Lolight

Syslog is the first indicator but only if you have persistent logging configured -- which most users don't. Honestly though -- it's pretty much impossible to know a boot drive is about to fail while it's still actively serving as one. By the time symptoms appear you're often already in trouble. The more practical approach is two-fold. First, check what you already have -- ChipGenius or Flash Drive Information Extractor will tell you what controller and NAND type is actually inside your current drive without disturbing it. That tells you a lot about what to expect long term. SanDisk being the exception since their controllers block identification tools entirely. If you want to actively test the drive you'd need to run H2testw on it -- full write and verify cycle that surfaces marginal NAND. Not practical for most people while it's the active boot device.

Exactly right on all counts -- you've worked through the logic correctly. The Innodisk drives are an excellent choice for a single boot drive setup. The same specifications that make them good for a mirror make them ideal as a single drive. A single Innodisk running Unraid will likely outlast everything else in the system. Buying an identical spare is exactly the right approach. On capacity -- yes, Unraid Flash Creator supports drives larger than 32GB now. 64GB is fine. But if you encounter any problems with formatting use Rufus first. Out of curiosity -- how did you confirm the Planar TLC on the SanDisk? Standard tools like ChipGenius or Flash Drive Info Extractor can't read SanDisk controller data since they lock out NAND identification. Wondering if you have a source for that specific model's internals.

Since you don't want to use fTPM, your license needs to remain on a USB drive. When you migrate to internal boot the two Innodisk drives handle the boot function -- but the license anchor stays on a separate USB drive that must remain permanently connected. So your actual configuration will be three USB drives total -- two Innodisk drives in the mirrored boot pool plus your existing USB drive repurposed as a dedicated license anchor. On hardware portability -- when you move to new hardware you'll need to bring all three USB drives along with your array and cache drives. The license travels with the dedicated USB license drive, not with the boot pool drives. Worth considering an alternative -- a discrete dTPM module for your motherboard's TPM header (if available) typically costs $10-20 and eliminates the third USB drive requirement entirely while avoiding the fTPM vulnerabilities the guide covers. If your motherboard has a physical TPM header this is worth looking at before committing to the three USB drive approach but your future migration would be restricted by the TPM header availability on the new machine. Now to your specific questions: The mirror continues operating on the surviving drive after a failure and Unraid notifies you. Be aware of a documented in an NVMe configuration bug in 7.3.1 where replacing the failed mirror member can leave a stale device entry requiring a terminal zpool detach command to clean up. Supposedly fixed by now. For port selection -- USB 2.0 ports for all three drives. USB 3.x drives in USB 3.x ports generate more heat. Use your motherboard's internal USB 2.0 headers with 9-pin adapter cables. On the partition structure -- the boot partition is ZFS not BTRFS. 8GB minimum -- sufficient for boot but it's up to you which size to select. The remainder becomes the data partition if you choose to use it. The Innodisk drives are the right choice for this application -- but make sure you account for the third drive in your planning before purchasing. One final note: the mirrored USB boot pool is the least utilized and tested configuration among all available Unraid boot options. Real world deployments are rare and documented cases even rarer. I've documented it in the guide based on JorgeB's confirmation that it's supported, but I don't use this configuration myself and can't speak from personal experience. Proceed with that context in mind and make sure you have solid backups of your config folder before attempting the migration. One more thought on the redundancy goal specifically -- it's worth stepping back and evaluating whether the additional complexity of this setup is justified for your situation. A mirrored boot pool protects against a single drive failure. But Unraid's config folder backup -- which you should be maintaining regardless of boot configuration -- already provides recovery from a complete boot drive failure in minutes. The question is whether the seamless failover of a mirror justifies three USB drives, the barely tested configuration complexity, and the migration process compared to simply maintaining a current config backup and keeping a spare USB drive on hand. For some use cases the mirror makes sense. For others the simpler answer is a quality boot drive with a current backup. Worth thinking through before committing

Fair question. The surplus origin isn't stated in the listing but it's inferable from several details. The drives carry LYTX DriveCam DVR-S70 custom branding -- LYTX is a commercial fleet dashcam company. Innodisk produces custom branded drives for OEM customers as a standard service. The drives arrive in individual anti-static bags which is Innodisk's factory new packaging. The iTracker output across tested units shows an erase count of 2 -- consistent with factory bench testing, not field deployment. The most likely scenario: LYTX transitioned to a different drive format for their DriveCam systems. The remaining inventory of custom branded Innodisk units had no further use to them and entered the secondary market through this eBay seller. The seller's other listings confirm the pattern -- they specialize in surplus industrial and laboratory equipment including Watlow process controllers, Laurel panel meters, and LabJack data acquisition hardware. This is an industrial surplus liquidation seller, not a consumer electronics reseller. None of this is stated explicitly in the listing -- it's assembled from the available evidence. But the combination of OEM branding, specific factory packaging, factory-baseline iTracker readings, and the seller's overall inventory profile makes the surplus origin the most coherent explanation by a significant margin.

Fair points all around -- let me address them. Your second point is the most valid one. The vulnerability window is when you plug the drive into your Windows PC to run the Flash Creator -- that's when a BadUSB attack would execute, not on the Unraid server itself. That's a legitimate concern. The reason I'm not worried about this specific listing is the iTracker testing. Eight drives from this listing were tested with that monitoring tool which communicates through Innodisk's own vendor protocol. All eight returned factory-correct output. A controller running BadUSB firmware alongside Innodisk's proprietary monitoring protocol simultaneously would require reverse engineering that vendor interface entirely -- an enormous (!!! x 100) engineering effort when generic $2 drives require none of that. iTracker is Innodisk's proprietary health monitoring utility -- not publicly downloadable, requires contacting Innodisk support directly with the drive's part number. https://www.innodisk.com/en/technology/itracker Happy to share my copy with you or anyone who wants to verify their own drives. Send me a PM. On the price -- this is surplus liquidation from LYTX, a commercial fleet dashcam company that transitioned to a different drive format. The same drive costs $120 or more at industrial distributors right now. Surplus liquidation pricing reflects what LYTX needed to clear inventory, not what the drives are worth.

The scenario you're describing -- a seller acquiring legitimate surplus drives and reprogramming the controllers with BadUSB firmware before resale. The SM3261 AB firmware can be updated using Silicon Motion's MPTool utilities. The tooling exists and 3,000 units is a scale where automated flashing rigs make it operationally feasible. So why am I not worried about this specific listing? Eight drives purchased from this listing were tested with Innodisk's proprietary iTracker utility -- the health monitoring tool that communicates with the Silicon Motion SM3261 controller through Innodisk's own vendor protocol. All eight returned consistent factory-baseline output: 99.93% health, average erase count of 2, firmware version O0917v1, controller identified as SMI3261. A reprogrammed controller running BadUSB firmware would not respond to Innodisk's proprietary monitoring commands. Getting a drive to simultaneously implement BadUSB behavior AND correctly respond to Innodisk's protocol would require much more sophisticated firmware development than standard BadUSB attacks involve. Eight units all returning consistent iTracker output is proven evidence against firmware tampering. There's also the operational context -- Unraid is Linux running headless without an active terminal session in normal operation. BadUSB keyboard injection needs an active session to land in. The attack surface on a headless NAS is essentially nonexistent regardless. The concern is legitimate as a general principle. The iTracker evidence makes it super unlikely in this specific case.

That's quite a characterization without a single specific example to back it up. You're also answering for Craigb, who expressed his own interest in the explanation. If Craigb wants to pass that's his call to make. Which post of mine was incorrect? Point to something specific and I'll address it directly.

Check if you can see the drive as a hardware device on Windows in disk management. If you can't then try different USB ports, also different machines if available (Windows, Mac, Linux). If the drive gets recognized then the controller is functional and files are potentially recoverable via software recovery tools. If it doesn't then it's bricked.

For a replacement stick choose a quality one, meaning an industrial USB flash drive.

Unfortunately iTracker isn't available as a public download -- you're correct that the product page exists but provides no download link. The only way to obtain it is by contacting Innodisk's web request team directly through their website contact form. In my experience they responded in 3 days. Be prepared to provide some information about yourself -- they asked for a company name and its address before sending the file. If providing that information isn't something you're comfortable with, feel free to send me a PM -- I'm happy to share my copy of iTracker with any forum member who needs it. The iTracker output on purchased units from the listing confirms 99.93% health and average erase count of 2 across all 8 tested drives -- factory new condition on every unit tested. A count of 2 is factory quality control testing and firmware initialization.

I'll make an attempt at explaining it, but let's get your current upgrade sorted first. Once everything is stable and the Sony is sent to its well-deserved retirement, let me know and I'll walk through what's likely behind its unusual longevity. Or show how to achieve similar longevity in the replacement drive. Don't just go out and buy another one without knowing the details

Yes, an 8GB capacity is more than enough for Unraid. Your planned combo of the SanDisk MobileMate on a USB 2.0 port paired with the industrial MLC card is a good setup. The MobileMate card reader should have a unique serial number (+VID/PID) Unraid requires for its license.

The occasional failure to enumerate on Windows is worth paying attention to -- that's an early warning signal of a drive approaching end of reliable operation rather than a random quirk. A Sony flash drive from 2010 has had a good run by any measure. But 15+ years puts any consumer USB drive well past its expected service life regardless of how it appears to be performing day to day. Enumeration failures are typically the first visible symptom before more serious read/write reliability issues develop. Before your next upgrade attempt it might be worth starting fresh with a known good quality drive rather than continuing to troubleshoot around a 15 year old flash drive that's already showing warning signs. There's a "PSA in Industrial MLC" in the Pre-Sales section covering what's currently available for boot drives if you haven't seen it -- including options that should last considerably longer than the Sony has.

The Kingston DT SE9 is a reasonable choice from that recommendation era but worth verifying it's genuine before trusting it as a long term boot device. I purchased the exact same model years ago on the same forum recommendation. Two years later I discovered greenish corrosion appearing inside the metal housing near the USB connector. Running ChipGenius on it revealed a FirstChip controller -- a controller never used in genuine Kingston DT SE9 production. It was a counterfeit that had a unique GUID and passed every casual inspection. A unique GUID by itself is not confirmation of a genuine drive. Counterfeits can and do register unique GUIDs. Worth running ChipGenius or Flash Drive Information Extractor on your drive to confirm what's actually inside. Genuine DT SE9 drives used Phison or Silicon Motion controllers depending on production era -- anything else warrants investigation. And if you do end up needing a replacement for the trial or eventually for a purchased license -- there's a PSA just below this thread covering what's currently available that goes significantly beyond the 2022 recommendations.

Your USB drive is now your license anchor -- if it fails your system won't be able to validate its license on the next reboot regardless of how healthy your SSD mirror is. Worth making sure whatever is in that slot is a quality drive.

Before you grab a new flash drive for the fresh trial -- what are you planning to use? If you're going to be purchasing a license anyway it's worth starting with a quality drive rather than whatever's convenient. There's a time-sensitive opportunity just below this thread that's directly relevant: "PSA on industrial MLC..."

USB drives from 15 years ago were far more durable and ideal for Unraid. Times have changed, but you can still grab an industrial drive for a bargain (always use USB 2.0 port or motherboard header) https://forums.unraid.net/topic/199642-psa-on-industrial-mlc-usb-boot-drives-mlc-nand-end-of-life-and-last-accessible-stock/#comment-1629106

This is a community heads-up for anyone running Unraid on USB boot or planning to. Kioxia (formerly Toshiba) issued a formal End-of-Life notice in March 2026 covering all TSOP MLC NAND production. https://www.semicone.com/article-412.html Final orders accepted until September 15, 2026. Final shipments by March 15, 2027. https://www.mouser.com/PCN/Kioxia_America_Inc_KIOXIA_TSOP_EOL_Notification_2026March.KAI..pdf Samsung completed final MLC shipments in June 2026. The last major MLC producers are done. Industrial MLC USB drives are entering permanent scarcity. The authorized distributor price for the Innodisk 3ME (DEUA1-64GI61BW1SC) currently sits at $120 at specialist distributors and $1,252 at DigiKey for the same drive. The opportunity: There is currently a surplus liquidation listing on eBay for this exact drive -- 64GB Innodisk 3ME. The drives originated from a LYTX commercial fleet dashcam application -- hardware specified for always-on duty far more demanding than a NAS boot device will ever impose. Brand new OEM in factory anti-static pouches -- 3K+ units sold across three previous listings. 🏷️ eBay item 327321790467 🏷️ https://www.ebay.com/itm/327321790467 Original listing price: $3.99 Previous listing price: $13.99 (as of August 17) Current listing price: $20.00 (as of September 7) Full specifications: https://www.innodisk.com/en/products/flash-storage/others-usb-usb-edc/usb-drive-3me For context on why this specific drive matters for Unraid boot: Toshiba MLC, 60-bit ECC, metal housing, 30μ gold contacts, industrial BOM, custom industrial firmware, -40°C to +85°C rating, S.M.A.R.T. capable* After Kioxia's March 2027 final shipment deadline new production of this NAND category ends permanently. The $3.99 price reflects surplus liquidation pricing set before the MLC EOL announcements created market awareness of the scarcity.For full technical context see: https://forums.unraid.net/topic/196967-unraid-boot-device-guide-usb-and-internal-boot-hardware-selection-and-risk-tradeoffs/ *Drive health monitoring available via Innodisk’s proprietary iTracker utility -- vendor-specific S.M.A.R.T. like metrics such as health percentage and erase count; not standard ATA S.M.A.R.T. Tool available from Innodisk support on request. This post has no commercial relationship with the seller. No financial interest in any form and shape. Just a community heads-up on a closing window that most people won't notice until it's gone.

The read-only protection mode your drive triggered is actually a meaningful safety feature that isn't widely understood in the community. When certain USB controllers (most) detect imminent failure they lock the drive read-only specifically to preserve existing data -- sacrificing bootability to protect configuration integrity. Your situation demonstrated it working exactly as designed. It's worth noting that NVMe drives in combined boot and cache configurations don't have standardized or guaranteed equivalent “fail-safe read-only mode". Instead, behavior varies significantly by manufacturer and firmware implementation. A failing NVMe in internal boot role is less likely to lock itself in read-only state to protect your data -- it just fails by disappearing from BIOS entirely, or showing up but throwing I/O errors on both reads and writes, or becoming stuck in a reset loop. Your recovery path -- new USB, copy config, boot -- is the USB boot architecture working as intended even in failure.

Historical unreliability of the current crop of consumer USB flash is real, but the opposite is also true for higher-quality legacy MLC and current industrial MLC/SLC drives. The historical record also shows that legacy consumer USB flash drives are very reliable -- some of those are still being used as Unraid boot devices 2 decades later. fTPM licensing (if you plan on using it) adds a layer that USB licensing doesn’t have -- BIOS updates, microcode patches, or CMOS clears that can reset the identifier. But those risks are not applicable to dTPM - the discrete TPM module. I'm just trying to present a bigger picture. You can only make an informed decision when you're aware of all known pro's and con's. One more thought -- internal boot is a brand new feature. The Unraid community's own golden rule applies here as much as anywhere: new major features need time to prove themselves in real world conditions across diverse hardware configurations. You mentioned no issues with your current USB setup. That's actually the most important data point. A working system with no complaints is not a problem requiring a solution -- especially when that solution involves a fundamental architectural shift rather than a routine upgrade. Unlike a software version that can be rolled back, migrating to internal boot changes your boot architecture, your licensing anchor, and your failure domain in ways that aren't easily reversed. The long term reliability picture for internal boot won't be clear for another year or two minimum! If your USB setup works, the most defensible position right now is to wait. Let others accumulate the real world experience first. Revisit in 12-24 months when the community has long term data rather than early adopter enthusiasm

I meant given that you're still in the planning stage -- it's easier to evaluate alternatives now than after the fact. The configuration you're describing, 8GB boot partition shared with docker containers and download storage on the same NVMe, means a single drive failure loses your OS configuration and all that data at the same time. That's a huge failure spread for one physical device. A mirrored USB boot pool handles the redundancy without touching your NVMe at all. Your NVMe stays dedicated to docker and storage with its full capacity available. The two functions stay independent -- if either fails the other is unaffected. https://forums.unraid.net/topic/196967-unraid-boot-device-guide-usb-and-internal-boot-hardware-selection-and-risk-tradeoffs/

Before going further with the cache drive migration -- did you know that Unraid 7.3 actually supports a mirrored USB boot pool? Two USB drives in a ZFS mirror, redundancy against single drive failure, zero impact on your cache drive or its data. Might be worth considering given where things stand right now.

Yes it was promised here: https://newsletter.unraid.net/p/unraid-2025-year-in-review Will it actually happen? Who knows...

Or how about keeping it reliable and simple by getting a super-reliable USB flash drive and attaching it to the motherboard USB 2.0 header? An original "internal boot" configuration!

The boot drive on the motherboard header is the right move. But I'd suggest getting a real industrial USB drive instead of the microSD card/reader combo. It's a much more durable option. Get a couple or more. You will not find them at this price again once the current stock is depleted. P/N: DEUA1-64GI61BW1SC ebay Item #326046070546