Dear users,
LizardFS 3.10.0 is now stable.
This release consists of:
erasure code goals (aka K+M goals)
per directory quotas
improved interaction with legacy chunkservers and mounts
ports for OSX and FreeBSD
updated mfsfileinfo output with xor/ec goals
many fixes
Details:
Erasure code goals
For each file using ec(K+M) goal, the system will split the file into K parts and generate M parity parts. Any K parts are required to be able to reproduce file contents.
If labels are specified, parts will be kept on chunkservers with these labels. Otherwise, default wildcard labels will be used.
This kind of goal allows M of K+M copies to be lost and the file would still remain accessible. Erasure code goal occupies M/K extra space.
Please note that minimal K value is 2 and minimal M value is 1.
Examples of new goal configuration:
5 ec32 : $ec(3,2){ A B _ _ _ }
6 ec43 : $ec(4,3)
Per directory quotas
It is now possible to set inode/size limits per directory.
In case of multiple quota rules, the most restrictive one is always effective.
Example:
mfssetquota -d 10M 100M 50M 1G /mnt/lizardfs/a/directory
Improved interaction with legacy chunkservers and mounts
Pre-2.6.0 mounts and chunkservers used to be not fully compatible with 3.9.x servers, which complicated the upgrade process. Those issues are now solved - new mounts are perfectly capable of communicating with legacy chunkservers and vice versa.
This rule has one exception: for consistency reasons, replication from new chunkservers to old ones is disabled.
Ports for OSX and FreeBSD
LizardFS is now buildable for those systems.
Fixes
denying replication during scan
atomicity of replication process
sending proper error codes to master
proper handling of replication limits during rebalancing
removal of splice call
removing possible livelock in master networking
adding missing update in inodes id pool
Next release (unstable now) will contain a massive set of patches which make the cluster more responsive.
Also, a great reduction of master's and chunkserver's RAM usage is to be expected (at least 30% less resources needed).
An improved client tool for managing goals/quotas/files etc. is coming as well.
Best regards,
Piotr Sarna
LizardFS Team