Slashdot Mirror
Linux 3.2 Has Been Released
diegocg writes "Linux 3.2 has been released. New features include support for Ext4 block size bigger than 4KB and up to 1MB, btrfs has added faster scrubbing, automatic backup of critical metadata and tools for manual inspection; the process scheduler has added support to set upper limits of CPU time; the desktop responsiveness in presence of heavy writes has been improved, TCP has been updated to include an algorithm which speeds up the recovery of connection after lost packets; the profiling tool 'perf top' has added support for live inspection of tasks and libraries. The Device Mapper has added support for 'thin provisioning' of storage, and a support for a new architecture has been added: Hexagon DSP processor from Qualcomm. New drivers and small improvements and fixes are also available in this release. Here's the full list of changes."
Short answer: no.
Long answer: Please! The more people exercising the code, the more bugs will be revealed, and the more confident the developers can be. But you will have to be ready for some performance regressions and data loss danger. For the brave.
I never did like the number "3.1" for some reason
There's no fsck.. So unless you're 100% sure your Linux machine never crashes and your power supply is never interrupted - don't.
Performance is still pretty bad, especially when deleting many small files. It can take minutes with BTRFS, while EXT4 does it almost instantly in comparison.
The fucking point is to encourage beta-testers. Bleeding-edge users who know what they are doing and don't care about data loss are being offered the chance to test a new and interesting filesystem and (ab)use it in ways that upstream developers had not thought of, hopefully uncovering major bugs before the thing will get marked as feature-complete and enabled by default for new installs by major distros.
Amen.
fsck's only job is to make that junk that was a filesystem look something like a filesystem again. Nothing in there about making it look like the particular filesystem you used to have. fsck is not backups. fsck will not (necessarily) get your data back. fsck may eat kittens on a bad day. What fsck does hand back to you should not be trusted and should certainly be verified.
If you think that pulling most of what was /home, /var, /srv or /opt out of lost+found is fun, just remember that corrupted directory and filenames get named after their inodes. Nothing like trying to figure out of 1234567 or 1234568 was the start of the quarterly financials report.
If you are relying upon a fsck to get your data back after a power outage, you have more faith in your filesystem than you should. It's a nice validation tool, with the caveat that a False Negative means you go back to using a damaged filesystem for more fun later, rather than now.
BUT if you have backups, please do test. Having talked to the BTRFS team directly at LINUXCON, Mr. Chacon and folks are pretty cool about getting feedback. And you can do nifty things like snapshots for backups on RAID10 or thin disks on virtual machines which don't inflate during formatting.
For many filesystems, failing a fcsk means reaching for the format tools and the last (verified) backup. You are backing up everything, right?
"You cannot have a General Will unless you have shared experiences. You cannot be fair to people you don't know."
I used to alias "rm" as "rm -i".
Then, one day, I was using someone else's computer. I used "rm" with the expectation that it would prompt me, but this person never bothered to set it up that way, and I had the fearful experience of worrying whether it was deleting too much. I hadn't been too careless that time, but it got me thinking. It's dangerous to use "rm" when I really mean "rm -i"; habits are strong things.
So I made a change that I still use. I now alias "r" as "rm -i". "r" by itself does not have default behavior on most computers. Now if I absent-mindedly type "r *.txt" on someone else's computer, I get "r: command not found" and I edit the command to say "rm -i".
I suppose I should have used "rmi" or something like that, just in case I am a guest somewhere that "r" was aliased to something crazy. In practice, it hasn't been a problem. I use more aliases than most people seem to; they seem to be content with the defaults. I seem to be the only one I know who likes one-letter aliases.
Hmm, I guess I might accidentally run the R statistics package someday?
steveha
lf(1): it's like ls(1) but sorts filenames by extension, tersely
BTRFS is journaled (the log tree), so you don't lose data due to a power failure. It just replays the journal when you mount it again.
btrfsck is only really needed to recover from something unanticipated happening. Like software bugs. The kind that you'd expect in a new filesystem. So it's absolutely not ready for prime-time until a fsck is released.
For non-critical systems it's completely usable for people who like to experiment with the new toys. I've been using it for a year on my laptop (including multiple power losses and other shenanigans) with no problems.
My experience so far:
Good:
Snapshots (and subvolumes) are a killer feature. Having hourly and daily versions of everything is wonderful. I have subvolumes for root (@) and /home (@home). If I want to roll back my entire system but keep my homedir, I can simply: "cd /snap ; mv @ @-2012.01.04-broken ; btrfs subv snap @-2012-01-02--00-40-39-apt @ ; reboot". Translated - Save a copy of my current root; create a new root from the last snapshot that was automatically created when I last ran apt-get ; then reboot (all shutdown continues in the @-2012.01.04-broken subvolume, it doesn't corrupt the new @). Killer. Feature.
Bad:
fsync is god.awful.slow. And dpkg does a whole lot of fsync. It's completely unacceptable performance, and either btrfs has to get faster or dpkg has to be a little more miserly with fsync. If dpkg could send write barriers instead of syncs it'd probably solve it, but who knows. For the time being: "apt-get install eatmydata; ln -s `which eatmydata` /usr/local/bin/aptitude ; ln -s `which eatmydata` /usr/local/bin/apt-get" . eatmydata preloads a library that overrides sync and fsync (they're simply ignored). dpkg is now screaming fast, but I run the risk of completely screwing its metadata if there's a crash or poweroff while it's doing something. So the solution is I have a /etc/apt/apt.conf.d/80-snapshot that creates a snapshot before every apt run. If something goes wrong, I just roll back the system like I mentioned under Good.
So in summary: Some good, some bad, definitely not fully baked yet, but completely usable if you're adventurous. No fsck required. Yet. Keep backups. :)