4

I have created a mdadm RAID 10 with 4 SSDs. 2 of those SSDs are consumer models and does not have the power loss data protection feature.

At the RAID creation i have put 1 SSD with power protection in every RAID 1 pair. So every RAID 1 pair has one consumer SSD and one enterprise SSD.

If the server has a power fault, how would the power loss protection affect the filesystem or the RAID consistency? I can imagine, that the RAID is not in sync anymore because SSD A - which is with power protection - has an other (newer) content than SSD B.

Could the power loss protection cause more damage or problems in this case?

1 Answer 1

2

Yes, you may experience data loss. A real power loss protection would imply capacitors on the SSD. That's what enterprise-class ones should have, or an alternate mechanism just as viable.

Any SSD should to have some kind of protection against FTL (flash translation layer) corruption, or the entire SSD will be damaged resulting in total data loss.

Standard SSDs use a technique known as 'journaling' to prevent their FTL from becoming corrupted. This means that upon sudden power-loss, the SSD practically reverts to an earlier state, which comes in conflict with the flush cache command. In client computing (desktop, notebooks), there are 2 types of data defined: “data at rest” (data that has been physically saved to the storage media) and “data in flight” (pending writes - writes sent to the drive from the host computer but not yet committed to the storage media, or any write that is in progress, but is not yet complete, or in cache). In the case of power failure, the “data in flight” is gone.

In the enterprise-class SSDs, the capacitor circuits enable the SSDs to protect itself, to complete the pending / in-progress writes. The capacity is also enough to ensure that the FTL addressing table is properly saved to the NAND.

So, getting back to your case, in an unexpected power loss situation, the enterprise-class SSD will manage to write everything and have no losses while the other will likely have data losses. This will lead to RAID inconsistency.

What you should do is pair up the standard ones as an array and the enterprise-class ones as another. Keep the critical data on the protected ones and not that needed data (and pagefiles) on the standard ones.

2
  • On research i did not found a clear information how mdadm would handle a inconsistent RAID. Would the filesystem / the RAID have huge problems and it will not be possible to boot anymore without manual repair or will mdma handle these situations automatically and guess which SSD has the current files? Dec 15, 2017 at 15:06
  • There is no clear answer because the result of using a mdadm RAID check utility may differ - it may find the problem (it will try to determine which disk is the damaged one which could be successful in your case) or not (if there are corruptions of area of unpredictable size and location in the array). Regarding booting issues, it should not be a problem since there is no reason for those sectors to be re-written, but the data resync may be bad.
    – Overmind
    Dec 18, 2017 at 6:46

You must log in to answer this question.

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