My first in-prod corrupted hard drive problem
blog.pavementlink.ch
blog.pavementlink.ch
Funny thing was, I didn't actually observe any data loss. I stressed the drive for several days, no errors. It went back in my daily driver for the next 5 years with no failure. It's been 15 years since that happened and the drive still hasn't failed.
I don't trust SMART anymore.
This doesn't quite seem to follow. As described, neither of the "recovery" methods actually restore lost data. So why weren't any of the SQL pages left in a bad state?
So if you keep rereading that section of the disk you eventually get all the data, save it somewhere, write a bunch of new patterns over it, then write the original data and verify it reads back correctly many times.
I believe the article's analysis about RAID is wrong though; most controllers will start resilvering or just fail a drive once it experiences too many IO errors.
> I did some research, and a RAID wouldn’t have saved it either, RAID protects against drive failure, not against silent page corruption that gets faithfully replicated to every mirror.
That being said, the article has some strong signal of AI writing in it. So it's possible the author isn't really learning well from the experience either. :(
I dispute this was a 'silent' drive error as many systems reported read errors. Silent data corruption on hard drives is extremely rare, due to the tons of checksums used on all data. Maybe I'm wrong but I bet there are read errors on the drive in the appropriate system logs.
I feel that people confuse regular 'bad blocks' with 'silent data corruption' and there is a huge difference[0].
[0]: https://louwrentius.com/what-home-nas-builders-should-unders...
I had another brand-new, identical Samsung SSD, so I hooked both the old and new drive up to a Linux laptop (with USB cases) and tried to dd the old drive to the new drive. That mostly worked, but VERY VERY slowly: it would run fast for 5 seconds and then have no activity for 30 seconds. I had a fan blowing on the old drive to keep it cool because it was running very hot.
The dd copy would eventually fail and then I'd restart it with appropriate iseek and oseek values. I also did a cmp /dev/zero with the new disk to verify that it was all zeroes (it was brand new), and that allowed me to use conv=sparse on the dd. The reason for that was to avoid writing to ever sector of the new disk; I didn't want to copy sectors from the old drive that had never been accessed (she only used about 250GB of the 1TB).
It took a couple of days and about 5 restarts to finish the copy, but it did work, and as a precaution, I made another copy of the drive and ran a cmp of the original drive and the 2nd copy (also having to restart cmp several times). Since that compare worked, I knew that all 3 drives had identical content. The new drive worked fine in her laptop and she was mighty glad to see her Windows login screen.
The thing that made this work, IMO, is that Linux has a longer timeout for errors than Windows apparently does, especially during the boot sequence. Plus Linux allows adjusting the drive timeout, so if the device is doing error recover, which is sometimes slow, it gives it time to finish rather than reporting an error.
One of my theories was that the bad SSD was overheating, but if that was the case, a cold boot should have worked, with the failures only coming later.
The other theory is that one of the chips on the SSD failed, so the drive was having to use the ECC codes to correct for the missing information, and the correction process was taking longer than Windows boot would tolerate.
You can also get 'partclone' to generate a list (in ddrescue format) of sectors containing data, so you don't need to try to read unused areas of the disk. For the partclone trick to work, the FS does need to be at least somewhat readable.
Disk errors logged in the system event log are from the I/O layer, low-level class driver (msahci.sys) / filter drivers. See Windows Storage Driver Architecture : https://learn.microsoft.com/en-us/windows-hardware/drivers/s...
A disk error of this type showing in the event log must immediately be treated as an actual disk issue. This is a low level issue below the actual filesystem and application/services. Seems here the .mdf/ldf of your SQL database used one or more bad sectors on the disk surface.
Your disk seems to be only one on the system, so the first thing to do is check SMART status, for example with Crystaldiskinfo (the most used and user-friendly free portable windows software).
It would very probably have shown a warning state for the internal disk, with probably one or more (judging the quantity of disk error entries in your log) for Attribute C5 "Current Pending sectors" and probably some in Att 05 - "Reallocated sectors count" and/or Att C4 - "Realloc event Count".
Second thing to do is trying to backup your data as fast as possible. In your case related to a Ms SQL database, trying to dump it / backup first was the good move. Sadly (DR pro experience here), weak surface / failing Head Stack Assembly of a traditionnal HDD from most vendors has more difficulties reading correctly a sector than writing it.
If the dump/backup fails, the second choice would have be to try to a sector-to-sector dump approach of the whole disk, with either a online (from OS) software capable of reading sectors from the boot disk (didn't try if HDD Raw Copy Tool 2.6 supports it), or an offline solution like Clonezilla, Acronis True Image, Aomei backupper, etc. But offline solution means offline computer and service...
I didn't exactly understood if you had an actual backup of the data or an image of the whole disk. Considering the critical usage of this station, you should have both running : daily data backup or more + up-to-date disk image ready. whatever the type of disk (HDD/SSD). And a spare, identical computer.
As for repair of HDDs "weak sectors" (meaning Current Pending Sectors), it is indeed possible, often with complete data recovery. If not, the sector will be left as is, or may be remapped if written to 0 (it will then shift from Current pending to realloc sector count).
Hard disk Sentinel Pro as such features (Disk repair, Quick Fix), it works quite well. The result vary greatly from one type of failure to another, as from one disk maker to another.
Note that if the SMART shows more than a little dozen of sectors, the head (amp/preamp) is probably failing, making weaker magnetically-wise sectors too difficult to read and/or write. In this case, the count of current and remapped increases every repair/check pass made by the tools. In this case, the drive is toast and must be replaced ASAP.
SSD are a complete different case for repair.
A older autonomous tool, Spinrite, was specialized for this usage (accurate recovery of data), but veeeeery slow.
RAID pertinence : fortunately, it is an expecteed case as most SATA disks are prone to HSA failure before not initialyzing at all. A RAID 1 mirror would have protected you from a mirrored defect accross the two disks.
The RAID controller (true hardware controller like LSI/Avago or Microsemi) or even fake raid like Intel RST / VROC maintain data integrity accross the array's disks. The defective disk will raise bad blocks (that will get marked in metadata of the Raid Volume), but the others disks are fine and the data can be read safely. If too many errors are reported on a disk (very few in fact on most controllers), it will be labelled as failed and taken down from the array.
What filters in the event log would you apply to find such errors?
Keep using the system from a disk showing read issues could trigger loss of more data, and one could always back up the SQL from the backup image later.
Over the last decade, I've ran hundreds of servers if not thousands, and I entirely stopped using hard drives, now it's solely SSD/NVMe where the failure rate in practice is incredibly lower, I've had my fair share of middle-night runs because websites are offline or whatever to end-up in a hard drive diagnosis circus.
Imo, the peace of mind you get worth the cost, it also allows you to rethink development entirely, typical example would be that suddenly, copying all node_modules or rust deps is a great idea with 10Gbit/s bandwidth and fast drives (yes, I expect people to shit on me for saying this, please give me the counterarguments if you downvote me), many things change if you have a higher base performance assumption, storage is relatively cheap as well. I would never advise anyone that wants to run continuously in prod with low friction to get servers with HDD.
I get that for some use cases it's not possible, but for large majority of use cases, it's clearly not HDD that is the cost burden. $50 servers gets you TBs of SSD, of course don't go with VPS or "Cloud" if you intend to change your development based on new performance assumptions, it blows my mind the numbers of people paying thousand of dollars just to handle what, 100K visitors a day? That fits on a $100 server and a bunch of Kimsufi hosted across the world as a CDN.
People are overcomplicating infrastructure, big time (which leads to more problems, higher maintenance, security issues and so on).
My experience is that (most) spinners give off reliable pre-failure indicators (if you take the time to look/script looking), but SSDs fail by disappearing from the bus. The SSDs do fail much less often, but they still fail from time to time and recovery is harder.
Either way, if your data is important to you/your customers, you really need a backup/recovery plan.
I dunno about recent pricing, but not so long ago, it felt like spinners had a pretty high price floor and SSDs didn't... If you don't need a lot of space, you could find a small SSD that was still around the same $/GB as a medium sized SSD, but for spinners, there's a floor in dollars and space. So if you don't need a lot of space, you save money with an SSD and get better perf for free... If you need a lot of space and not a lot of perf, big spinners are more attainable than big SSDs.
I'm not a pro, just a smalltime dork with a homelab. I use cheap WD HDDs on my NAS system connected to an LSI hardware RAID controller. I'll boast that I have a 100% record so far of preventing downtime and data loss by simply listening for the controller's audible alarm and swapping drives right away (I keep brand new spares). I also have offline backups, but have so far never needed them. Not sure how this would change if I moved to SSDs.
Coordinated failure is a thing, unfortunately. Drives bought together tend to fail together.
SATA disk are indeed generally more predictable failure-wise. Most issues are related to a failing head stack assembly. Rarely platter demagnetization for some disks (Toshiba laptop).
Other failure issues are usually related to a friggin' manufactured firmware issue from Dell, HP or Lenovo corp.
> Either way, if your data is important to you/your customers, you really need a backup/recovery plan.
You'd be surprised at how many devs/companies walk on eggshells all the time (praying that the fatal moment never arrive) because they aren't "brave" enough to do a proper backup system, which is often few minutes/hours of setup only.