So I had an odd issue with a dell poweredge 2950 server with 5 hard drives in it. They were Hitachi drives 2TB each. I would format the server and use software raid to create a raid 5. Upon turning off the server and booting back up one of the hard drives would be “removed” from the array.
I later realized a complete off shutdown -h now and a cold start were how to reproduce the issue.
I realized that hitachi had a firmware update for my hard drives since they had an issue in some 3ware controllers. My controller a perc 5i is a dell re-branded LSI controller – so I gave it a shot anyway. After flashing the drives I had the same result. I think you should flash the drives to bring them up to date if you are reading this.
What the actual cause of the issue was; I only plugged in one of the two redundant power supplies for the server. In the past, we ran this server with 2 hard drives and had no issue with it being fully functional running on one of the two PSU’s. So when we un-racked the server and brought it to the office for this rebuild we only plugged in one PSU. After all outlets are valuable! Well I plugged in the second PSU into the other wall jack/strip (a different one – I didn’t even bother trying the same jack/strip) and I have not seen the drive go into a “removed” state yet.
So far I have concluded this was the issue. My conclusion is the fact the issue was consistent before and no longer occurs. The reason behind this I think is that these 2GB drives draw alot of power from the server – the initial boot from a cold state requires that the drives spin up. All 5 drives, all fans, and all cpu’s are drawing much more power than regular at the same time during the boot from a cold start. This must have been draining the 750w power supply with my server and causing the PERC 5i to not see one of the drives or to mark it as removed since it did not gain a response in adequate time.
Well that was a waste of about 16-20 hours of work – no I can begin installing the actual software on this machine knowing my raid will be reliable.