OnPremise - Incorrect hostname when device behind a wifi network login page.

[farrantch]

We have installed the agent on a device located on a wifi network that redirects to a login page before allowing access to the network.

What I believe is happening - First the agent checks to see if it's an EC2 instance by trying to hit the meta-data service, but the local network is redirecting this request to a login page. The entire contents of this page is then saved as the device hostname.

We are using the hostname within our log-group name. So then an error is thrown when trying to create the log group because the name limit is 512 characters and the content of the web page is much longer than that.

Potential solution - Parse the returned page to make sure it is the ec2 meta-data service that is responding.

[pingleig]

😅 The logic is

• get wifi login https://github.com/aws/amazon-cloudwatch-agent/blob/master/translator/util/ec2util/ec2util.go#L57-L57
• use it as hostname https://github.com/aws/amazon-cloudwatch-agent/blob/master/translator/translate/util/placeholderUtil.go#L52-L52

But I think it won't get triggered if the agent is configured as onPrem using -m onPremise I think you are using default, which is auto detection without validation

func initEC2UtilSingleton() (newInstance *ec2Util) {
    newInstance = &ec2Util{Region: "", PrivateIP: ""}
    if context.CurrentContext().Mode() == config.ModeOnPrem {
        return
    }

    if info, e := md.GetMetadata("hostname"); e == nil {
        newInstance.Hostname = info
    } else {
        log.Println("E! getting hostname from EC2 metadata fail: ", e)
    }
}

https://github.com/aws/amazon-cloudwatch-agent/blob/master/packaging/dependencies/amazon-cloudwatch-agent-ctl#L56-L59

        -m: mode
            ec2:                                    indicate this is on ec2 host.
            onPremise:                              indicate this is on onPremise host.
            auto:                                   use ec2 metadata to determine the environment, may not be accurate if ec2 metadata is not available for some reason on EC2.

There are some inconsistencies in our logic, the credential has fall backed to static creds instead of using ec2 instance profile, while the hostname logic took wifi logic page as valid ec2 metadata api response. I think we can add a logic to avoid case like this, though the best approach for your use case should be specifying onPremise

btw: in aws go sdk, the ec2metadata API just copy the content without decoding

[farrantch]

Ahh, I can definitely move forward with the -m option. Thank you!!

[farrantch]

So turns out I was already using the -m onPremise option :( So I am going to re-open this ticket.

Perhaps potential solution now is to verify -m onPremise uses os.hostname by default?

[jhnlsn]

I'm assuming that the current logic relies on the request to the metadata endpoint to fail as the decision to use alternative means to lookup the hostname. This should be pretty easy to replicate by spoofing the response from the metadata endpoint ip to verify this behavior.

[SaxyPandaBear]

@farrantch Have you tried using the {local_hostname} placeholder instead of {hostname}?

I didn't know about this until just now as I was looking at the code in ./translator/translate/util/placeholderUtil.go

func getHostName() string {
    if hostname, err := os.Hostname(); err == nil {
        return hostname
    } else {
        log.Println("E! getHostName: ", err)
        return unknownHostname
    }
}