From Fedora Project Wiki
 
Line 10: Line 10:
== Current status ==
== Current status ==
* Targeted release: [[Releases/12 | Fedora 12 ]]
* Targeted release: [[Releases/12 | Fedora 12 ]]
* Last updated: 8/06/2009
* Last updated: 9/29/2009
* Percentage of completion: 90%
* Percentage of completion: 100%
** I've have had problems with the stability of the rawhide kernel and some clvmd problems activating volumes. So, I haven't been able to adequately test clustered-samba features. I have seen gfs2 hang with a failover test and I'm debugging this right now. Testing/stabilization work ongoing
* I've been able to test the following:
** Simultaneous samba exports of the same share from multiple cluster nodes (active/active samba)
** Export the shares through a public IP address and mount from a client node using this public IP address
** Run successful smbtorture nbench tests from a client
** Active failover of samba service from one node (when killed) to another. I've _not_ tested this in f12 as I encountered a gfs2 recovery hang (it's a gfs2 bug I'm working on, has nothing to do with clustered samba), but I have seen this work with RHEL5 testing of gfs2-clustered-samba and should work in f12 as well.


== Detailed Description ==
== Detailed Description ==

Latest revision as of 14:52, 29 September 2009

Clustered Samba for GFS2

Summary

Support for clustered samba (including active/active configuration) over GFS2 using CTDB.

Owner

Current status

  • Targeted release: Fedora 12
  • Last updated: 9/29/2009
  • Percentage of completion: 100%
  • I've been able to test the following:
    • Simultaneous samba exports of the same share from multiple cluster nodes (active/active samba)
    • Export the shares through a public IP address and mount from a client node using this public IP address
    • Run successful smbtorture nbench tests from a client
    • Active failover of samba service from one node (when killed) to another. I've _not_ tested this in f12 as I encountered a gfs2 recovery hang (it's a gfs2 bug I'm working on, has nothing to do with clustered samba), but I have seen this work with RHEL5 testing of gfs2-clustered-samba and should work in f12 as well.

Detailed Description

With the availability of a clustered database (CTDB) for Samba, multiple instances of smbd across different nodes of a cluster are able to share state. It is now possible to export a gfs2 filesystem through one or more nodes in the cluster in a active/passive or active/active configuration to provide a highly-available Samba service.

Benefit to Fedora

Samba was not cluster-aware until the introduction of CTDB and hence it was not possible to have a highly available samba service. Particularly, active/active configuration was not possible. With this feature, it will be possible to run highly available samba service(s) with different configurations (active/active, active/passive) over a shared gfs2 filesystem on the Red Hat cluster infrastructure.

Scope

  • Verify simple functionality by exporting samba shares over gfs2 and running smbtorture tests (in progress)
  • fcntl posix locks performance: Improve upon gfs2's posix lock performance (single-node & clustered) as smbd uses them extensively when posix locking is turned on. (in progress)
  • Get CTDB and CTDB-supported Samba packages into F12. (complete)
  • Get Conga interface ready for manipulating the various config files (/etc/cluster/cluster.conf, /etc/sysconfig/ctdb, /etc/samba/smb.conf) through web-based GUI. (Alternative to hand-editing configs) (in progress)
  • Verify working of clustered samba use cases such as service failover, recovery, etc., and document respective configuration parameters. (in progress)

This feature will not affect the operation of non-clustered samba setups or other gfs2-based setups. It will provide the capability to setup clustered samba over gfs2.

How To Test

Hardware Requirements

  • cluster nodes (2 or more) connected to shared storage
  • client node(s) to run client loads from.

Packages

  • kernel with gfs2 and dlm enabled
  • Red Hat cluster packages cman/openais/corosync/lvm2-cluster etc. See usage.txt for more information.
  • Latest samba and ctdb packages.

Setup

  • Setup the cluster.conf file and start up the cluster 'service cman start'
  • If you're using clustered lvm, 'service clvmd start'
  • mount a gfs2-formatted clustered volume at the same mountpoint on all nodes.
  • Export this mountpoint (or a directory within it) as a samba share. All smb.conf files in all nodes should be identical
  • Set ctdb configuration. See ctdb setup
  • start the ctdb service and check 'ctdb status' to make sure the ctdb cluster is up.
  • Now you should simultaneously be able to access the samba share through all the exporting nodes and run operations on them simultaneously.

Tests

  • The smbtorture test suite in the samba package has a few scenarios RAW_BENCH_LOCK, RAW_BENCH_OPEN, BENCH_NBENCH etc. See testing ctdb
  • There's also the ping_pong test at: ping_pong doc
  • Any other test you can think of with multiple clients (windows users mounting the samba share) that stresses ctdb/gfs2.

User Experience

Users not using gfs2, samba or ctdb should notice no change. Users will now have the ability to run a clustered samba setup exporting gfs2 shares.

Dependencies

Most of the kernel support (gfs2 filesystem) is already upstream. ctdb and samba packages have been included in Fedora.

Contingency Plan

In it's current state, a basic clustered samba setup should work. Enhancements such as posix lock performance & conga interface and testing of different use-cases would be nice to get done with. If not done with these tasks, in it's current form, the feature is still acceptable.

Documentation

Release Notes

  • Fedora now has support for clustered samba over gfs2

Comments and Discussion