Monitor Metadata agent health via the HTTP layer

[Edin S]

We've found a few cases (e.g. LP#1907686) where monitoring the the wellness of the systemd service "nova-api-metadata" to be inadequate in determining if the metadata service is working; the systemd service might be reported as active/running, but it may be failing to receive and respond to connections/requests.

I'd like to request a check for monitoring a metadata agent's health via the HTTP layer (e.g. make a request to http://169.254.169.254/ and potentially check for good values). I understand that there are complexities that arise from network namespaces, and scaling this solution to hundreds (or thousands) of networks -- but perhaps there's a middle ground that can be found.