Btrfs RAID 5/6 Code Found To Be Very Unsafe & Will Likely Require A Rewrite

[starshipeleven]

Why would you think Reiser is insane? Something to do with the filesystem or just the fact (beyond reasonable doubt...) he killed his wife?

If someone cannot understand that to get a filesystem mainlined you need to be nice to mantainers AND also obey to their directives on code quality standards for mainline (just like everyone else), but only answers by trolling and calling "political issue" the whole matter, also someone that states a filesystem is "stable" at random when it clearly is not, yes I question his sanity.

Then there is the whole stuff he did at the trial, he acted like a person with issues (not unlike what he did with kernel mailing list).

Then at the end there is also the fact that he is a person that can't even fucking kill another person properly (i.e. without getting caught), which is another obvious evidence of lack of self-control and rational thought.

[Brane215]

Then at the end there is also the fact that he is a person that can't even fucking kill another person properly (i.e. without getting caught), which is another obvious evidence of lack of self-control and rational thought.

Reiser kept a journal. It was his nature.

[Brane215]

Then at the end there is also the fact that he is a person that can't even fucking kill another person properly (i.e. without getting caught), which is another obvious evidence of lack of self-control and rational thought.

His wife was not just another person. The fact that she was banging his friend was probably not insignificant etc. Not many people act rationally when it comes to family violence.
At some point he couldn't take that sh*t any more and there is always some fan nearby, so it was "kill -9" time.

[starshipeleven]

Reiser kept a journal. It was his nature.

which is another obvious evidence of lack of self-control and rational thought.

At some point he couldn't take that sh*t any more and there is always some fan nearby, so it was "kill -9" time.

which is another obvious evidence of lack of self-control and rational thought.

Seriously, you should try harder.

[justmy2cents]

only btrfs raid 0 and 1 is meant for use and it is stable and not facing rewrite

and i only use 5

[Brane215]

which is another obvious evidence of lack of self-control and rational thought.
which is another obvious evidence of lack of self-control and rational thought.

Seriously, you should try harder.

No, it was main selling point. At the time, ext2/3 had no journal.

[Zan Lynx]

could someone explain to me why there even is work on brtfs raid when there is software raid that is better to manage than hardware raid.

Linux software RAID can be good but too many distributions set it up completely incompetently.

For one, for software RAID-5/6 to be safe at all it requires a write-ahead journal which has to be enabled when setting it up. Otherwise your 5 drive RAID-5 array can end up with 2 disks with new data and 3 with old data, and without the RAID journal the array has to either discard the entire stripe or rebuild it corrupted. With pieces of the stripe that the database or whatever didn't even know it was writing and can't recover. See https://code.facebook.com/posts/4147...ite-ahead-log/

Software RAID also requires monitoring scripts that will inform the administrator when a drive fails. It needs decent scripts to assist an admin in repairing a failed MD array. The system initramfs needs to have support built in to boot and recover a failed MD root. It needs cron jobs to run patrol reads to look for failed blocks before they have time to stack up and create massive failure cascades. Ideally the distribution would set up SMART, including monitoring and scheduled full test runs.

BTRFS and ZFS improve on RAID a lot by including checksums so the filesystem knows when an update is complete and correct. And Copy on Write with generation counts lets it know during recovery which pieces to use by finding the highest generation with the correct checksums.

[Brane215]

+Zan lynx:

Agree on everything else, but I'm not so sure in that solution that was linked, with log on SSD. This means constant wear & tear of SSD.
Optimal solution for home and SOHO, at least for my taste, would be a MB or two of reserved kernel space in RAM, reserved for just such purpose, ofcourse with UPS on which one can rely. So user could then chose to either forego bitmap alltogether and risk whole array rebuild if that protection croaks or use bitmap and accept extra r/w.

[Zan Lynx]

+Zan lynx:

Agree on everything else, but I'm not so sure in that solution that was linked, with log on SSD. This means constant wear & tear of SSD

If it wears out after you write two or three hundred terabytes to it, get a new one. SSDs are like $50 now.

[starshipeleven]

No, it was main selling point. At the time, ext2/3 had no journal.

ext3 arrived at nearly the same time as reiserfs and had metadata journaling. Journaling was the main feature of ext3 over ext2.