发布于 2014-06-26 02:53:38 | 544 次阅读 | 评论: 0 | 来源: 网友投递
GlusterFS 集群文件系统
Gluster 是一个集群的文件系统，支持 PB 级的数据量。GlusterFS 通过 RDMA 和 TCP/IP 方式将分布到不同服务器上的存储空间汇集成一个大的网络并行文件系统。
分布式文件系统GlusterFS 3.5.1发布。2014-06-25 上个版本是2014-04-18的3.5.0。被Redhat收购后，GlusterFS发展还是很快的。相当简单好用的分布式文件系统。可用来做虚拟化 后端并且支持FUSE.在要求不高的情况下个人觉得是首选的分布式文件系统。
完全改进：
This is mostly a bugfix release. The Release Notes for 3.5.0contain a listing of all the new features that were added.
There are two notable changes that are not only bug fixes, or documentation additions:
a new volume optionserver.manage-gidshas been added This option should be used when users of a volume are in more than approximately 93 groups (Bug 1096425)
Duplicate Request Cache for NFS has now been disabled by default, this may reduce performance for certain workloads, but improves the overall stability and memory footprint for most users
765202: lgetxattr called with invalid keys on the bricks
833586: inodelk hang from marker_rename_release_newp_lock
859581: self-heal process can sometimes create directories instead of symlinks for the root gfid file in .glusterfs
986429: Backupvolfile server option should work internal to GlusterFS framework
1039544: [FEAT] "gluster volume heal info" should list the entries that actually required to be healed.
1046624: Unable to heal symbolic Links
1046853: AFR : For every file self-heal there are warning messages reported in glustershd.log file
1063190: Volume was not accessible after server side quorum was met
1064096: The old Python Translator code (not Glupy) should be removed
1066996: Using sanlock on a gluster mount with replica 3 (quorum-type auto) leads to a split-brain
1071191: [3.5.1] Sporadic SIGBUS with mmap() on a sparse file created with open(), seek(), write()
1078061: Need ability to heal mismatching user extended attributes without any changelogs
1078365: New xlators are linked as versioned .so files, creating .so.0.0.0
1086743: Add documentation for the Feature: RDMA-connection manager (RDMA-CM)
1086748: Add documentation for the Feature: AFR CLI enhancements
1086749: Add documentation for the Feature: Exposing Volume Capabilities
1086750: Add documentation for the Feature: File Snapshots in GlusterFS
1086751: Add documentation for the Feature: gfid-access
1086752: Add documentation for the Feature: On-Wire Compression/Decompression
1086754: Add documentation for the Feature: Quota Scalability
1086755: Add documentation for the Feature: readdir-ahead
1086756: Add documentation for the Feature: zerofill API for GlusterFS
1086758: Add documentation for the Feature: Changelog based parallel geo-replication
1086760: Add documentation for the Feature: Write Once Read Many (WORM) volume
1086762: Add documentation for the Feature: BD Xlator - Block Device translator
1086766: Add documentation for the Feature: Libgfapi
1086774: Add documentation for the Feature: Access Control List - Version 3 support for Gluster NFS
1086781: Add documentation for the Feature: Eager locking
1086782: Add documentation for the Feature: glusterfs and oVirt integration
1086783: Add documentation for the Feature: qemu 1.3 - libgfapi integration
1088848: Spelling errors in rpc/rpc-transport/rdma/src/rdma.c
1089054: gf-error-codes.h is missing from source tarball
1089470: SMB: Crash on brick process during compile kernel.
1089934: list dir with more than N files results in Input/output error
1091340: Doc: Add glfs_fini known issue to release notes 3.5
1091392: glusterfs.spec.in: minor/nit changes to sync with Fedora spec
1095256: Excessive logging from self-heal daemon, and bricks
1095595: Stick to IANA standard while allocating brick ports
1095775: Add support in libgfapi to fetch volume info from glusterd.
1095971: Stopping/Starting a Gluster volume resets ownership
1096040: AFR : self-heal-daemon not clearing the change-logs of all the sources after self-heal
1096425: i/o error when one user tries to access RHS volume over NFS with 100+ GIDs
1099878: Need support for handle based Ops to fetch/modify extended attributes of a file
1101647: gluster volume heal volname statistics heal-count not giving desired output.
1102306: license: xlators/features/glupy dual license GPLv2 and LGPLv3+
1103413: Failure in gf_log_init reopening stderr
1104592: heal info may give Success instead of transport end point not connected when a brick is down.
1104915: glusterfsd crashes while doing stress tests
1104919: Fix memory leaks in gfid-access xlator.
1104959: Dist-geo-rep : some of the files not accessible on slave after the geo-rep sync from master to slave.
1105188: Two instances each, of brick processes, glusterfs-nfs and quotad seen after glusterd restart
1105524: Disable nfs.drc by default
1107937: quota-anon-fd-nfs.t fails spuriously
1109832: I/O fails for for glusterfs 3.4 AFR clients accessing servers upgraded to glusterfs 3.5
1110777: glusterfsd OOM - using all memory when quota is enabled
The following configuration changes are necessary for qemu and samba integration with libgfapi to work seamlessly:
More details are also documented in the Gluster Wiki on the Libgfapi with qemu libvirt page.
gluster volume set <volname> server.allow-insecure on
restarting the volume is necessary
gluster volume stop gluster volume start
Edit/etc/glusterfs/glusterd.volto contain this line:
option rpc-auth-allow-insecure on
restarting glusterd is necessary
service glusterd restart
For Block Device translator based volumes open-behind translator at the client side needs to be disabled.
libgfapi clients callingglfs_finibefore a successfullglfs_initwill cause the client to hang has been reported by QEMU developers. The workaround is NOT to callglfs_finifor error cases encountered before a successfullglfs_init. Follow Bug 1091335 to get informed when a release is made available that contains a final fix.
After enablingserver.manage-gids, the volume needs to be stopped and started again to have the option enabled in the brick processes
gluster volume stop <volname> gluster volume start <volname>
下载：http://download.gluster.org/pub/gluster/glusterfs/LATEST/glusterfs-3.5.1.tar.gz