zlacker

[parent] [thread] 64 comments
1. davedu+(OP)[view] [source] 2022-07-08 20:41:05
> Double disk failure is improbable but not impossible.

It's not even improbable if the disks are the same kind purchased at the same time.

replies(6): >>spiffy+15 >>kabdib+7t >>0xbadc+jA >>adrian+nB >>bink+qK >>perilu+pD1
2. spiffy+15[view] [source] 2022-07-08 21:00:26
>>davedu+(OP)
Yep: if you buy a pair disks together, there's a fair chance they'll both be from the same manufacturing batch, which correlates with disk defects.
replies(5): >>clinto+T7 >>bragr+N8 >>GekkeP+gr >>sofixa+tH1 >>dspill+lu3
◧◩
3. clinto+T7[view] [source] [discussion] 2022-07-08 21:14:24
>>spiffy+15
This makes total sense but I've never heard of it. Is there any literature or writing about this phenomenon?

I guess proper redundancy is having different brands of equipment also in some cases.

replies(6): >>AceJoh+A8 >>eganis+L8 >>davedu+aa >>toast0+Xh >>atheno+vj >>mceach+Lm
◧◩◪
4. AceJoh+A8[view] [source] [discussion] 2022-07-08 21:16:54
>>clinto+T7
I don't know about literature, but in the world of RAID this is a common warning.

Having a RAID5 crash and burn because the backup disk failed during the reconstruction phase after a primary disk failed is a common story.

◧◩◪
5. eganis+L8[view] [source] [discussion] 2022-07-08 21:17:50
>>clinto+T7
Not sure about literature, but past anecdotes and HN threads yes.

https://news.ycombinator.com/item?id=4989579

◧◩
6. bragr+N8[view] [source] [discussion] 2022-07-08 21:17:53
>>spiffy+15
Yeah just coming here to say this. Multiple disk failures are pretty probable. I've had batches of both disks and SSDs with sequential serial numbers, subjected to the same workloads, all fail within the same ~24 hour periods.
replies(2): >>mpyne+pb >>schroe+bd
◧◩◪
7. davedu+aa[view] [source] [discussion] 2022-07-08 21:24:41
>>clinto+T7
I hadn't heard of it either until disks in our storage cluster at work started failing faster than the cluster could rebuild in an event our ops team named SATApocalypse. It was a perfect storm of cascading failures.

https://web.archive.org/web/20220330032426/https://ops.faith...

replies(1): >>Flott+lj
◧◩◪
8. mpyne+pb[view] [source] [discussion] 2022-07-08 21:29:37
>>bragr+N8
Seems like it was only a few days ago that there was a comment from a former Dropbox engineer here pointing out that a lot of disk drives they bought when they stood up their own datacenter had been found to all have a common flaw involving tiny metal slivers.
◧◩◪
9. schroe+bd[view] [source] [discussion] 2022-07-08 21:36:43
>>bragr+N8
Had the same experience with (identical) SSDs, two failures within 10 minutes in a RAID 5 configuration.

(Thankfully, they didn't completely die but just put themselves into read-only)

◧◩◪
10. toast0+Xh[view] [source] [discussion] 2022-07-08 21:53:32
>>clinto+T7
I also don't know about literature on this phenomenon, but i recall HP had two different SSD recalls because when the uptime counter rolled over, they would fail. That's not even load dependent, just did you get a batch and power them on all at the same time. Uptime is too high causing issues isn't that unusual for storage, unfortunately.

It's not always easy, but if you can, you want manufacturer diversity, batch diversity, maybe firmware version diversity[1], and power on time diversity. That adds a lot of variables if you need to track down issues though.

[1] you don't want to have versions with known issues that affect you, but it's helpful to have different versions to diagnose unknown issues.

replies(1): >>GekkeP+wr
◧◩◪◨
11. Flott+lj[view] [source] [discussion] 2022-07-08 21:58:07
>>davedu+aa
Great read, thank you!
◧◩◪
12. atheno+vj[view] [source] [discussion] 2022-07-08 21:58:41
>>clinto+T7
Not sure about literature but that was a known thing in the Ops circles I was in 10 years ago: never use the same brand for disk pairs, to minimize wear-and-tear related defects from arising at the same time.
replies(1): >>cestit+Ey4
◧◩◪
13. mceach+Lm[view] [source] [discussion] 2022-07-08 22:10:42
>>clinto+T7
Wikipedia has a section on this. It's called "correlated failure." https://en.wikipedia.org/wiki/RAID#Correlated_failures
◧◩
14. GekkeP+gr[view] [source] [discussion] 2022-07-08 22:27:10
>>spiffy+15
Eek - now I'm glad I wait a few months before buying each disk for my NAS.

Not doing it for this reason but rather financial ones :) But as I have a totally mixed bunch of sizes I have no RAID and a disk loss would be horrible.

replies(1): >>bragr+iP
◧◩◪◨
15. GekkeP+wr[view] [source] [discussion] 2022-07-08 22:28:14
>>toast0+Xh
The crucial M4 had this too but it was fixable with a firmware update.

https://www.neoseeker.com/news/18098-64gb-crucial-m4s-crashi...

replies(1): >>toast0+FU
16. kabdib+7t[view] [source] 2022-07-08 22:34:21
>>davedu+(OP)
I once had a small fleet of SSDs fail because they had some uptime counters that overflowed after 4.5 years, and that somehow persistently wrecked some internal data structures. It turned them into little, unrecoverable bricks.

It was not awesome seeing a bunch of servers go dark in just about the order we had originally powered them on. Not a fun day at all.

replies(2): >>mikiem+A91 >>rbanff+Z05
17. 0xbadc+jA[view] [source] 2022-07-08 23:05:04
>>davedu+(OP)
Even if they're not the same, they're written at the same time and rate, meaning they have the same wear over time, subject to the same power/heat issues, etc.
replies(1): >>pmoria+UL1
18. adrian+nB[view] [source] 2022-07-08 23:09:38
>>davedu+(OP)
I learned this principle by getting a ticket for a burnt out headlight 1 week after I replaced the other one.
replies(1): >>hallwa+ff1
19. bink+qK[view] [source] 2022-07-08 23:51:30
>>davedu+(OP)
Or even if the power supplies were purchased around the same time. I had a batch of servers that as soon as they arrived started chewing through hard drives. It took about 10 failed drives before I realized it was a problem with the power supplies.
◧◩◪
20. bragr+iP[view] [source] [discussion] 2022-07-09 00:19:48
>>GekkeP+gr
Have to be careful doing that too or you'll end up with subtly different revisions of the same model. This may or may not cause problems depending on the drives/controller/workload but can result in you chasing down weird performance gremlins or thinking you have a drive that's going bad.
◧◩◪◨⬒
21. toast0+FU[view] [source] [discussion] 2022-07-09 01:07:36
>>GekkeP+wr
That one looks not too bad, seems like you can fix it with a firmware update after it fails. A lot of disk failures due to firmware bugs end up with the disk not responding to the bus, so it becomes somewhere between impossible and impractical to update the firmware.
◧◩
22. mikiem+A91[view] [source] [discussion] 2022-07-09 03:05:29
>>kabdib+7t
You are never going to guess how long the HN SSDs were in the servers... never ever... OK... I'll tell you: 4.5years. I am not even kidding.
replies(3): >>kabdib+bb1 >>chinat+GA1 >>muttan+nn3
◧◩◪
23. kabdib+bb1[view] [source] [discussion] 2022-07-09 03:20:11
>>mikiem+A91
Let me narrow my guess: They hit 4 years, 206 days and 16 hours . . . or 40,000 hours.

And that they were sold by HP or Dell, and manufactured by SanDisk.

Do I win a prize?

(None of us win prizes on this one).

replies(6): >>mikiem+ae1 >>dang+zh1 >>Amfy+di1 >>mkl+dq1 >>agileA+Rs6 >>pankaj+MN6
◧◩◪◨
24. mikiem+ae1[view] [source] [discussion] 2022-07-09 03:49:12
>>kabdib+bb1
These were made by SanDisk (SanDisk Optimus Lightning II) and the number of hours is between 39,984 and 40,032... I can't be precise because they are dead and I am going off of when the hardware configurations were entered in to our database (could have been before they were powered on) or when we handed them over to HN, and when the disks failed.

Unbelievable. Thank you for sharing your experience!

◧◩
25. hallwa+ff1[view] [source] [discussion] 2022-07-09 04:00:57
>>adrian+nB
Anyone familiar with car repair will tell you that if one headlight burns out you should just go ahead and replace both, because of this exact phenomenon. I suppose with LEDs we may not have to worry about it anymore
◧◩◪◨
26. dang+zh1[view] [source] [discussion] 2022-07-09 04:22:15
>>kabdib+bb1
Wow. It's possible that you have nailed this.

Edit: here's why I like this theory. I don't believe that the two disks had similar levels of wear, because the primary server would get more writes than the standby, and we switched between the two so rarely. The idea that they would have failed within hours of each other because of wear doesn't seem plausible.

But the two servers were set up at the same time, and it's possible that the two SSDs had been manufactured around the same time (same make and model). The idea that they hit the 40,000 hour mark within a few hours of each other seems entirely plausible.

Mike of M5 (mikiem in this thread) told us today that it "smelled like a timing issue" to him, and that is squarely in this territory.

replies(2): >>mikiem+ci1 >>tempes+qF1
◧◩◪◨⬒
27. mikiem+ci1[view] [source] [discussion] 2022-07-09 04:29:05
>>dang+zh1
This morning, I googled for issues with the firmware and the model of SSD, I got nothing. But now I am searching for "40000 hours SSD" and a million relevant results. Of course, why would I search for 40000 hours.

This thread is making me feel a lot less crazy.

replies(2): >>boulos+Qq1 >>dredmo+m44
◧◩◪◨
28. Amfy+di1[view] [source] [discussion] 2022-07-09 04:29:25
>>kabdib+bb1
is this leased to HN as dedicated/baremetal servers or colocation aka HN owns the hardware?
replies(1): >>dang+hj1
◧◩◪◨⬒
29. dang+hj1[view] [source] [discussion] 2022-07-09 04:38:03
>>Amfy+di1
The former.
◧◩◪◨
30. mkl+dq1[view] [source] [discussion] 2022-07-09 05:47:14
>>kabdib+bb1
I wonder if it might be closer to 40,032 hours. The official Dell wording [1] is "after approximately 40,000 hours of usage". 2^57 nanoseconds is 40031.996687737745 hours. Not sure what's special about 57, but a power of 2 limit for a counter makes sense. That time might include some manufacturer testing too.

[1] https://www.reddit.com/r/sysadmin/comments/f5k95v/dell_emc_u...

replies(2): >>boulos+Cq1 >>gomija+je4
◧◩◪◨⬒
31. boulos+Cq1[view] [source] [discussion] 2022-07-09 05:51:27
>>mkl+dq1
See! People should register via mail for those important notifications! (Or alternatively do quarterly checks that your firmware is up to date).
replies(1): >>sqldba+pO4
◧◩◪◨⬒⬓
32. boulos+Qq1[view] [source] [discussion] 2022-07-09 05:52:59
>>mikiem+ci1
I'm hoping that deep in your spam folder is a critical firmware update notice from Dell/EMC/HP/SanDisk from 2 years ago :).
◧◩◪
33. chinat+GA1[view] [source] [discussion] 2022-07-09 07:41:07
>>mikiem+A91
How many other customers will/have hit this?
replies(1): >>qu1j0t+ns3
34. perilu+pD1[view] [source] 2022-07-09 08:09:26
>>davedu+(OP)
There's a principle in aviation of staggering engine maintenance on multiple-engined airplanes to avoid maintenance-induced errors leading to complete power loss.

e.g. Simultaneous Engine Maintenance Increases Operating Risks, Aviation Mechanics Bulletin, September–October 1999 https://flightsafety.org/amb/amb_sept_oct99.pdf

◧◩◪◨⬒
35. tempes+qF1[view] [source] [discussion] 2022-07-09 08:34:14
>>dang+zh1
This kind of thing is why I love Hacker News. Someone runs into a strange technical situation, and someone else happens to share their own obscure, related anecdote, which just happens to precisely solve the mystery. Really cool to see it benefit HN itself this time.
replies(1): >>dang+bc3
◧◩
36. sofixa+tH1[view] [source] [discussion] 2022-07-09 09:04:10
>>spiffy+15
That's why serious SAN vendors take care to provide you a mix of disks (e.g. on a brand new NetApp you can see that disks are of 2-3 different types, and with quite different serial numbers).
◧◩
37. pmoria+UL1[view] [source] [discussion] 2022-07-09 10:01:37
>>0xbadc+jA
Hopefully, regularly checking the disks' S.M.A.R.T status will help you stay on top of issues caused by those factors.

Also, you shouldn't wait for disks to fail to replace them. HN's disks were used for 4.5 years, which is greater than the typical disk lifetime, in my experience. They should have replaced them sooner, one by one, in anticipation of failure. This would also allow them to stagger their disk purchases to avoid similar manufacturing dates.

replies(1): >>justso+TZ1
◧◩◪
38. justso+TZ1[view] [source] [discussion] 2022-07-09 12:23:24
>>pmoria+UL1
https://news.ycombinator.com/reply?id=32033520&goto=item%3Fi...

I've seen too many dead disks with a perfect SMART. When the numbers go down (or up) and triggers are fired then you are surely need to replace the disk[0], but SMART without warnings just means nothing.

[0] my desktop run for years entirely on the disks removed from the client PCs after a failure. Some of them had a pretty bad SMART, on a couple I needed to move the starting point of the partition a couple GBs further from the sector 0 (otherwise they would stall pretty soon), but overall they worked fine - but I never used them as a reliable storage and I knew I can lose them anytime.

Of course I don't use repurposed drives in the servers.

PS and when I tried to post it I received " We're having some trouble serving your request. Sorry! " Sheesh.

◧◩◪◨⬒⬓
39. dang+bc3[view] [source] [discussion] 2022-07-09 20:29:34
>>tempes+qF1
It's also an example of the dharma of /newest – the rising and falling away of stories that get no attention:

HPE releases urgent fix to stop enterprise SSDs conking out at 40K hours - https://news.ycombinator.com/item?id=22706968 - March 2020 (0 comments)

HPE SSD flaw will brick hardware after 40k hours - https://news.ycombinator.com/item?id=22697758 - March 2020 (0 comments)

Some HP Enterprise SSD will brick after 40000 hours without update - https://news.ycombinator.com/item?id=22697001 - March 2020 (1 comment)

HPE Warns of New Firmware Flaw That Bricks SSDs After 40k Hours of Use - https://news.ycombinator.com/item?id=22692611 - March 2020 (0 comments)

HPE Warns of New Bug That Kills SSD Drives After 40k Hours - https://news.ycombinator.com/item?id=22680420 - March 2020 (0 comments)

(there's also https://news.ycombinator.com/item?id=32035934, but that was submitted today)

replies(3): >>dredmo+t44 >>yahelc+eN4 >>winter+ue6
◧◩◪
40. muttan+nn3[view] [source] [discussion] 2022-07-09 22:01:35
>>mikiem+A91
It's concerning that a hosting company was unaware of the 40,000 hour situation with SSD it was deploying. Anyone in hosting would have been made aware of this, or at least should have kept a better grip on happenings in the market.
replies(1): >>dogeco+lH3
◧◩◪◨
41. qu1j0t+ns3[view] [source] [discussion] 2022-07-09 22:39:04
>>chinat+GA1
Every large DC will have hit it (Amazon, Facebook, Google, etc). But it's a shame that all their operational knowledge is kept secret.
replies(1): >>exikyu+3l5
◧◩
42. dspill+lu3[view] [source] [discussion] 2022-07-09 22:57:16
>>spiffy+15
This is why I try to mismatch manufacturers in RAID arrays. I'm told there is a small performance hit (things run towards the speed of the slowest, separately in terms of latency and throughput) but I doubt the difference is high and I like the reduction in potential failure-during-rebuild rates. Of course I have off-machine and off-site backups as well as RAID, but having to use them to restore a large array would be a greater inconvenience than just being able to restore the array (followed by checksum verifies over the whole lot for paranoia's sake).
◧◩◪◨
43. dogeco+lH3[view] [source] [discussion] 2022-07-10 01:10:29
>>muttan+nn3
Yeah, this is why you run all equipment in a test environment for 4.5 years before deploying it to prod. Really basic stuff.
replies(1): >>muttan+WJ3
◧◩◪◨⬒
44. muttan+WJ3[view] [source] [discussion] 2022-07-10 01:38:59
>>dogeco+lH3
The HD makers started issuing warnings in 2020... this was foreseeable
◧◩◪◨⬒⬓
45. dredmo+m44[view] [source] [discussion] 2022-07-10 06:31:42
>>mikiem+ci1
There are times I don't miss dealing with random hardware mystery bullshit.

This one is just ... maddening.

◧◩◪◨⬒⬓⬔
46. dredmo+t44[view] [source] [discussion] 2022-07-10 06:32:41
>>dang+bc3
Popularity is a very poor relevance / truth heuristic.
replies(1): >>gpshea+6C5
◧◩◪◨⬒
47. gomija+je4[view] [source] [discussion] 2022-07-10 08:59:22
>>mkl+dq1
It might not be nanoseconds, but something that's a power of 2 number of nanoseconds going into an appropriately small container seems likely. For example, a 62.5MHz counter going into 53 bits breaks at the same limit. Why 53 bits? That's where things start to get weird with IEEE doubles - adding 1 no longer fits into the mantissa and the number doesn't change. So maybe someone was doing a bit of fp math to figure out the time or schedule a next event? Anyway, very likely some kind of clock math that wrapped or saturated and broke a fundamental assumption.
replies(1): >>dreamc+Bg6
◧◩◪◨
48. cestit+Ey4[view] [source] [discussion] 2022-07-10 12:49:01
>>atheno+vj
We used to use the same brand, but different models or at least ensure they were from different manufacturing batches.
◧◩◪◨⬒⬓⬔
49. yahelc+eN4[view] [source] [discussion] 2022-07-10 14:34:30
>>dang+bc3
Easy to imagine why this didn’t capture peoples’ attention in late March 2020…
replies(2): >>dang+TQ5 >>mcv+8W6
◧◩◪◨⬒⬓
50. sqldba+pO4[view] [source] [discussion] 2022-07-10 14:42:21
>>boulos+Cq1
A lot of companies have teams dedicated to hardware that don’t give a shit about it. And their managers don’t give a shit.

Then the people under them who do give a shit, because they depend on those servers, aren’t allowed to register with HP etc for updates, or to apply firmware updates, because “separation of duties”.

Basically, IT is cancer from the head down.

◧◩
51. rbanff+Z05[view] [source] [discussion] 2022-07-10 16:13:25
>>kabdib+7t
I had a similar issue, but it was a single RAID-5 array and wear of some other manufacture defect. They were the same brand, model, and batch. When the first failed and the array got in recovery mode I ordered 3 replacements and upped the backup frequency. It was good that I did that because the two remaining drives died shortly after.

The lesson I learned is that the three replacements went to different arrays and we never again let drives from the same batch be part of the same array.

◧◩◪◨⬒
52. exikyu+3l5[view] [source] [discussion] 2022-07-10 18:09:38
>>qu1j0t+ns3
I understand BackBlaze is more HDD rather than SSD, but perhaps they might have some level of awareness.
◧◩◪◨⬒⬓⬔⧯
53. gpshea+6C5[view] [source] [discussion] 2022-07-10 19:50:58
>>dredmo+t44
I wanted to upvote this comment but that just feels wrong.
replies(1): >>dredmo+X36
◧◩◪◨⬒⬓⬔⧯
54. dang+TQ5[view] [source] [discussion] 2022-07-10 21:25:02
>>yahelc+eN4
Yes, an enterprisey firmware update - all very boring until BLAM!
◧◩◪◨⬒⬓⬔⧯▣
55. dredmo+X36[view] [source] [discussion] 2022-07-10 22:54:16
>>gpshea+6C5
You're a good man, Charlie Brown.
◧◩◪◨⬒⬓⬔
56. winter+ue6[view] [source] [discussion] 2022-07-11 00:11:50
>>dang+bc3
Interesting how something that is so specifically and unexpectedly devastating, yet known for such a long time without any serious public awareness from companies involved, is referred to as a "bug".

It makes you lose data and need to purchase new hardware, where I come from, that's usually referred to as "planned" or "convenient" obsolescence.

replies(1): >>mulmen+Kl6
◧◩◪◨⬒⬓
57. dreamc+Bg6[view] [source] [discussion] 2022-07-11 00:31:35
>>gomija+je4
53 is indeed a magic value for IEEE doubles, but why would anybody count an inherently integer value with floating-point? That's a serious rookie mistake.

Of course there's no law that says SSD firmware writers can't be rookies.

replies(1): >>lultim+SQ6
◧◩◪◨⬒⬓⬔⧯
58. mulmen+Kl6[view] [source] [discussion] 2022-07-11 01:17:46
>>winter+ue6
The difference between planned and convenient seems to be intent. And in this context that difference very much matters. I wouldn’t conflate the two.
replies(1): >>winter+Kn6
◧◩◪◨⬒⬓⬔⧯▣
59. winter+Kn6[view] [source] [discussion] 2022-07-11 01:38:57
>>mulmen+Kl6
Depends on who exactly we are talking about as having the intent...

Both planned and convenient obsolescence are beneficial to device manufacturers. Without proper accountability for that, it only becomes a normal practice.

replies(1): >>mulmen+Vn6
◧◩◪◨⬒⬓⬔⧯▣▦
60. mulmen+Vn6[view] [source] [discussion] 2022-07-11 01:41:15
>>winter+Kn6
> Depends on who exactly we are talking about as having the intent...

The manufacturer, obviously. Who else would it be?

Could be an innocent mistake or a deliberate decision. Further action should be predicated on the root cause. Which includes intent.

◧◩◪◨
61. agileA+Rs6[view] [source] [discussion] 2022-07-11 02:36:33
>>kabdib+bb1
Bang on!

https://news.ycombinator.com/item?id=32048148

◧◩◪◨
62. pankaj+MN6[view] [source] [discussion] 2022-07-11 06:24:08
>>kabdib+bb1
Do they use SSD on space missions aswell?
replies(1): >>galois+oc7
◧◩◪◨⬒⬓⬔
63. lultim+SQ6[view] [source] [discussion] 2022-07-11 06:49:08
>>dreamc+Bg6
Full stack JS, everything is a double down to the SSD firmware!
◧◩◪◨⬒⬓⬔⧯
64. mcv+8W6[view] [source] [discussion] 2022-07-11 07:46:59
>>yahelc+eN4
Was HN an indirect casualty of Covid?
◧◩◪◨⬒
65. galois+oc7[view] [source] [discussion] 2022-07-11 10:36:01
>>pankaj+MN6
Only for 4 years, 206 days and 16 hours.
[go to top]