Last night was the first time I got all build servers plugged in and going at once. A picture of the cluster as it looks now in its nice, new, handcrafted rack can be found here:
OK, the rack looks a little hokey, but $25 for wood and screws, it beats what a real rack would cost.
If you examine the photo you'll notice that there are three empty slots. Hint, hint, hint… I would gladly accept donations of matching iMX53 QSB and SATA drives to complete the cluster of eight build servers. Once the major build push is done in a few months I'll send you the iMX53 QSB and hard drive back to you if you wanted it for your own use. My cost for a iMX53 QSB and hard drive is about $230 after taxes and shipping.
Having them all working I couldn't resist running a test overnight. After nine hours, the cluster built 285 source packages yielding 360 binary packages to be added to the repository.
On a sad note, Mr. Murphy payed me a visit this morning. I needed to adjust something with the the Build00 server and with power going to the cluster I unplugged it's molex connector and when plugging it back in I accidentally had the plug reversed. The molex connectors won't plug in that way, but the pins stick out just far enough that 12V was momentarily connected the Build00 unit on it's 5v input. Instantly the power supply made a popping noise and all the build servers instantly switched off. Seeing what I had done I let out a big "OH! SH**!!!" Dumb, dumb, dumb…
Fortunately, the 12v surge was isolated to just Build00 instantly frying the hard drive, but somehow the iMX53 QSB survived. Being a development board it must have some type of protection against a klutz like me attaching it to 12v. All the systems going down and the popping noise was just the PC power supply circuit protection kicking in. A new Build00 hard drive is already on order with Amazon.
Unfortunately, Build00 was my prototype system and where I kept all my Linux kernel builds of which, of course, I hadn't yet made backups of. I have notes of that work, but those may be incomplete and I'll probably be in for some heartache trying to replicate it as I sort out the kernel issues I came across late last week. Losing my kernel work is not good, but I'm very happy I didn''t accidentally fry $1000 worth of ARM development boards and hard drives.
Needless to say, I clipped the 12v line going from the power supply to the molex connectors and I'll be much more careful about doing anything while power is applied to the cluster. I'll also be spending some time putting a backup regime in place.
Oh well, it's been a two steps forward, one step backward morning for me.