August 5, 2010 at 3:48 PM

To begin, very early this morning at 3:04AM CDT (GMT-5) we received an alert that Node2 had gone offline. This was a hard crash, unlike most that are a matter of a simple reboot, after this crash, the server did not come back up without reverting to a previous kernel due to issues with the newer kernels loading. At the time, we just loaded up the old kernel and let it be to keep downtime to a minimum. This initial outage lasted about 40 minutes  (~60 minutes for an individual VPS + quota re-check). After a few more hours, at 7:56AM CDT we received another notification that Node2 had gone offline. We immediately began going over possible problems and solutions to determine what wasn't working and what could help get the server back online and fully functional to avoid further outages or any issues with service speed.

It was quickly determined that the only working kernel was an old kernel that seemed to have trouble allowing the server to utilize all of the installed RAM. For a VPS node, it is imperative that all of the installed RAM is available to the node without issue, otherwise we encounter subsequent crashes due to the RAM + SWAP being completely utilized.

Throughout the morning we have been going over many different possible solutions. Our first plan to resolve the issue came in figuring out what was (seemingly) wrong with the kernels that wouldn't boot. This came down to trying many different kernels, kernel parameters, and even compiling software that seemed to be interfering with the boot process. Despite all of this, nothing at all changed the behavior of the problems we were seeing.

After many hours of trying for a software solution, it was clear something wasn't right with the RAM. We tried changing BIOS options, but still the entirety of the RAM was not being recognized by the server. Ultimately the problem was pinpointed to be a faulty RAM stick. This led to all of the problems we saw throughout the software troubleshooting process. After hours of troubleshooting, replacing the RAM at around 1:50PM CDT finally resolved the problems on Node2.

We have seen very few hardware failures present as software/kernel boot problems in this nature. We admittedly spent far too much time banging our heads against the wall with a number of solutions that were not showing any promise, when we should have been questioning what triggered the initial outage earlier in the night. In the end, Node2 was brought back online at 1:57PM CDT, with VPS services coming back online shortly thereafter.

We would like to thank everyone on Node2 for their patience throughout the problems today. As our way of thanking you for your patience and honoring our SLA, you may request a one month service credit for this outage by e-mailing [email protected].