Старый жесткий диск с подозрительным шумом прошел все тесты, но эксперт решил его заменить

A little while ago, I was gifted an old Intel NUC mini PC. I had no idea what it did before it landed in my hands, but I eventually gave it the job of being a media PC for my aging smart TV. Its 1TB mechanical hard drive has plenty of space for a local media library, which would save me from having to rely on external storage. The problem is that the hard disk is 13 years old, and every so often it makes concerning noises that make me glance down from the TV. A quick check with basic tools still reported PASSED, and Linux’s disk utility agreed. Rather than trust one reassuring test, I installed the free and open source app Scrutiny to see whether the drive was genuinely healthy or on its way to failure. My first health checks all came back as healthy The reassuring green verdict left out more history than I expected I started with the quickest check I could think of: sudo smartctl -H /dev/sda The result was rather reassuring and a little suspicious: SMART overall-health self-assessment test result: PASSED Linux Mint’s Disks utility was also in agreement. According to both, there was nothing catastrophically wrong with the HGST (Hitachi) drive. Digging into the individual S.M.A.R.T (Self-Monitoring, Analysis, and Reporting Technology) attributes didn’t immediately expose a disaster either. My baseline looked like this: SMART Attribute | Value Reallocated sectors | Current pending sectors | Offline uncorrectable sectors | CRC errors | Power-on hours | 11,136 hours Load/unload cycles | 403,715 Power-off retract count | 66 Historical maximum temperature | 53°C Previous SMART self-tests | None recorded I filtered out the usual wall of SMART output down to more useful attributes: sudo smartctl -A /dev/sda | grep -E 'Reallocated_Sector|Power_On_Hours|Load_Cycle|Current_Pending|Offline_Uncorrectable' Scrutiny made the distinction much clearer. Its normal table only showed a Load Cycle Count of 60, but that’s just a normalized SMART value. Expanding the attribute exposed the raw count of 403,715 underneath. That number looked scary, but it didn’t necessarily mean the disk was failing. HGST actually specifies at least 600,000 normal load/unload cycles for this type of disk. What it did mean, though, was that this disk had already accumulated about two-thirds of that figure. For reference, this is what I generally consider warning signs: SMART item | My result | What would concern me Reallocated sectors | | ANY new reallocations Pending sectors | | ANY non-zero value Offline uncorrectable | | ANY non-zero value UDMA CRC errors | | A rising count which would indicate a SATA cable issue Power-on hours | 11,136 | Any number above the manufacturer's threshold Load/unload cycles | 403,715 | Any number above the manufacturer's threshold for assumed normal cycles Power-off retracts | 66 | A number above 20,000 emergency unloads Temperature | Max 53°C (127.4°F) | Any number exceeding the safe operation limit for the disk HGST’s own documentation puts the drive’s service life at approximately five years, or 20,000 power-on hours under assumed workloads. It also specifies 0–60°C (32–140°F) operating conditions. Those figures are based on a lot of assumptions, so they’re not countdown timers. Just because the drive hits 20,000 hours doesn’t mean it will fall over at 20,001. Scrutiny showed me the difference between health and wear The dashboard made the numbers understandable instead of burying them in SMART output smartctl had already given me plenty of raw data, but if I’m honest, the results looked like they were meant for a hard drive engineer and not me. Installing Scrutiny made that same information much easier to interpret, so I installed version 0.9.3 on the NUC with Docker Compose and gave it access to the physical drive: services: scrutiny: image: ghcr.io/analogj/scrutiny:v0.9.3-omnibus cap_add: - SYS_RAWIO devices: - /dev/sda:/dev/sda volumes: - /run/udev:/run/udev:ro Then I started it with: docker compose up -d Scrutiny offered a history that SMART couldn’t. Instead of checking once and forgetting about it, Scrutiny stored each collection and plotted how the drive changed over time. That made it much easier to answer questions I needed answers to most: Was the temperature rising under sustained use? Yes Were the load cycles climbing too quickly? No Had there been any pending or reallocated sectors? No Were any attributes getting worse between checks? No It also helped explain a few SMART results that initially didn’t make sense. Scrutiny showed Power-On hours as 75, but expanding that row revealed the figure was actually more than 11,136 hours. The same thing happened with the Load Cycle Count, where a normalized value of 60 hid the actual raw count above 403,000. Again, this is context, and not a countdown. Scrutiny’s failure-rate figures weren’t a prediction that my hard drive has a x% chance of failing. That context was exactly what I needed to see because it showed me what the hard drive had already lived through. Take a baseline SMART snapshot before stress-testing an old drive. If there are pending, reallocated, or uncorrectable sector counts in the results, consider the disk failing. Don’t perform any more tests and start backing up and removing your data. Scrutiny Platforms Linux, macOS, FreeBSD, Docker, and Windows binaries Developer(s) Jason Kulatunga (AnalogJ), with maintenance by Aram Akhavan Scrutiny is a free, open-source hard drive health dashboard that turns S.M.A.R.T data into readable health checks, historical trends, temperature graphs, and failure warnings. It can monitor multiple drives through a web interface and run in Docker or as a manual installation. Price model Free and open source, MIT licensed I made the old drive prove it could still do the job It survived every self-test, then sounded much less convincing under a sustained write Scrutiny had shown me how much life this disk had already had, but I stopped trusting the easy PASSED verdict and made it do some real work. I started with SMART’s built-in self tests: sudo smartctl -t short /dev/sda sudo smartctl -l selftest /dev/sda The short test was completed in two minutes without error at 11,139 hours. That was fine for a short test, but I wanted something more thorough: sudo smartctl -t long /dev/sda sudo smartctl -l selftest /dev/sda This test took around three hours to complete, all the while the hard disk was making a lot of unsettling sounds. The extended test was completed without any errors at 11,143 hours. That result was more important because the drive had actually scanned its media rather than just reporting that no failure thresholds were crossed. After both tests, the important counters were still squeaky clean. Reallocated sectors were still 0, and pending sectors too. Offline uncorrectable sectors stayed at 0, load cycles were still 403,715, and the drive had settled back to its normal operating temperature of 39–40°C (102.2–104°F). I then imitated the job I actually wanted this drive to do. I started a large sequential write to mimic gradually filling it up with media: dd if=/dev/zero of=~/hdd-media-test/media-test.bin bs=64M count=800 oflag=direct status=progress Scrutiny showed the temperature climbing to 46°C (114.8°F), but the SMART counters kept giving the all clear. The problem at this stage was that the drive itself was making horrendous mechanical noises, badly enough for me to call off the read test. What frustrated me was that I could hear the disk was not OK, but the deeper checks still came back clean. The SMART error logs still showed No Errors Logged, and the kernel also showed no I/O errors or SATA resets. Ironically, the most worrying result of the whole testing process was one no diagnostic tool could show me. Every test passed, and I’ll still be replacing the drive Healthy enough to boot my TV box isn't the same as healthy enough for new storage There’s a perfectly reasonable argument for keeping this old hard drive. It has no reallocated, pending, or uncorrectable sectors, no interface errors, and no logged SMART errors of any kind. Both the short and extended tests were completed without issue, and their temperatures even stayed within a normal range. To say the least, it’s frustrating to see these results while still knowing there is an issue with the drive because of the noise. If its only job were to keep booting Linux and running the TV interface, I probably wouldn’t replace it just because it’s got over 10K power-on hours. The problem was that I was about to give it a much bigger responsibility. Building a media library means copying hundreds of gigabytes to it, then relying on it every day to actually keep that data safe. At that point, the 400,000+ load and unload cycles, its age, plus the horrible sounds it made during sustained writes say a lot more than the green health badge. SMART can flag most common failure signs, but can’t predict mechanical, electronic, or sudden failure. And honestly, I don’t want to sit there watching my favorite sci-fi shows on Netflix while listening to the hard drive straining in the background. For now, I’m keeping Scrutiny installed and letting the drive continue with its light TV-box duties, but I’ll be imaging it with my FOG server and looking for a new SSD drive for it. Losing totally replaceable movies wouldn’t be the end of the world, but having to rebuild that library would still be a huge pain. In the end, the drive passed every test I threw at it. Scrutiny didn’t prove my intuition regarding its imminent failure, but it gave me enough context to realize that “not failing” isn’t a standard for keeping a hard disk going.