message names the expression that failed, followed by the parser's error. Unbalanced braces and a missing comma between object fields are the common causes. |
| An expression is nested too deeply. | ValidationException |
Flatten it. Bind intermediate values with := instead of
nesting them. |
| The request expression, or a passthrough function's response expression, is valid JSONata but fails when it runs. | ValidationException |
The message gives the reason and, for a guardrail, its code. Common causes are a built-in function that isn't on the allowlist, a guardrail limit, and a media value that is altered or placed more than once. See Allowed built-in functions and Expression limits and guardrails. |
| For a model target, the request expression doesn't produce a JSON object. | ValidationException |
The request body sent to the model must be an object. The message reports what the expression produced instead. A path that doesn't exist produces no value rather than an error, so check the paths the expression reads. |
| A passthrough function produces no value: the request expression yields nothing, or the response expression yields nothing when applied to the request expression's result. | ValidationException |
Such a function would never write metadata. Make the response expression
read the fields that the request expression builds. A model-style path such
as $.output.text finds nothing in a passthrough result. |
The request expression references two feature contexts – two of
media.transcript, media.contextual, and
media.clip. |
ValidationException |
Reading data from more than one existing feature is not supported. Split
the work into two functions. You can combine any one of them with
media.frames or media.video. |
A passthrough function references media.frames or
media.video. |
ValidationException |
A passthrough function invokes no model, so it can't send media anywhere. Remove the reference, or change the target to a model. |
| The name, model, or description doesn't match its required pattern or length. | ValidationException |
See the property table in Creating a function. |
| You changed the name or the target of a function that a feed is using. | ValidationException |
Remove the function output from every feed that uses the function, or create a new function. |
You changed which media context the request expression reads
while a feed is using the function. |
ValidationException |
Each feed's trigger and source outputs are set up for the function's current context. Remove the function output from the feeds, update the function, then add the output again. |
| You deleted a function that a feed is using. | ConflictException |
The message reports how many feeds hold it, and the function's
references field lists them. Remove the function output from each
of those feeds. A disabled function output still holds the
reference. |
Creating or updating a feed with a function output
Unless the table says otherwise, the error is ValidationException.
| Message reports | What to do |
|---|---|
| The function ARN isn't a valid function ARN. | Check the ARN. It must have the form
arn:aws:elemental-inference:. |
| The function can't be found. | Confirm that the function exists in the same Region as the feed. |
| The function belongs to a different AWS account. | A feed can reference only functions in its own account. Create the function in the feed's account. |
The function output invokes a model, so the feed must specify an
accessRoleArn. |
Add accessRoleArn to the request. UpdateFeed
clears the role if you omit it. See Setting up the access role. |
| A feed with an enabled function output must also have an enabled cropping, clipping, subtitling, or contextual metadata output. | Enable a smart crop, event clipping, smart subtitles, or contextual metadata output on the feed, or disable the function output. |
| The function uses transcript context but names no subtitling output, or names one that isn't an enabled subtitling output on this feed. | Add subtitlingSettings.outputName, and confirm that the output
it names is a smart subtitles output on this feed with
"status": "ENABLED". |
| Transcript context isn't supported with this trigger. | A transcript function must use fixedInterval. |
Contextual metadata context is only available on a perScene
or perShot trigger. |
Use perScene, or perShot with
contextualMetadataSettings.outputName set. |
| The scene or contextual metadata source isn't set, or isn't an enabled contextual metadata output. | Add contextualMetadataSettings.outputName, naming an enabled
contextual metadata output on this feed. |
Clip context is only available on a perClip or
fixedInterval trigger, or the clipping source isn't set. |
Use perClip or fixedInterval, and set
clippingSettings.outputName to an enabled event clipping output
on this feed. |
| The output requests more frame and video media than fits a single inference request. | The frames and the clip together are too large for one model request.
Reduce the frameSettings count, the videoSettings
count, or both, or set a smaller frameSettings.frameSize. |
ServiceQuotaExceededException for enabled function
outputs. |
Disable a function output, or request a quota increase. See Quotas and limits for functions. |
ServiceQuotaExceededException for enabled outputs, or for
outputs enabled and disabled combined. |
Disable or remove another output. These quotas can't be increased. See Quotas and limits for functions. |
