For literally the past few days, a new stable version of ZFSonLinux has been released, a project that is now central in the world of OpenZFS development. Farewell, OpenSolaris, hello fierce GPL-CDDL incompatible world of Linux.
Below is a review of the most interesting things (well, 2200 commits!), and for dessert â a bit of intrigue.
New features
Of course, the most anticipated one is native encryption. Now you can encrypt only the necessary datasets with the built-in ZFS encryption, and (in my opinion â the main thing) â you can send encrypted data through zfs send and check the integrity of the data WITHOUT decryption using built-in tools, all the possibilities for maintaining ZFS data integrity will be at your disposal!
Next in importance, it is worth mentioning the long-awaited TRIM. Yes, it took a long time to reach production. Partly because for CoW file systems, the issue of SSD wear is not so critical. But now we can all breathe easy â zpool trim will save our precious flash drives.
Now you can remove accidentally added vdev arrays from the pool (but only if it's sparse or mirror). A useful little thing.
Next in our hit parade is pool checkpoints. Briefly â snapshots for the entire state of the pool, BUT allowing the rollback of changes not only in data but also in enabled pool features and structural changes. Another way to secure yourself.
Pool initialization â filling the underlying storage with zeros. Useful for operation in environments with thin provisioned disks to clearly allocate space and exclude unexpected performance drops in the future.
Project accounting and quota â in the existing quota mechanism, it is now possible to use project segmentation.
Channel programs â the ability to perform administrative tasks atomically using Lua scripts. There are limits on execution time and memory. If you are into automation â this is for you.
Direct IO â for simplicity, we rolled out Direct IO functionality, nothing has changed internally (calls go straight past the cache), but now software wishing to operate in this mode will not suffer.
Projekt Pyzfs is now integrated into the main repository and taken under the wing of the ZFSonLinux project. There are now more tools for management from Python (and support for the module will be less of a concern). Many Python scripts have also been adapted for Python 3.
And now for the delicious part â performance
Nu kan metadata fĂžrst lĂŠses ved scrub- og resilver-operationer, og derefter dataene i den maksimale seriel form â pĂ„ den mĂ„de foregĂ„r genopretning af arrayet og integritetskontrol med maksimal hastighed.
Allokeringsklasser â vdev-arrays har nu en ny type lager, sĂ„ metadata/ deduplikerings-tabeller (DDT) / datablokke mindre end X KB kan flyttes til et separat vdev-array fra mere effektive diske. Mere hastighed til hastighedsguden! (og forresten â denne mulighed vil vĂŠre meget praktisk i fremtiden ).
Mange administrative kommandoer arbejder nu hurtigere pÄ grund af punktcache af metadata (for eksempel, zfs list, zfs get).
Dataallokeringsprocessen er paralleliseret, nu oprettes der flere allokeringsenheder for hver sektion af frit plads (metaslab). Selv om det ikke vil udnytte NVME til fulde, vil det blive bedre.
Forsinket integritetsgenopretning af arrayet vil ikke belaste arrayet med samtidig rekonstruktion af flere diske, men vil gÞre det sekventielt. Dette vil bÄde reducere indflydelsen pÄ ydeevnen og tiden til rekonstruktion.
Ved import af puljer med et stort antal volumes er hastigheden for deres registrering i systemet Ăžget.
Derudover QAT tillader nu aflastning af krypterings- og checksumberegning.
Og til dessert â intriger
Selvom ZFSonLinux hurtigt tilfĂžjer support til friske Linux-kerner (i Ăžjeblikket understĂžttes versioner 2.6.32 â 5.1*), viser kernelmaintainere Ă„benlyst ikke interesse i at hjĂŠlpe tredjepartsmoduler (« â greg k-h»). SĂ„ de nĂždvendige systemopkald til effektiv drift i version 5.0 . I kerner med denne patch vil ydeevnen af ZFS vĂŠre betydeligt dĂ„rligere. Det redder, at denne funktionalitet kan implementeres fra modul-siden, hvilket sandsynligvis vil blive gjort. Indtil da kan du tage et eksempel fra NixOS â de đ
Projektet har ogsĂ„ fĂ„et en Code of Conduct, hvilket udlĂžste en bĂžlge af diskussioner. Men vi stod imod đ
Ănsker alle en stabil backup og stabile versioner!
Nyttige links:
â
â
Kilde: habr.com
