Hey All,
I hope everyone is doing well. 
I have a PMM2 server running on an Azure VM and now I’m trying to monitor a database instance that is running on a GCP VM.
I installed the pmm-client and ran the pmm-admin config command. At first it looked like it worked just fine, but I can’t see anything on my Grafana dashboards.
Here is the output of my “pmm-admin status”:
root@my_vm01 server]# pmm-admin status
Agent ID: /agent_id/4c691837-e7a0-4e2d-be82-a37facdf2b10
Node ID : /node_id/aa7777a0-28d6-4d66-873b-a07ca6d333d8
PMM Server:
 URL :
 Version: 2.9.1
PMM Client:
 Connected : true
 Time drift : -74.059µs
 Latency : 8.460801ms
 pmm-admin version: 2.9.1
 pmm-agent version: 2.9.1
 /agent_id/62c9d70a-a7bd-4396-a3d2-1309f086050a node_exporter Running
Below is what I see on Grafana:

Check https://<pmmserver>/prometheus/targets im going to guess it shows a bunch of connection timeout issues. Pmm client talks to the server on port 443 but pmm server needs to talk back to the exporters on ports 42001,2,3 etc so you will need to open that channel up in reverse. We are working to change this in a future release. 

I’m seeing tons of “context deadline exceeded” messages. Thanks for pointing me in the right direction, Steve. 
– L

No worries, been my biggest frustration but relief is on the way.  For those following along, we’ll still allow direct access to the exporters in the traditional “pull” model but we’ll also have a simpler “push” so we can re-use the path you’ve already confirmed to work (client --> server on TCP 443) with the pmm-client registration step!

Now I have a different issue to solve…
Long story short, my PMM-Server is on Azure and it has a public IP address. I just learned the system I’m trying to monitor on GCP doesn’t have a public IP address.
The endpoint is registered in the server using a GCP private IP address, so my understanding is that when the PMM-Server tries to reach out to the endpoint on port 42000, it just can’t reach it. :frowning:
I’m starting to think about creating an instance on GCP, but I wouldn’t like to go on this route, though. 
Have anybody here seen a use case like this? Any guesses on how I could solve this?
– L