DCP2000 repair disk procedure question. (urgent)

[James Gardiner]

a DCP2000 has a bad disk. It was replaced.
a Repair was started in the WEB interface..
/data is now healthy. But /opt is still degraded. If I select /opt in the touch screen interface, and repair, it only gives me the option of reinitialising, and deleting all data.

image.png

The interesting issue is. If I reboot the DCP2000, when it comes up, it reports its rebuilding the /opt array. then when it gets to 100% it then goes into the error state seen above. it has [U__] indicating 2 disks are uninitialized/missing.

I have not done this on a DCP2000 before. I have used a lot of IMB2000 and they restore both partitions and it all just works.


What is the correct course of action? (Urgent of course)

Can I just initialize the /opt array? What will be the effect of doing this? Will the player operate under these condition?

[Leo Enticknap]

Suggest emailing cinemasupport [at] dolby.com. They may have a more elegant solution than nuking the entire RAID volume and starting over (which I'm guessing you don't want to do because you can't reingest all the content that you'd lose).

[Marcel Birgelen]

You're right regarding the two members missing: [3/1] [U__] => two of the three members are missing. Since this is a RAID 5, with two missing members, you can't rebuild the RAID. I can't imagine the /opt partition to be even accessible in this case. I guess another disk went bad, if this wasn't signaled before, chances this happened during the increased stress of the rebuild. So, the only valid option I do see is to re-initialize the /opt RAID5 array, I doubt Dolby can really help otherwise in this case to be honest. Still, it can't hurt to ask them.

[James Gardiner]

We just went with replacing all disks.

The Problem with the error given is md0 and md1 are on the same disk as show by the devices list. And it's working for md0, but not md1. VERY STRANGE. looks like an OS bug.
A rebuild with full set of new disks was done and all good.

[Marcel Birgelen]

The Doremi machines use a rather strange software RAID configuration. They create two partitions on al your disks and create two separate RAID5 arrays on top of those partitions. That way, both /opt and /data are two separate RAID5 arrays on top of the same physical disks. A bunch of bad sectors clustered together may affect one partition, but not the other and in such affect one of the two RAID5 arrays. In practice, it will most likely also hurt the performance of the other array though.

I don't really like this design, as it can create this strange inconsistencies where one RAID is functioning and the other is not, like you just experienced. Also, vanilla Linux software RAID is really bad at detecting and handling I/O issues that lead to unacceptable read/write delays. It was engineered for consistency in "office" applications, not for performance-critical tasks. Disks that are near-death often aren't ejected from the RAID, leading to performance degradation to a point that fluent playback can't be guaranteed.