Java 2 Ada

Solving Linux system lock up when intensive disk I/O are performed

By stephane.carrez

When a system lock up occurs, we often blame applications but when you look carefully you may see that despite your multi-core CPU, your applications are sleeping! No cpu activity! So what happens then? Check the I/Os, it could be the root cause!

With Ubuntu 10.04, my desktop computer was freezing when the ReadyNAS Bacula backup was running. Indeed, the Bacula daemon was performing intensive disk operations (on a fast SATA hard disk). The situation was such that it was impossible to use the system, the interface was freezing for a several seconds then working for a few seconds and freezing again.

Linux I/O Scheduler

The I/O scheduler is responsible for organizing the order in which disk operations are performed. Some algorithms allow to minimize the disk head moves, other algorithms tend to anticipate read operations,

When I/O operations are not scheduled correctly, an interactive application such as a desktop or a browser can be blocked until its I/O operations are scheduled and executed (the situation can be even worse for those applications that use the O_SYNC writing mode).

By default, the Linux kernel is configured to use the Completely Fair Queuing scheduler. This I/O scheduler does not provide any time guarantee but it gives in general good performances. Linux provides other I/O schedulers such as the Noop scheduler, the Anticipatory scheduler and the Deadline scheduler.

The deadline scheduler puts an execution time limit to requests to make sure the I/O operation is executed before an expiration time. Typically, a read operation will wait at most 500 ms. This is the I/O scheduler we need to avoid the system lock up.

Checking the I/O Scheduler

To check which I/O scheduler you are using, you can use the following command:

$ cat /sys/block/sda/queue/scheduler
noop anticipatory deadline [cfq]

where sda is the device name of your hard disk (or try hda).

The result indicates the list of supported I/O scheduler as well as the current scheduler used (here the Completely Fair Queuing).

Changing the I/O Scheduler

To change the scheduler, you can echo the desired scheduler name to activate it (you must be root):

# echo deadline >  /sys/block/sda/queue/scheduler

To make sure the I/O scheduler is configured after each system startup, you can add the following lines to your /etc/rc.local startup script:

test -f /sys/block/sda/queue/scheduler &&
  echo deadline > /sys/block/sda/queue/scheduler

test -f /sys/block/sdb/queue/scheduler &&
   echo deadline > /sys/block/sdb/queue/scheduler

test -f /sys/block/hda/queue/scheduler &&
   echo deadline > /sys/block/hda/queue/scheduler

You may have to change the sda and sdb into hda and hdb if you have an IDE hard disk.

Conclusion

After changing the I/O scheduler to use the Deadline scheduler, the desktop was not freezing any more when backups are running.

To add a comment, you must be connected. Login to add a comment

One year of data backup with Bacula on a ReadyNAS duo

By stephane.carrez

After one year of daily and weekly backup using Bacula on a ReadyNAS duo, I wanted to share information about this success story. Bacula is a network backup solution that I installed on a ReadyNAS duo. Bacula allows to make full as well as incremental backups of remote machines. It uses a MySQL database that also runs on the ReadyNAS (see Installing Mysql server on a ReadyNAS duo) and it stores backups on media such as tapes, CDs, DVDs or files.

Backup Architecture

The Bacula software is running directly on the ReadyNAS duo. The backup is configured to backup my desktop which is accessed locally, and it also backups a server running on the Internet (vacs.fr). Since the ReadyNAS is behind my Livebox, it connects to the Internet server by using a secure tunnel with OpenVPN.

Network Backup with Bacula on a ReadyNAS

The ReadyNAS duo has two 1To hard disks configured as RAID 1 mirrors.

  • Bacula director and bacular storage daemons are running on the ReadyNAS duo
  • Bacula client is running on each machine that must be backed up (Desktop and Remote Server).

Backup Pools and Strategy

Bacula is configured to create backups on file tapes. Each tape is a flat file stored on the ReadyNAS duo in some directory. I've configured file tapes so that they do not extend 4.3G (so that copying and burning DVDs could be possible).

File tapes are grouped in several pools. Each pool represent a class of backup. My primary backup strategy is split in 3 backup grades:

A-Grade backups represent critical files that must not be lost at all. They represent the files that I really care and for which I want to have one year of backup. The retention policy is set to one year with one full backup per month. In short, it means I can restore the data I had anytime during the last year. Basically it contains my full desktop home directory as well as specific directories (private photos and so on).

B-Grade backups represent less critical files for which I may not need to restore an old version. The retention policy is 180 days. This backup grade is used for software or files that I download from Internet.

C-Grade backups have a 65-days retention policy and they are used for the system. Basically, re-installation of a server or desktop from scratch is always possible but keeping the configuration files in the backup is very helpful.

A Pool is defined for each of these grades:

# A-Grade pool: 1 year retention, 12 full backups (1 full bkp/month)
Pool {
  Name = A-Full-Pool
  Pool Type = Backup
  # Bacula can automatically recycle Volumes
  Recycle = yes 
  AutoPrune = yes # Prune expired volumes
  Volume Retention = 360 days
  Label Format=A-Full-
  # 100 volumes of 4G (expecting 8 volumes/full backup)
  Maximum Volumes=100
}

# B-Grade pool: 6 months retention, 3 full backups (1 full bkp/2 months)
Pool {
  Name = B-Full-Pool
  Pool Type = Backup
  Recycle = yes
  AutoPrune = yes # Prune expired volumes
  Volume Retention = 180 days
  Label Format=B-Full-
  Maximum Volumes=40
}

# C-Grade pool: 2 months retention, 2 full backups (1 full bkp/45 day)
Pool {
  Name = C-Full-Pool
  Pool Type = Backup
  Recycle = yes
  AutoPrune = yes
  Volume Retention = 65 days # 2 months
  Label Format=C-Full-
  # 5 volumes of 4G (expecting 2 volumes/full backup)
  Maximum Volumes=5
}

In addition to these pools, an incremental and a differential pool must be defined.

Bacula FileSet

The Bacula FileSet represent the file patterns that have to be backed up. I have defined one FileSet for each machine and backup grade combination. The filesets are compressed. Files matching some patterns are excluded (

  • .o, *.log, *.bak, *~). The FileSet below is for my desktop and for the A-Grade backup. Directories /home, /data and /photos will be taken into account in the backup.
# List of files to be backed up
FileSet {
  Name = "Zebulon A-Grade"     
  Include {                    
  Options {                  
      signature=SHA1           
      compression=GZIP         
      verify = pins1           
      onefs = yes              
      WildFile = "*~"          
      WildFile = "*.bak"       
      WildFile = "*.log"       
      WildFile = "*.o"         
      Exclude = yes            
    }
    File = /home
    File = /data
    File = /photos
  }
}

Other FileSets are defined for the same machine but for different files. They will be used for other backup grades.

Backup Schedule

The schedule defines when the backup has to be executed. Each backup grade has its own schedule. This allows to run B-Grade and C-Grade backups less frequently than A-Grade.

The A-Grade backups have a full backup schedule the first Saturday of each month. A full backup of the desktop takes arround 5 hours and uses 57Go (compressed). A differential backup takes arround 2 hours and uses 10Go (compressed). The incremental backup uses 2-4Go (compressed) and 5 to 15 minutes. (these numbers depend on what is being backed up). The schedule hours are defined according to this.

Schedule {
  Name = "Weekly-A-Grade"
  Run = Full 1st sat at 23:05
  Run = Differential 2nd-5th sun at 22:10
  Run = Incremental sun-fri at 22:10
}
Schedule {
  Name = "Weekly-B-Grade"
  Run = Full jan 1st sat at 23:05
  Run = Full mar 1st sat at 23:05
  Run = Full may 1st sat at 23:05
  Run = Full jul 1st sat at 23:05
  Run = Full sep 1st sat at 23:05
  Run = Full nov 1st sat at 23:05
  Run = Differential 2nd-5th sun at 22:10
  Run = Incremental wed at 22:10
}
Schedule {
  Name = "Weekly-C-Grade"
  Run = Full jan 1st sat at 2:05
  Run = Full mar 1st sat at 2:05
  Run = Full may 1st sat at 2:05
  Run = Full jul 1st sat at 2:05
  Run = Full sep 1st sat at 2:05
  Run = Full nov 1st sat at 2:05
  Run = Differential 2nd-5th sat at 2:10
  Run = Incremental sat at 2:10
}

Bacula Job

The Bacula Job describes what must be backed up (FileSets), when (Schedule) and where (Pools). There is one job definition for each fileset.

Job {
  Name = "Zebulon-A"
  Type = Backup
  Client = zebulon-fd
  FileSet = "Zebulon A-Grade"
  Schedule = "Weekly-A-Grade"
  Storage = File
  Messages = Standard
  Pool = Default
  Full Backup Pool = A-Full-Pool
  Incremental Backup Pool = Incr-Pool
  Differential Backup Pool = Diff-Pool
  Priority = 8
}

Some Statistics

After more than one year of backups, the total storage space used is now 599G, each tape is 4.3G. The storage space used by file pools is as follows:

A Grade Full Tapes  73  313Go
B Grade Full Tapes  28  120Go
C Grade Full Tapes   4   17Go
Differential tapes  22   94Go
Incremental tapes   13   55Go

The MySQL database has grown a lot and is quite large. The InnoDB database file only contains the bacula database and it has grown up to 2Go now. The filename table references 885527 records and the path table references 546784 rows.

Conclusion

Bacula is not easy to configure but when you do it right it provides a performant backup solution. To learn more about the configuration, have a look at Bacula Documentation. Installed on a ReadyNAS duo, it proved to be a robust solution for a backup of a small set of machines. You cannot expect big performances during backup or restore. The performance bottleneck is the MySQL database which runs on the ReadyNAS.

Restoring files from the backup is quite easy but this is another story...

To add a comment, you must be connected. Login to add a comment

Restoring a complete system after a hard disk failure: bacula to the rescue!!!

By stephane.carrez

Step 1: Boot on your Ubuntu 8.04 CD

Since the disk that crashed contained the system, my computer was not even able to boot. A first step for me was to boot on the Ubuntu CDrom without installing Ubuntu again. After booting I was able to check my other disks, look at the kernel logs to realize that the disk was really completely dead without any hope to recover anything. By looking at my second hard disk, I was able to evaluate what was lost and needed to be recovered. If you have no other disk, you have to setup a new disk to proceed. Booting on the CD also helped me discover some room on my second disk where I would install a new system.

Step 2: Install the system

If the system has gone, you may have to re-install it from scratch. This is what I had to do. Having found an old debian partition on my second hard disk, I decided to install Ubuntu 8.0.4 Desktop on it. After 15 minutes, my computer was working again, running Ubuntu 8.0.4 as before. Still, my data were lost.

Step 3: Restore with bacula

Bacula is a great network backup solution that I put in place 2 years ago. Every night my bacula server is creating an incremental, differential or full backup of my computer (zebulon). It is the first time thought that I had to recover a full content. For the recovery, you have to use the Bacula Console and use the restore command.

ciceron $ bconsole

Every action made in bacula creates a job that is recorded in the database. The first thing is to identify those jobs that did the full, differential and incremental backups.

  • list jobs
 | JobId | Name      | StartTime           | Type | Level | JobFiles  | JobBytes       | JobStatus |
 |   877 | Zebulon   | 2007-12-02 02:22:27 | B    | F     | 1,245,258 | 31,026,036,274 | T         |
 | 1,067 | Zebulon   | 2008-02-03 00:52:18 | B    | F     |         0 |              0 | f         |
 | 1,319 | Zebulon   | 2008-04-26 22:28:29 | B    | D     |   207,801 |  6,048,511,830 | T         |
 | 1,328 | Zebulon   | 2008-04-29 22:17:04 | B    | I     |         0 |              0 | E         |
 | 1,331 | Zebulon   | 2008-04-30 22:17:04 | B    | I     |     1,025 |    761,323,545 | T         |
 | 1,511 | Zebulon   | 2008-06-29 22:47:57 | B    | I     |    77,997 |  9,050,108,256 | T         |
 | 1,514 | Zebulon   | 2008-06-30 22:16:40 | B    | I     |       968 |    613,957,318 | T         |
 | 1,517 | Zebulon   | 2008-07-01 22:16:38 | B    | I     |    16,710 |    866,232,575 | T         |
 | 1,520 | Zebulon   | 2008-07-02 22:17:00 | B    | I     |    11,530 |    887,021,057 | T         |

In result above is just an extract of the list command. Job 877 is a full backup (level F) and I had no other recent full backups than this one. It must be restored first. Since bacula has pruned the files, it has lost all the information about its contain (my backup could have been improved). Anyway, it is possible to restore completely this full backup. Jobs 1067 and 1328 cannot be used because they were in errors (I had many of them because the computer is off when the daily backup is started or for some other reasons). This is not a problem, bacula just ignores those jobs for the restore. To restore the full backup use the restore command:

  * __restore__
 
  First you select one or more JobIds that contain files
  to be restored. You will be presented several methods
  of specifying the JobIds. Then you will be allowed to
  select which files from those JobIds are to be restored.

After this, the bacula restore command prompts for a restore method. You can restore a files selectively, find files or restore a complete job or complete client. For me, I had to restore the full backup (job 877) so I selected the Enter list of comma separated JobIds to select method with my full backup job id:

 To select the JobIds, you have the following choices:
   1: List last 20 Jobs run
   2: List Jobs where a given File is saved
   3: Enter list of comma separated JobIds to select
   4: Enter SQL list command
   5: Select the most recent backup for a client
   6: Select backup for a client before a specified time
   7: Enter a list of files to restore
   8: Enter a list of files to restore before a specified time
   9: Find the JobIds of the most recent backup for a client
  10: Find the JobIds for a backup for a client before a specified time
  11: Enter a list of directories to restore for found JobIds
  12: Cancel
 Select item:  (1-12): __3__
   Enter JobId(s), comma separated, to restore: __877__
   You have selected the following JobId: 877
   
   Building directory tree for JobId 877 ...
   There were no files inserted into the tree, so file selection
   is not possible.Most likely your retention policy pruned the files
   
   Do you want to restore all the files? (yes|no):     __yes__

After this step, bacula searches which volumes (backup files, DVD, tapes) contain the backup:

   Bootstrap records written to /var/lib/bacula/janus-dir.restore.12.bsr
   
   The job will require the following
     Volume(s)            Storage(s)                SD Device(s)
   ===================================================
   
     Full-0013            File                      FileStorage
     Full-0014            File                      FileStorage
     Full-0015            File                      FileStorage
     Full-0016            File                      FileStorage
     Full-0017            File                      FileStorage
     Full-0035            File                      FileStorage
     Full-0036            File                      FileStorage
     Full-0037            File                      FileStorage
   
   1,245,258 files selected to be restored.

Now, I had to choose the client for the restore. For some reasons, I had to choose my crashed computer (zebulon):

   Defined Clients:
       1: janus-fd
       2: zebulon-fd
   Select the Client (1-2):    __2__

Bacula describes the restore job and you have a chance to change some parameters. In general, the restore process is made by the bacula daemon on the computer that you want to restore (ie, the client). This is natural, your computer X crashed and you want to recover on it. In my case, I wanted to recover on bacula server (called janus).

    Run Restore job
    JobName:         RestoreFiles
    Bootstrap:       /var/lib/bacula/janus-dir.restore.13.bsr
    Where:           /tmp/bacula-restores
    Replace:         always
    FileSet:         Janus Files
    Backup Client:   zebulon-fd
    Restore Client:  zebulon-fd
    Storage:         File
    When:            2008-07-05 14:16:28
    Catalog:         MyCatalog
    Priority:        10
    OK to run? (yes/mod/no): __mod__
    Parameters to modify:
     1: Level
     2: Storage
     3: Job
     4: FileSet
     5: Restore Client
     6: When
     7: Priority
     8: Bootstrap
     9: Where
    10: File Relocation
    11: Replace
    12: JobId
    Select parameter to modify (1-12): __5__
    The defined Client resources are:
     1: janus-fd
     2: zebulon-fd
   Select Client (File daemon) resource (1-2):__ 1__
    Run Restore job
    JobName:         RestoreFiles 
    Bootstrap:       /var/lib/bacula/janus-dir.restore.13.bsr
    Where:           /tmp/bacula-restores
    Replace:         always
    FileSet:         Janus Files
    Backup Client:   zebulon-fd
    Restore Client:  janus-fd
    Storage:         File
    When:            2008-07-05 14:16:28
    Catalog:         MyCatalog
    Priority:        10
    OK to run? (yes/mod/no):  __yes__

The restore process runs in background and a message and an email are sent after the restore job has finished. In my case, the files were restored on my bacula server in a /tmp/bacula-restores directory. When the restore process finished, that directory contained all my files.... back in December 2007. The differential backup was restored in the same say because the files were pruned too. Other jobs were restored as follows, using the same restore command:

    * __restore__
   
    First you select one or more JobIds that contain files
    to be restored. You will be presented several methods
    of specifying the JobIds. Then you will be allowed to
    select which files from those JobIds are to be restored.
   
    To select the JobIds, you have the following choices:
     1: List last 20 Jobs run
     2: List Jobs where a given File is saved
     3: Enter list of comma separated JobIds to select
     4: Enter SQL list command
     5: Select the most recent backup for a client
     6: Select backup for a client before a specified time
     7: Enter a list of files to restore
     8: Enter a list of files to restore before a specified time
     9: Find the JobIds of the most recent backup for a client
    10: Find the JobIds for a backup for a client before a specified time
    11: Enter a list of directories to restore for found JobIds
    12: Cancel
    Select item:  (1-12):__ 3__
    Enter JobId(s), comma separated, to restore: __1331,1511,1514,1517,1520__
    You have selected the following JobIds: 1331,1511,1514,1517,1520
   
    Building directory tree for JobId 1331 ...
    Building directory tree for JobId 1511 ...  +++++++++++++++++++++++++++++++++
    Building directory tree for JobId 1517 ...  +++++++++++++++++++++++++ 
    Building directory tree for JobId 1520 ...  +++++++++++++++++++++++++++++
    5 Jobs, 75,552 files inserted into the tree.
   
    You are now entering file selection mode where you add (mark) and
    remove (unmark) files to be restored. No files are initially added, unless
    you used the "all" keyword on the command line.
    Enter "done" to leave this mode.
   
    cwd is: /
    $ __mark *__
    79,536 files marked.
    $ __done__
    Bootstrap records written to /var/lib/bacula/janus-dir.restore.14.bsr
   
    The job will require the following
   Volume(s)            Storage(s)                SD Device(s)
    ======================================================
   
   Incr-0002            File                      FileStorage
   Incr-0005            File                      FileStorage
   Incr-0001            File                      FileStorage
   Incr-0006            File                      FileStorage
   
   79,536 files selected to be restored.

After the restore jobs finished, all my files were restored back to July 2nd 2008.

Lesson learned and conclusion

  1. Backup is vital in computer world. You don't want to loose your photos, emails and documents. When you loose one of them, you just cry. When you loose everything, you....die.
  2. My bacula configuration is not perfect. In particular it should do a full backup every 3 or 6 months. In the past I only used some file recovery but I've never tested a full recovery. This was an error (without bad consequences hopefully). Every change in bacula configuration must be followed by a full recovery test.
  3. The system partitions (/ and /usr) were not backup. Even if we can restore them with an installation, this may not be a good idea. You loose the configuration files and the knowledge of all the packages you have installed. Loosing this is not a big deal but it is a matter of time.
  4. It is necessary to test on a regular basis that we can recover from the backup. The problem is absolutely not the software itself. The problem is the backup configuration and backup needs that change over the time.

I am very thankful to the Bacula development team for their software. It is really a professional backup solution. I knew that for sure but now I can say I tested it in real situation. The hard disk failure only costs me time: time to install, time to recover the backup and time to write this story....

  • Page 1