0

Just today I set up a Google Cloud Compute instance to run my web server, which I was moving from another VPS vendor I'd had a pretty bad experience with.

Every once in a while (perhaps 3 or 4 times since this morning), the Google server is going unreachable from the outside. Normally I can ping it, ssh to it from a Cox Cable address, and get web pages from it on http. When it's having the problem, I can't do any of those three: ping gets 100% packet loss, ssh hangs (and existing ssh connections drop), and http in a web browser also hangs.

To make the problem stranger, during the problem if I ssh to the VPS from google's web interface, I can log in and ping google's nameserver at 8.8.8.8, and I can also ping out to www.yahoo.com. So it's only incoming traffic that's having trouble, not outgoing traffic.

It's acting a little like there's an IP address conflict, but only for the external IP, not for the internal IP. But I'm having a hard time convincing myself that could be the problem.

I've only changed the firewall in one way: I added one firewall rule to allow in http.

The external IP is 34.94.223.166

Can someone please let me know how I can make things more reliable?


During the last event, the system was down for about an hour.

Looking at /var/log/syslog* from during the last event, there is nothing interesting - though the messages don't stop ether.

1 Answer 1

1

Could be any number of performance or availability problems. Determine when this happens, if services are down for everyone, or only for certain routes across the Internet.

Monitoring of your instance is required to see problems. Read your log files.

As you are on GCP, consider using Stackdriver. Stackdriver agent captures instance metrics, can monitor Apache httpd or nginx, and can ingest syslog.

Down detection, try accessing it from as many ISPs as possible, such as from instances in other clouds. Manually or use a monitoring service. Just one does not distinguish between one provider having a problem or your services being down.

High availability beyond one instance is achieved with load balancers. However you may want to refine your monitoring processes with one instance first.

Not the answer you're looking for? Browse other questions tagged .