Catalog Services

General

This chapter is intended to be a technical discussion of the Catalog services and as such is not targeted at end users but rather at developers and system administrators that want or need to know more of the working details of Bacula.

The Bacula Catalog services consist of the programs that provide the SQL database engine for storage and retrieval of all information concerning files that were backed up and their locations on the storage media.

We have investigated the possibility of using the following SQL engines for Bacula: Beagle, mSQL, GNU SQL, PostgreSQL, Oracle, and MySQL. Each presents certain problems with either licensing or maturity. At present, we have chosen for development purposes to use MySQL and PostgreSQL. MySQL was chosen because it is fast, proven to be reliable, widely used, and actively being developed. MySQL is released under the GNU GPL license. PostgreSQL was chosen because it is a full-featured, very mature database, and because Dan Langille did the Bacula driver for it. PostgreSQL is distributed under the BSD license.

The Bacula SQL code has been written in a manner that will allow it to be easily modified to support any of the current SQL database systems on the market (for example: mSQL, iODBC, unixODBC, Solid, OpenLink ODBC, EasySoft ODBC, InterBase, Oracle8, Oracle7, and DB2).

If you do not specify either -with-mysql or -with-postgresq on the ./configure line, Bacula will use its minimalist internal database. This database is kept for build reasons but is no longer supported. Bacula requires one of the three databases (MySQL, PostgreSQL, or SQLite) to run.

Filenames and Maximum Filename Length

In general, either MySQL or PostgreSQL permit storing arbitrary long path names and file names in the catalog database. In practice, there still may be one or two places in the Catalog interface code that restrict the maximum path length to 512 characters and the maximum file name length to 512 characters. These restrictions are believed to have been removed. Please note, these restrictions apply only to the Catalog database and thus to your ability to list online the files saved during any job. All information received and stored by the Storage daemon (normally on tape) allows and handles arbitrarily long path and filenames.

Installing and Configuring MySQL

For the details of installing and configuring MySQL, please see the Installing and Configuring MySQL chapter of the Bacula Enterprise Main manual.

Installing and Configuring PostgreSQL

For the details of installing and configuring PostgreSQL, please see the Installing and Configuring PostgreSQL chapter of the Bacula Enterprise Main manual.

Internal Bacula Catalog

Please see the Internal Bacula Database chapter of the Bacula Enterprise Miscellaneous guide for more details.

Database Table Design

All discussions that follow pertain to the MySQL database. The details for the PostgreSQL databases are quite similar.

Because the Catalog database may contain very large amounts of data for large sites, we have made a modest attempt to normalize the data tables to reduce redundant information. While reducing the size of the database significantly, it does, unfortunately, add some complications to the structures.

In simple terms, the Catalog database must contain a record of all Jobs run by Bacula, and for each Job, it must maintain a list of all files saved, with their File Attributes (permissions, create date, ...), and the location and Media on which the file is stored. This is seemingly a simple task, but it represents a huge amount interlinked data. Note: the list of files and their attributes is not maintained when using the internal Bacula database. The data stored in the File records, which allows the user or administrator to obtain a list of all files backed up during a job, is by far the largest volume of information put into the Catalog database.

Although the Catalog database has been designed to handle backup data for multiple clients, some users may want to maintain multiple databases, one for each machine to be backed up. This reduces the risk of confusion of accidental restoring a file to the wrong machine as well as reducing the amount of data in a single database, thus increasing efficiency and reducing the impact of a lost or damaged database.

Sequence of Creation of Records for a Save Job

Start with StartDate, ClientName, Filename, Path, Attributes, MediaName, MediaCoordinates. (PartNumber, NumParts). In the steps below, “Create new” means to create a new record whether or not it is unique. “Create unique” means each record in the database should be unique. Thus, one must first search to see if the record exists, and only if not should a new one be created, otherwise the existing RecordId should be used.

1. Create new Job record with StartDate; save JobId
2. Create unique Media record; save MediaId
3. Create unique Client record; save ClientId
4. Create unique Filename record; save FilenameId
5. Create unique Path record; save PathId
6. Create unique Attribute record; save AttributeId store ClientId, FilenameId, PathId, and Attributes
7. Create new File record store JobId, AttributeId, MediaCoordinates, etc
8. Repeat steps 4 through 8 for each file
9. Create a JobMedia record; save MediaId
10. Update Job record filling in EndDate and other Job statistics

Database Tables

The Path table contains the path or directory names of all directories on the system or systems. The filename and any MSDOS disk name are stripped off. As with the filename, only one copy of each directory name is kept regardless of how many machines or drives have the same directory. These path names should be stored in Unix path name format.

Some simple testing on a Linux file system indicates that separating the filename and the path may be more complication than is warranted by the space savings. For example, this system has a total of 89,097 files, 60,467 of which have unique filenames, and there are 4,374 unique paths.

Finding all those files and doing two stats() per file takes an average wall clock time of 1 min 35 seconds on a 400MHz machine running RedHat 6.1 Linux.

Finding all those files and putting them directly into a MySQL database with the path and filename defined as TEXT, which is variable length up to 65,535 characters takes 19 mins 31 seconds and creates a 27.6 MByte database.

Doing the same thing, but inserting them into Blob fields with the filename indexed on the first 30 characters and the path name indexed on the 255 (max) characters takes 5 mins 18 seconds and creates a 5.24 MB database. Rerunning the job (with the database already created) takes about 2 mins 50 seconds.

Running the same as the last one (Path and Filename Blob), but Filename indexed on the first 30 characters and the Path on the first 50 characters (linear search done there after) takes 5 mins on the average and creates a 3.4 MB database. Rerunning with the data already in the DB takes 3 mins 35 seconds.

Finally, saving only the full path name rather than splitting the path and the file, and indexing it on the first 50 characters takes 6 mins 43 seconds and creates a 7.35 MB database.

The File table contains one entry for each file backed up by Bacula. Thus a file that is backed up multiple times (as is normal) will have multiple entries in the File table. This will probably be the table with the most number of records. Consequently, it is essential to keep the size of this record to an absolute minimum. At the same time, this table must contain all the information (or pointers to the information) about the file and where it is backed up. Since a file may be backed up many times without having changed, the path is stored in separate tables.

This table contains by far the largest amount of information in the Catalog database, both from the stand point of number of records, and the stand point of total database size. As a consequence, the user must take care to periodically reduce the number of File records using the retention command in the Console program.

The Job table contains one record for each Job run by Bacula. Thus normally, there will be one per day per machine added to the database. Note, the JobId is used to index Job records in the database, and it often is shown to the user in the Console program. However, care must be taken with its use as it is not unique from database to database. For example, the user may have a database for Client data saved on machine Rufus and another database for Client data saved on machine Roxie. In this case, the two database will each have JobIds that match those in another database. For a unique reference to a Job, see Job below.

The Name field of the Job record corresponds to the Name resource record given in the Director's configuration file. Thus it is a generic name, and it will be normal to find many Jobs (or even all Jobs) with the same Name.

The Job field contains a combination of the Name and the schedule time of the Job by the Director. Thus for a given Director, even with multiple Catalog databases, the Job will contain a unique name that represents the Job.

For a given Storage daemon, the VolSessionId and VolSessionTime form a unique identification of the Job. This will be the case even if multiple Directors are using the same Storage daemon.

The Job Type (or simply Type) can have one of the following values:

Note, the Job Type values in table (here) noted above are not kept in an SQL table.

The JobStatus field specifies how the job terminated, and can be one of the following:

The FileSet table contains one entry for each FileSet that is used. The MD5 signature is kept to ensure that if the user changes anything inside the FileSet, it will be detected and the new FileSet will be used. This is particularly important when doing an incremental update. If the user deletes a file or adds a file, we need to ensure that a Full backup is done prior to the next incremental.

The JobMedia table contains one entry at the following: start of the job, start of each new tape file, start of each new tape, end of the job. Since by default, a new tape file is written every 2GB, in general, you will have more than 2 JobMedia records per Job. The number can be varied by changing the "Maximum File Size" specified in the Device resource. This record allows Bacula to efficiently position close to (within 2GB) any given file in a backup. For restoring a full Job, these records are not very important, but if you want to retrieve a single file that was written near the end of a 100GB backup, the JobMedia records can speed it up by orders of magnitude by permitting forward spacing files and blocks rather than reading the whole 100GB backup.

The Volume tablenoteInternally referred to as the Media table contains one entry for each volume, that is each tape, cassette (8mm, DLT, DAT, ...), or file on which information is or was backed up. There is one Volume record created for each of the NumVols specified in the Pool resource record.

The Pool table contains one entry for each media pool controlled by Bacula in this database. One media record exists for each of the NumVols contained in the Pool. The PoolType is a Bacula defined keyword. The MediaType is defined by the administrator, and corresponds to the MediaType specified in the Director's Storage definition record. The CurrentVol is the sequence number of the Media record for the current volume.

The Client table contains one entry for each machine backed up by Bacula in this database. Normally the Name is a fully qualified domain name.

The Storage table contains one entry for each Storage used.

The Counter table contains one entry for each permanent counter defined by the user.

The JobHisto table is the same as the Job table, but it keeps long term statistics (i.e. it is not pruned with the Job).

The Log table contains a log of all Job output.

The Location table defines where a Volume is physically.

The Location Log table contains a log of all Job output.

The Version table defines the Bacula database version number. Bacula checks this number before reading the database to ensure that it is compatible with the Bacula binary file.

The BaseFiles table contains all the File references for a particular JobId that point to a Base file - i.e. they were previously saved and hence were not saved in the current JobId but in BaseJobId under FileId. FileIndex is the index of the file, and is used for optimization of Restore jobs to prevent the need to read the FileId record when creating the in memory tree. This record is not yet implemented.