Skip to content

Latest commit

 

History

History
148 lines (103 loc) · 5.79 KB

Maintenance.md

File metadata and controls

148 lines (103 loc) · 5.79 KB
Raw

Common MogileFS Maintenance

  • Running a Filesystem Check - See the FSCK guide.
  • Rebalancing Files - See the Rebalance guide.
  • Managing Tracker Jobs - See the Jobs guide.

Managing Hosts

Adding Hosts

Adding hosts is straightforward via mogadm host add

$ mogadm host add mystorage --ip=10.0.0.1

Note that it is highly recommended (and will become required) that you specify the IP address of the host when adding it. The IP address mogilefs uses is what your webservers and trackers will use to talk to the storage nodes.

It's common for people to alias "mystorage" to "127.0.0.1" via a hosts file, which will cause mogilefs to issue paths for "127.0.0.1", unless you specify an IP address.

NOTE When adding new hosts to a cluster, it could take a minute or two for them to be recognized as caches expire.

Taking Hosts Down Temporarily

If you need to do maintenance on a host (upgrade RAM, replace OS, etc) that involves shutting it off, it's recommended you first mark it as "down". While MogileFS is built to be resilient to accidental failures, there's nothing wrong in giving it a heads up before taking a host offline.

$ mogadm host mark mystorage down
... do work....
$ mogadm host mark mystorage alive

Taking Hosts Down Permanently

If you mark a host as 'dead', it is not recommended to ever mark it as 'alive' or 'down' again. You must only ever mark things as "dead" which you don't expect to ever bring back to life again.

If you mark a host as 'dead', mogilefs will not re-replicate files on that host. You must mark the individual drives as 'dead'.

$ mogadm host mark mystorage dead

Managing Devices

Numbering Devices

New devices should always have incremental devids. When you add new disks, give them new, unique numbers.

If you replace a dead disk with a new disk, always give it a new devid. never reuse the old devid. This is so you can replace a disk in mogilefs before all of the files from the old device have finished re-replicating. Otherwise you could end up clobbering fids uploaded to the replaced device, or never re-replicate old files. Bad! Don't do it!

Adding Devices

$ mogadm device add mystorage 5 --status=alive
... or...
$ mogadm device add mystorage 5 --status=down

The above shows adding a storage device (#5) to storage host (mystorage) either defaulting to alive or down. Good practice is to add a new set of devices as "down", then quickly review everything works and allow caches to pull in the new devices. Then mark them all as alive when you're ready for them to be used.

It's important that you not run the storage nodes as root if you can avoid it, and that the underlying directories below where the devices are stored are owned by root and not the mogstored user. If you unmount a device, MogileFS will happily write to the underlying filesystem if it still can. There's no way for it to know if the directory is a mount point or not.

Device Maintenance

MogileFS does some automatic testing of devices and hosts. Each tracker has a monitor job which will periodically probe devices and ensure it can, write, read, and delete small files on the device. In the case of a bad filesystem failure causing a filesystem to be remounted read-only, or disappearing entirely, mogile should stop trying to write new files into the device.

In the case of a temporary downtime, such as taking time to unmount the filesystem and run the OS fsck on it, you can mark the device as down and then later alive again. In the case of filesystem corruption, IO errors, etc. It's safest to immediately mark the device as dead and allow MogileFS to make new copies from still-good devices.

Temporary Maintenance

$ mogadm device mark mystorage 5 down
... later...
$ mogadm device mark mystorage 5 alive

Please avoid leaving devices or hosts marked as down for long periods of time. MogileFS might queue up deletes, replications, fsck checks, etc, waiting for the device to come back up. If a device is going away forever, you must mark it as dead. This doesn't mean there's a set time limit in minutes or hours. That all depends on how fast things change in your setup and how big you want your queues to be.

Readonly and Drain modes

If you want to freeze all files on a device the way they are, you would use readonly mode. This will stop MogileFS from putting new files on the device, but it will also prevent deletes from removing files. Instead the deletes will queue up and wait for you to mark the device as alive or drain.

$ mogadm device mark mystorage 5 readonly
$ mogadm device mark mystorage 5 drain

drain mode, as of version 2.40 and higher, tells MogileFS that no new files should be written to a device. However files may be deleted while in drain mode.

So if you have devices where you do not wish to put new files, mark them all as drain.

NOTE In earlier versions of MogileFS, drain mode would actively remove fids from a device. It has since been replaced by Rebalance.

Re-Replicating Files

If a drive dies, mogilefs will attempt to avoid the device, but it will not automatically re-replicate files which were on the device. You must manually mark it as 'dead' via mogadm. As soon as you do so, mogilefs will start to remove files that were on the device and attempt to re-replicate them.

Managing Databases

MogileFS must have all trackers point to the same database at all times. There are no exceptions! Transactions and exclusive locks are used to coordinate individual processes and trackers, so running trackers against multiple, master-master replicated databases, will cause sporadic failures.

See MySQL for more detailed information specific to MySQL.