Upgrading to a New Version of Lustre

(Updated: Dec 2009)

This section provides information about supported upgrades, upgrade paths, and interoperability of nodes with different versions of Lustre™ installed. It also describes procedures for upgrading your Lustre file system to a new version of Lustre. For information about available releases, see Lustre Release Information.

Supported Upgrades
For Lustre 1.8.x, the following upgrades are supported:
 * Lustre 1.6.x (latest version) to Lustre 1.8.x (latest version).
 * Lustre 1.8.x (any minor version) to Lustre 1.8.x (latest version).

Lustre Component Interoperability
Lustre interoperability enables 1.8.x servers ("new" servers) to work with 1.6.x clients ("old" clients), 1.6.x servers ("old" servers) to work with 1.8.x clients ("new" clients), and "mixed" environments with both 1.6.x and 1.8.x servers. For example, half of each OSS failover pair could be upgraded to enable a quick reversion to 1.6 by powering down the 1.8 servers.

The table below describes the interoperability between clients, OSTs, and MDTs with different versions of Lustre installed.

Upgrading Lustre 1.6.x to Lustre 1.8.x
Two Lustre upgrade paths are supported to meet the upgrade requirements of different Lustre environments.


 * Entire file system upgrade - All servers and clients are shut down and upgraded at the same time. See Performing a Complete File System Upgrade.
 * Rolling upgrade - Individual servers (or their failover partners) and clients are upgraded one at a time and restarted, so that the file system never goes down. See Performing a Rolling Upgrade.

Note: If you upgrade some Lustre components to 1.8.x but not others (such as running 1.8 clients in a file system with 1.6 OSTs), and run a mixed environment, you may see one or more warnings similar to this:

LustreError: 3877:0:(socklnd_cb.c:2228:ksocknal_recv_hello) Unknown protocol version (2.x expected) from 192.168.2.43

This warning is given when the 1.6 and 1.8 components use different protocols. It can be safely ignored because the Lustre components negotiate a common protocol. In this example, the 1.8 clients fall back to use the 1.6 protocol with the 1.6 OSTs.

Performing a Complete File System Upgrade
This procedure describes a complete file system upgrade in which 1.8.x Lustre packages are installed on multiple 1.6.x servers and clients, requiring a file system shut down. If you want to upgrade one Lustre component at a time and avoid the shutdown, see Performing a Rolling Upgrade.

Tip: In a Lustre upgrade, the package install and file system unmount steps are reversible; you can do either step first. To minimize downtime, this procedure first performs the 1.8.x package installation, and then unmounts the file system.

1. Make a complete, restorable file system backup before upgrading Lustre.

2. Install the 1.8.x packages on the Lustre servers and/or clients. Some or all servers can be upgraded. Some or all clients can be upgraded. For help determining where to install a specific package, see Lustre Packages.


 * a. Install the kernel, modules and ldiskfs packages. For example:

$ rpm -ivh kernel-lustre-smp- \ kernel-ib- \ lustre-modules- \ lustre-ldiskfs-


 * b. Upgrade the utilities/userspace packages. For example:

$ rpm -Uvh lustre-


 * c. If a new e2fsprogs package is available, upgrade it. For example:

$ rpm -Uvh e2fsprogs-


 * There may or may not be a new e2fsprogs package with a Lustre upgrade. The e2fsprogs release schedule is independent of Lustre releases.


 * d. (Optional) If you want to add optional packages to your Lustre system, install them now.

3. Shut down the file system. Shut down the components in this order: clients, then the MDT, then OSTs. Unmounting a block device causes Lustre to be shut down on that node.


 * a. Unmount the clients. On each client node, run:

umount


 * b. Unmount the MDT. On the MDS node, run:

umount


 * c. Unmount the OSTs (be sure to unmount all OSTs). On each OSS node, run:

umount

4. Unload the old Lustre modules by either:
 * Rebooting the node
 * - OR -


 * Removing the Lustre modules manually. Run lustre_rmmod several times and use lsmod to check the currently loaded modules.

5. Start the upgraded file system. Start the components in this order: OSTs, then the MDT, then clients.
 * a. Mount the OSTs (be sure to mount all OSTs). On each OSS node, run:

mount -t lustre


 * b. Mount the MDT. On the MDS node, run:

mount -t lustre


 * c. Mount the file system on the clients. On each client node, run:

mount -t lustre &#58;/

If you have a problem upgrading Lustre, contact us by submitting a bug to our bug tracker Bugzilla.

Performing a Rolling Upgrade
This procedure describes a rolling upgrade in which one Lustre component (server or client) is upgraded and restarted at a time while the file system is running. If you want to upgrade the complete Lustre file system or multiple components at a time, requiring a file system shutdown, see Performing a Complete File System Upgrade.

Note: If the Lustre component to be upgraded is an OSS in a failover pair, follow these special upgrade steps to minimize downtime:

1. Fail over the server to its peer server, so the file system remains available.

2. Install the Lustre 1.8.x packages on the idle server.

3. Unload the old Lustre modules on the idle server by either:
 * Rebooting the node.
 * - OR -
 * Removing the Lustre modules manually by running the lustre_rmmod command several times and checking the currently loaded modules with the lsmod command.

4. Fail back services to the now upgraded server.

5. Repeat Steps 1 to 4 on the peer server. This limits the outage (per OSS) to a single server for as long as it takes to fail over.

To perform a rolling upgrade:

1. Make a complete, restorable file system backup before upgrading Lustre.

2. Install the 1.8.x packages on the Lustre component (server or client). For help determining where to install a specific package, see Lustre Packages.


 * a. Install the kernel, modules and ldiskfs packages. For example:

$ rpm -ivh kernel-lustre-smp- \ kernel-ib- \ lustre-modules- \ lustre-ldiskfs-


 * b. Upgrade the utilities/userspace packages. For example:

$ rpm -Uvh lustre-


 * c. If a new e2fsprogs package is available, upgrade it. For example:

$ rpm -Uvh e2fsprogs-

There may or may not be a new e2fsprogs package with a Lustre upgrade. The e2fsprogs release schedule is independent of Lustre releases.


 * d. (Optional) If you want to add optional packages to your Lustre system, install them now.

3. Unload the old Lustre modules by either:
 * Rebooting the node
 * - OR -
 * Removing the Lustre modules manually. Run lustre_rmmod several times and use lsmod to check the currently-loaded modules.

4. If the upgraded component is a server, fail back services to the new server.

If you have a problem upgrading Lustre, contact us via the Bugzilla bug tracker.

Upgrading Lustre 1.8.x to the Next Minor Version
To upgrade Lustre 1.8.x to the next minor version, for example, Lustre 1.8.0.1 > 1.8.x, follow these procedures:
 * To upgrade the complete file system or multiple file system components at the same time, requiring a file system shutdown, see Performing a Complete File System Upgrade
 * To upgrade one Lustre component (server or client) at a time, while the file system is running, see Performing a Rolling Upgrade

Downgrading from Lustre 1.8.x
For Lustre 1.8.x, the following downgrades are supported:
 * If you upgraded from Lustre 1.6.x > 1.8.x, you can downgrade to version 1.6.x.
 * If you upgraded from one minor version to the next (for example, Lustre 1.8.0 > 1.8.x), you can downgrade to the earlier minor version.

For a procedure to downgrade from Lustre 1.8.x to Lustre 1.6.x, see Section 13.5: Downgrading from Lustre 1.8.x to Lustre 1.6.x in the Lustre Operations Manual.

Caution: A new installation of Lustre 1.8.x is not guaranteed to be downgradable to an earlier Lustre version.