Fedora ARM Secondary Architecture/ARM Fedora Backup Management
It is still under construction and content may change. Do not rely on this information.
Contents
Seneca CDOT ARM Project: Management Server Backup and Recovery plan
Contact info:
- Phone: X33463
- Email: fedora-arm at senecacollege dot ca
Scope
This document is intended for those who are familiar with administering the ARM Project's management servers. Having knowledge of the CDOT ARM standard operation procedures may lead to a better understanding of this document.
Purpose
The purpose of this document is to describe the process for the backup and synchronization of critical data and provide example of some possible disaster recovery scenarios.
Introduction
The Fedora ARM Project is a highly active R & D project. Because of the rapid change, the project required some specific backup strategies to be implemented on top of regular backup strategies. In the event of an emergency, this document should provide enough information to re-create a similar management platform that will keep the project running.
Background
Like all Fedora projects, the ARM Project has a small yet active global community. The build firm is located in Ontario, Canada. Infrastructure support for the build firm is provided by the Seneca Center for Development and Technology at Seneca College in Toronto. The number of available ARM builders in the firm can vary. The builders are not backed up as those are easily recreated from available images.
As for the management servers, there are five major ones that are critical to this operation. For a complete list of servers and the services those offer please consult the necessary documents. These four servers are Australia (Koji builders temporary work storage), Chile (The complete ARM repository), Hongkong (Koji hub and web interface), Iraq (Central backup management and storage) and Ireland (Koji database).
Description
The backup system has been broken into two custom scripts. First one is to create a local copy of the configurations and to back up the repository to removable drives, it is called blaze. The second script is called wildfire and it is to synchronize those local copies to different locations over the network. The servers get the copies of those scripts from the central backup management and storage server (iraq) via scp and runs blaze first and then wildfire.
Backup Script Source
The scripts are located at (any server with backup enabled):
/usr/local/bin/
Backup Directory Structure
The backups (both local copies and networked synced copies) are stored inside /archive/blaze-backup (/archive/repo-backup for the repository sync) in each servers respective directory. Example:
[root@australia ~]# tree /archive /archive ├── blaze-backup │ ├── australia │ │ ├── australia-cron-Feb-27-2012.tar.bz2 │ │ ├── australia-etc-Feb-27-2012.tar.bz2 │ │ └── australia-userlocalbin-Feb-27-2012.tar.bz2 │ ├── chile │ │ ├── chile-cron-Feb-27-2012.tar.bz2 │ │ ├── chile-etc-Feb-27-2012.tar.bz2 │ │ └── chile-userlocalbin-Feb-27-2012.tar.bz2 │ └── ireland │ ├── ireland-cron-Feb-27-2012.tar.bz2 │ ├── ireland-etc-Feb-27-2012.tar.bz2 │ ├── ireland-postgresql-Mar-27-2012.sql.bz2 │ └── ireland-userlocalbin-Feb-27-2012.tar.bz2 └── lost+found
Space Consideration and Allocation
For backing up the configurations and keeping copies of those for the previous 30 days and also allocating room for possible synced copies from other servers, minimum space requirements is 10GB. To accommodate for the synced copy of the repository that could go up to 2TB
Data Retention
- Removable drives will keep data up to 3 days
- Network synced drives/servers will keep data up to 30 days
Initial Setup
This applies to any new server that will be backed up (this process could become a script eventually)
- Create a user named "backup"
- Give that user certificate access through ssh (to and from) the other servers (in the backup group). Tutorials available here
- Also give cert access to the root of each server so that those can remote access to the "backup" user of other servers.
- Add a new lvm called archive that will be mounted as /archive on startup (add an entry to /etc/fstab). LVM tutorial available here
- Create the following directories and give "backup" write permission to those directories
/archive/blaze-backup /archive/repo-backup /var/log/blaze-backup
- Create the following file and give "backup" write permission to that file and also add the hostname for the admin server (iraq) in that file (please see offered services section for other options)
/etc/sysconfig/blaze
- In the admin host, add the hostname in the buddyhost section in blaze and wildfire with a random syncpeer. (optional, please see offered services section for more options)
- Add the following entry to /etc/aliases (use proper email format):
blaze-wildfire: fedora-arm at senecacollege dot ca
- Schedule backup as root (crontab entry). It will look something similar to this:
20 2 * * * ADMHOST=$(cat /etc/sysconfig/blaze); BKU=backup; export ADMHOST BKU; scp $BKU@$ADMHOST:/usr/local/bin/blaze \ /usr/local/bin/; chown -R $BKU:$BKU /usr/local/bin/blaze; chmod 755 /usr/local/bin/blaze; /usr/local/bin/blaze
- If the server is going to be a reposync host then follow the first three steps to create a user account called apache and give it access to and from the repohost
Scheduling
- In Chile blaze runs at 01:35 AM UTC
- In Hongkong blaze runs at 02:05 AM UTC
- In Australia blaze runs at 02:20 AM UTC
- In Ireland blaze runs at 03:35 AM UTC
- In Scotland blaze runs at 12:50 AM EDT
- In Romania blaze runs at 01:50 AM EDT
- In Iraq blaze runs at 08:30 AM EDT
Offered/Backedup Servers/Services
- Please add the service type in the appropriate row in the "Define the names of the remote machines and services" section in each script
- PostgreSQL Database Server (DBHOST)(Possibly one - Ireland)
- Repository (REPOHOST) (Possibly one - Chile)
- Server that has backup to removable drive capabilities (RBDHOST) (Chile or Ireland or Scotland)
- Central Sync Server (SYNCHOST) (Iraq, Scotland and Romania)
- Repository Backup (COPYREPOHOST) (Scotland Romania)
- Sync Peer (BUDDYHOST) (All)
- Custom Log and Mail Server (AGGREGHOST) (Iraq)
- Administrator Host (ADMHOST) (Iraq, for now)
- All related servers under blaze wildfire system (ALLSYS) (Australia Chile Hongkong Iraq Ireland Scotland Romania)
Repo Sync
- Repo sync process is done via rsync
- Repo sync via LVM snapshot is under consideration
Backup to Removable Drives
- Removable drives (2x1T or 1x2T) can be attached to the repo server or any server that has a copy of the repo
- Those servers have the capability to auto backup overnight as long as the drives are labeled as cdot-backup-nnn (where nnn could be 001..999)
- Multiple removable drives should be used and one set of drives containing the full repo copy should be always offsite (daily rotation schedule)
Offsite Backup
- Offsite backup over the network is still under consideration
Restore from Backups
This process is not automated yet. If an older copy is required for the postgresql dump or any scripts or configuration files, please copy the necessary zip file from /archive/blaze-backup/[hostname]/ location and unzip it to a new location and copy the desired file. Please follow the backup layout table provided in the Appendix. PostgreSQL Restore is done via the following command (after unzipping):
pg_restore
Conclusion
ARM is still a developing technology and expectations are that The Fedora ARM Project at Seneca CDOT will be highly active for the next few years. Depending on the level of activity the backup strategy in place may become out of date in a matter of months or it may need to be supplemented with other solutions.
Appendix
Admin Configuration
This file specifies which server is the admin host. It is available in each host at
/etc/sysconfig/blaze
Backup Layout
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Backup Location Capacity
hongkong Filesystem Type Size Mounted on /dev/mapper/hk-archive ext4 5.0G /archive
iraq Filesystem Type Size Mounted on /dev/mapper/iraq-archive ext4 50.0G /archive
ireland Filesystem Type Size Mounted on /dev/mapper/ireland-archive ext4 1.2T /archive
australia Filesystem Type Size Mounted on /dev/mapper/australia-archive ext4 9.9G /archive
chile Filesystem Type Size Mounted on /dev/mapper/chile-archive ext4 9.9G /archive
scotland Filesystem Type Size Mounted on /dev/mapper/scotland-archive ext4 2.2T /archive
romania Filesystem Type Size Mounted on /dev/mapper/yoda-archive ext4 2.0T /archive
Log Files
Available on all servers running blaze
/var/log/blaze-backup.log /var/log/wildfire-sync.log
Available on central/aggregator/admin server
/var/log/blaze-wildfire.log
.
