appsync: Log group for API is not created at deployment leading to "The specified log group does not exist" error when applying metric filter

[jzybert]

Describe the bug

I am deploying a GraphqlApi construct with a logConfig prop set with retention days as well. I am also calling api.logGroup.addMetricFilter in the same stack. However, I get this error Resource handler returned message: "The specified log group does not exist. It appears that the log group is not created even when the logConfig prop is provided upon deployment. This is causing newly deployed stacks to fail.

Expected Behavior

I would have expected the log group to be created when the API is created so that metric filters could be applied.

Current Behavior

I get a CloudFormation error of Resource handler returned message: "The specified log group does not exist and my stack fails to depoy.

Reproduction Steps

const api = new GraphqlApi(this, 'API', {
  name: 'api',
  schema: SchemaFile.fromAsset('schema.graphql'),
  logConfig: {
    fieldLogLevel: FieldLogLevel.ALL,
    retention: RetentionDays.ONE_WEEK,
    excludeVerboseContent: true
  },
});

api.logGroup.addMetricFilter('MetricFilter', {
  filterPattern: {
    logPatternString: '{ $.fieldName = "myQuery" && $.fieldInError IS TRUE }'
  },
  metricName: 'ErrorCount',
  metricNamespace: 'MyNamespace,
  metricValue: '1',
});

Possible Solution

It appears that the code is simply looking for a log group name but not actually generating its own log group when the API is created. See here. A fix for this would be to ensure the log group is actually created rather than trying to load it by name.

Additional Information/Context

Some additional context is that I'm using VTL templates for some of my API resolvers. Because they don't natively have their own log groups or metrics, I'm leveraging adding a metric filter to the main API log group which is where the template resolvers log their logs/errors to.

CDK CLI Version

2.88.0

Framework Version

No response

Node.js Version

18

OS

macOS Ventura 13.4

Language

Typescript

Language Version

5.0.4

Other information

No response

[pahud]

Looks like it assums the log group already exists.

aws-cdk/packages/aws-cdk-lib/aws-appsync/lib/graphqlapi.ts

Lines 566 to 568 in 960d14a

	const logGroupName = `/aws/appsync/apis/${this.apiId}`;
	this.logGroup = LogGroup.fromLogGroupName(this, 'LogGroup', logGroupName);

Reading the doc I guess you will need to enable the logging first?
https://docs.aws.amazon.com/appsync/latest/devguide/monitoring.html

If this works for you, we probably should add this notice in the CDK doc.

[jzybert]

@pahud The logConfig is added if that's what you mean by enable logging first

[pahud]

@jzybert

No from what I see in the source code, adding the logConfig will not create log group for you from CDK. Instead, that log group seems to be managed by the AppSync service and will be created only when you turn on Enable Logs from the console as described in the doc. I am not sure what is the best way to deal with it but CDK appsync construct library at this moment seems to expect the log group exists so it can import that with fromLogGroupName() method.

[jzybert]

@pahud ah, I see. Unfortunately that doesn't entirely work for us, because our CloudFormation stack deployment would fail after deploying the API but before enabling the logs then. Hoping for a CDK update here to support turning it on immediately after the API is created rather than needing to be a manual process

[idclipr]

We are hitting this bug too. Can GraphqlApi construct be updated to actually create the log group instead of assuming it exists? i.e. use new LogGroup() instead of fromLogGroupName()?

[yashmardikar]

For me even after logging was turned on, the log group was not getting created. I ran some queries and then log group was created. Is log group created lazily? This is not ideal.

For first time deployment in new account/region, deployment fails and then rollback takes at least 20 mins before i can try another deployment. Please Fix this.