Real-World ZFS Insights: Mistakes, Lessons & Future 🔧

https://lawrence.video/

In this video, I’m joined by ZFS expert Allan Jude with Klara Systems to discuss some very technical details of ZFS:

- Fine-tuning alignments for performance
- Drive considerations
- Exciting new developments like ZFS working with containers
- The roadmap to getting more performance from NVMe drives

Klara Systems OpenZFS Horror Stories:
https://klarasystems.com/webinars/openzfs-horror-stories/

Klara Systems Isolating Containers with ZFS and Linux Namespaces:
https://klarasystems.com/articles/isolating-containers-with-zfs-and-linux-namespaces/

Connect With Us
---------------------------------------------------
+ Hire Us for a project: https://lawrencesystems.com/hire-us/
+ Toms' Twitter 🐦 https://twitter.com/TomLawrenceTech
+ Our Website https://www.lawrencesystems.com/
+ Our Forums https://forums.lawrencesystems.com/
+ Instagram https://www.instagram.com/lawrencesystems/
+ Facebook https://www.facebook.com/Lawrencesystems/
+ GitHub https://github.com/lawrencesystems/
+ Discord https://discord.gg/ZwTz3Mh

Lawrence Systems Shirts and Swag
---------------------------------------------------
►👕 https://lawrence.video/swag/


AFFILIATES & REFERRAL LINKS
---------------------------------------------------
Amazon Affiliate Store
🛒 https://www.amazon.com/shop/lawrencesystemspcpickup

UniFi Affiliate Link
🛒 https://lawrence.video/unifi-affiliate

All Of Our Affiliates help us out and can get you discounts!
🛒 https://lawrencesystems.com/partners-we-love/

Gear we use on Kit
🛒 https://kit.co/lawrencesystems

Use OfferCode LTSERVICES to get 10% off your order at
🛒 https://www.techsupplydirect.com?aff=2

Digital Ocean Offer Code
🛒 https://m.do.co/c/85de8d181725

HostiFi UniFi Cloud Hosting Service
🛒 https://hostifi.net/?via=lawrencesystems

Protect your privacy with a VPN from Private Internet Access
🛒 https://www.privateinternetaccess.com/pages/buy-vpn/LRNSYS

Patreon
💰 https://www.patreon.com/lawrencesystems

Chapters
00:00 Intro – ZFS mistakes, snapshots & recovery questions
01:03 ZFS Expansion – why the slow, careful rollout
02:28 Testing new features – prebuilt Ubuntu master packages
03:16 Block Reference Tree & cloning – VMs and synthetic full backups
04:02 Dedup reality – benefits, costs & scaling limits
07:40 Shorn writes – partial file writes & corruption
09:14 Crash recovery model – TXGs vs fsck
10:14 5-second TXGs – batching changes to cut write amplification
10:37 SLOG/Intent Log – when it helps and when it doesn’t
12:45 Databases & sync writes – when durability really matters
14:23 Mirrors, failure patterns – HDD noise vs SSD “sudden death”
15:55 RAIDZ straggler detection – bypassing slow disks
17:26 1 PB museum design – off-site real-time replica with ZFS send
18:23 Tuning TXGs – should you change the 5s default? trade-offs
22:27 ZIL on main devices – NVMe pools and million-IOPS reality
23:38 CPU vs device latency – when optimizations don’t pay
24:50 NVMe queues & queue depth – feeding 64×64 command lanes
27:04 ZFS overhead – checksums, compression & memory bandwidth
34:59 NFS recordsize mismatch – why 64K broke 128K records
36:39 Zero reads/writes & checksums – integrity side effects
38:55 Design for worst-case – scrub under load, still hit targets
40:29 Mirrors vs RAIDZ – IOPS math and efficiency trade-offs
42:23 HA split-brain import – TXG rewind to limit damage
44:26 Why rewind works – block pointers across drives
48:05 Too many datasets/snapshots – send/receive indexing costs
50:35 Copy-on-write saves the day – rewinding TXGs
51:42 Delegating datasets to containers – LXD/Docker with ZFS