CloudWatch Log Group not automagically created

[claranet]

CloudWatch Log Group not automagically created

[raymondbutcher]

Hi.

You can create the log group yourself with https://www.terraform.io/docs/providers/aws/r/cloudwatch_log_group.html and Lambda will just log to that if the name matches.

Does aws_cloudwatch_log_subscription_filter actually fail if the function hasn't run yet and the log group doesn't exist? I'm just wondering if outputting the log group name would be enough, or if you actually need the resource.

I'm in 2 minds about adding the resource to this module. Terraform can get pretty slow when there are a lot of resources in a project. On the other hand, having Lambda create log groups means there are things not managed by Terraform.

But ultimately it would be nice to have retention_in_days easily set by this module, so it's probably worth adding the resource.

[jeohist]

It does fail unfortunately:

• aws_cloudwatch_log_subscription_filter.foo_to_bar: ResourceNotFoundException: The specified log group does not exist.

Creating the CloudWatch Log Group yourself with the same name that would be generated by the module works just fine. It doesn't actually take up a lot of time (2s creation time), though obviously it'd be another resource to create. I agree that having retention_in_days would be a nice-to-have though :-)

[raymondbutcher]

I'm leaning towards yes for creating the log group resource in this module.

The only problem I can think of is that enabling it might break existing deployments, because Terraform would want to create a log group that the Lambda function has already created. I suspect that it would break, but haven't tested it. Maybe it needs a var.create_cloudwatch_log_group to make it optional.

[lorengordon]

We can always import the resource when we update to the new version. Folks ought to know to pin module versions.

[jeohist]

I'm leaning towards yes for creating the log group resource in this module.

The only problem I can think of is that enabling it might break existing deployments, because Terraform would want to create a log group that the Lambda function has already created. I suspect that it would break, but haven't tested it. Maybe it needs a var.create_cloudwatch_log_group to make it optional.

Maybe something like var.use_existing_cloudwatch_log_group? This would allow users who've already used the module to simply specify the existing CloudWatch Log Group, would allow new users to specify their own CloudWatch Log Group (for whatever reason), and if not defined would explicitly create a CloudWatch Log Group.

It's a shame there's no plural data source for aws_cloudwatch_log_group allowing you to check whether the CloudWatch Log Group already exists, and make the decision based on that...

edit: Speak of the devil - hashicorp/terraform-provider-aws#7928

[raymondbutcher]

I'm thinking about whether to add this as part of the next release that adds Terraform 0.12.x support. Since it's a big breaking change, we can worry less about this breaking things.

But there's one thing I hadn't considered about this before: I don't necessarily want to delete the log group when deleting the Lambda function. There may be compliance issues, e.g. PCI DSS has log retention rules.

Maybe something like var.use_existing_cloudwatch_log_group?

This would get around that problem, but I'm still not sure about it. So I need to think about this some more!

[lorengordon]

But there's one thing I hadn't considered about this before: I don't necessarily want to delete the log group when deleting the Lambda function. There may be compliance issues, e.g. PCI DSS has log retention rules.

Such requirements could use kinesis streaming to forward logs for retention compliance to wherever they need to store them.

[raymondbutcher]

That's assuming that CloudWatch Logs is not the desired place to store them.

There may be some wisdom behind the way that Lambda functions create their own log groups that never get deleted. The default behaviour is very safe.

I think I would rather it just output the log group name, and then creating the log group is still pretty easy and most importantly optional:

module "lambda" { ... }

resource "aws_cloudwatch_log_group" "lambda" {
  name = module.lambda.log_group_name
  retention_in_days = 7
}

[jeohist]

Wouldn't that fail if the CloudWatch Log Group already exists, because the Lambda function has somehow already been triggered?

[raymondbutcher]

It would, but if you’re creating the function right next to it then it’s unlikely to have been triggered.

If it were created as part of the module then I suppose we could add the log group to depends_on for all out the outputs, which would make it even less likely for something to trigger it.

[zkghost]

Hey all, just wanted to add I think I have a use case which encounters this issue in a 😢 way:

So we have some .tf that creates a template lambda function, as part of that we create some metric alarms in Cloudwatch.

I'm trying to create a log_metric_filter that tracks the log output from the lambda and counts occurrences of a few terms. The issue I'm hitting is that I can't seem to get my lambda function to write to my log_stream/log_group. It only writes to the default stream that it is created with. The end result is that my metric filter is created and watches a log group, but the log group gets no output and the metric never sees a value / the alarm never fires.

I wouldn't mind just putting my log_metric_filter on the default log_group that gets created when the lambda does, but I have no idea how to reference it. I've tried using the default name /aws/lambda/<function-name> but that returns:

 The specified log group does not exist.

Is this a symptom of the same issue as is discussed in this thread?