System Administration is all about maintenance, so it is only fair that we discuss how to correctly maintain an OpenLDAP deployment.
Backup strategies largely depend on the amount of change in the database and how much of that change an administrator might be willing to lose in a catastrophic failure. There are two basic methods that can be used:
1. Backup the Berkeley database itself and periodically back up the transaction log files:
Berkeley DB produces transaction logs that can be used to reconstruct changes from a given point in time. For example, if an administrator were willing to only lose one hour's worth of changes, they could take down the server in the middle of the night, copy the Berkeley database files offsite, and bring the server back online. Then, on an hourly basis, they could force a database checkpoint, capture the log files that have been generated in the past hour, and copy them offsite. The accumulated log files, in combination with the previous database backup, could be used with db_recover to reconstruct the database up to the time the last collection of log files was copied offsite. This method affords good protection, with minimal space overhead.
2. Periodically run slapcat and back up the LDIF file:
Slapcat can be run while slapd is active. However, one runs the risk of an inconsistent database- not from the point of slapd, but from the point of the applications using LDAP. For example, if a provisioning application performed tasks that consisted of several LDAP operations, and the slapcat took place concurrently with those operations, then there might be inconsistencies in the LDAP database from the point of view of that provisioning application and applications that depended on it. One must, therefore, be convinced something like that won't happen. One way to do that would be to put the database in read-only mode while performing the slapcat. The other disadvantage of this approach is that the generated LDIF files can be rather large and the accumulation of the day's backups could add up to a substantial amount of space.
You can use slapcat(8) to generate an LDIF file for each of your slapd(8) back-bdb or back-hdb databases.
slapcat -f slapd.conf -b "dc=example,dc=com"
For back-bdb and back-hdb, this command may be ran while slapd(8) is running.
MORE on actual Berkeley DB backups later covering db_recover etc.
Berkeley DB log files grow, and the administrator has to deal with it. The procedure is known as log file archival or log file rotation.
Note: The actual log file rotation is handled by the Berkeley DB engine.
Logs of current transactions need to be stored into files so that the database can be recovered in the event of an application crash. Administrators can change the size limit of a single log file (by default 10MB), and have old log files removed automatically, by setting up DB environment (see below). The reason Berkeley DB never deletes any log files by default is that the administrator may wish to backup the log files before removal to make database recovery possible even after a catastrophic failure, such as file system corruption.
Log file names are log.XXXXXXXXXX (X is a digit). By default the log files are located in the BDB backend directory. The db_archive tool knows what log files are used in current transactions, and what are not. Administrators can move unused log files to a backup media, and delete them. To have them removed automatically, place set_flags DB_LOG_AUTOREMOVE directive in DB_CONFIG.
Note: If the log files are removed automatically, recovery after a catastrophic failure is likely to be impossible.
The files with names __db.001, __db.002, etc are just shared memory regions (or whatever). These ARE NOT 'logs', they must be left alone. Don't be afraid of them, they do not grow like logs do.
To understand the db_archive interface, the reader should refer to chapter 9 of the Berkeley DB guide. In particular, the following chapters are recommended:
- Database and log file archival - http://www.oracle.com/technology/documentation/berkeley-db/db/ref/transapp/archival.html
- Log file removal - http://www.oracle.com/technology/documentation/berkeley-db/db/ref/transapp/logfile.html
- Recovery procedures - http://www.oracle.com/technology/documentation/berkeley-db/db/ref/transapp/recovery.html
- Hot failover - http://www.oracle.com/technology/documentation/berkeley-db/db/ref/transapp/hotfail.html
- Complete list of Berkeley DB flags - http://www.oracle.com/technology/documentation/berkeley-db/db/api_c/env_set_flags.html
Advanced installations can use special environment settings to fine-tune some Berkeley DB options (change the log file limit, etc). This can be done by using the DB_CONFIG file. This magic file can be created in BDB backend directory set up by slapd.conf(5). More information on this file can be found in File naming chapter. Specific directives can be found in C Interface, look for DB_ENV->set_XXXX calls.
Note: options set in DB_CONFIG file override options set by OpenLDAP. Use them with extreme caution. Do not use them unless You know what You are doing.
The advantages of DB_CONFIG usage can be the following:
- to keep data files and log files on different mediums (i.e. disks) to improve performance and/or reliability;
- to fine-tune some specific options (such as shared memory region sizes);
- to set the log file limit (please read Log file limits before doing this).
To figure out the best-practice BDB backup scenario, the reader is highly recommended to read the whole Chapter 9: Berkeley DB Transactional Data Store Applications. This chapter is a set of small pages with examples in C language. Non-programming people can skip these examples without loss of knowledge.
If you put "checkpoint 1024 5" in slapd.conf (to checkpoint after 1024kb or 5 minutes, for example), this does not checkpoint every 5 minutes as you may think. The explanation from Howard is:
'In OpenLDAP 2.1 and 2.2 the checkpoint directive acts as follows - *when there is a write operation*, and more than <check> minutes have occurred since the last checkpoint, perform the checkpoint. If more than <check> minutes pass after a write without any other write operations occurring, no checkpoint is performed, so it's possible to lose the last write that occurred.''
In other words, a write operation occurring less than "check" minutes after the last checkpoint will not be checkpointed until the next write occurs after "check" minutes have passed since the checkpoint.
This has been modified in 2.3 to indeed checkpoint every so often; in the meantime a workaround is to invoke "db_checkpoint" from a cron script every so often, say 5 minutes.
The simplest steps needed to migrate between versions or upgrade, depending on your deployment type are:
- Stop the current server when convenient
- slapcat the current data out
- Clear out the current data directory (/usr/local/var/openldap-data/) leaving DB_CONFIG in place
- Perform the software upgrades
- slapadd the exported data back into the directory
- Start the server
Obviously this doesn't cater for any complicated deployments like MirrorMode or N-Way Multi-Master, but following the above sections and using either commercial support or community support should help. Also check the Troubleshooting section.