path: root/doc/man/man5/slapd-mdb.5

.TH SLAPD-MDB 5 "RELEASEDATE" "OpenLDAP LDVERSION"
.\" Copyright 2011-2022 The OpenLDAP Foundation All Rights Reserved.
.\" Copying restrictions apply.  See COPYRIGHT/LICENSE.
.\" $OpenLDAP$
.SH NAME
slapd\-mdb \- Memory-Mapped DB backend to slapd
.SH SYNOPSIS
.B ETCDIR/slapd.conf
.SH DESCRIPTION
The \fBmdb\fP backend to
.BR slapd (8)
uses OpenLDAP's Lightning Memory-Mapped DB (LMDB) library to store data.
It relies completely on the underlying operating system for memory
management and does no caching of its own. It is the recommended
primary database backend.
.LP
The \fBmdb\fP backend uses a hierarchical database layout which
supports subtree renames.
.SH CONFIGURATION
These
.B slapd.conf
options apply to the \fBmdb\fP backend.
That is, they must follow a "backend mdb" line and
come before any subsequent "backend" or "database" lines.
.TP
.BI idlexp \ <exp>
Specify a power of 2 for the maximum size of an index slot.
The default is 16, yielding a maximum slot size of 2^16 or 65536.
Once set, this option applies to every \fBmdb\fP database instance.
The specified value must be in the range of 16-30.
.LP

These
.B slapd.conf
options apply to the \fBmdb\fP backend database.
That is, they must follow a "database mdb" line and
come before any subsequent "backend" or "database" lines.
Other database options are described in the
.BR slapd.conf (5)
manual page.
.TP
.BI checkpoint \ <kbyte>\ <min>
Specify the frequency for flushing the database disk buffers.
This setting is only needed if the \fBdbnosync\fP option is used.
The checkpoint will occur if either \fI<kbyte>\fP data has been written or
\fI<min>\fP minutes have passed since the last checkpoint.
Both arguments default to zero, in which case they are ignored. When
the \fI<min>\fP argument is non-zero, an internal task will run every 
\fI<min>\fP minutes to perform the checkpoint.
Note: currently the \fI<kbyte>\fP setting is unimplemented.
.TP
.B dbnosync
Specify that on-disk database contents should not be immediately
synchronized with in memory changes.
Enabling this option may improve performance at the expense of data
security. In particular, if the operating system crashes before changes are
flushed, some number of transactions may be lost.
By default, a full data flush/sync is performed when each
transaction is committed.
.TP
.BI directory \ <directory>
Specify the directory where the LMDB files containing this database and
associated indexes live.
A separate directory must be specified for each database.
The default is
.BR LOCALSTATEDIR/openldap\-data .
.TP
\fBenvflags \fR{\fBnosync\fR,\fBnometasync\fR,\fBwritemap\fR,\fBmapasync\fR,\fBnordahead\fR}
Specify flags for finer-grained control of the LMDB library's operation.
.RS
.TP
.B nosync
This is exactly the same as the
.I dbnosync
directive.
.RE
.RS
.TP
.B nometasync
Flush the data on a commit, but skip the sync of the meta page. This mode is
slightly faster than doing a full sync, but can potentially lose the last
committed transaction if the operating system crashes. If both
.I nometasync
and
.I nosync
are set, the
.I nosync
flag takes precedence.
.RE
.RS
.TP
.B writemap
Use a writable memory map instead of just read-only. This speeds up write operations
but makes the database vulnerable to corruption in case any bugs in slapd
cause stray writes into the mmap region.
.RE
.RS
.TP
.B mapasync
When using a writable memory map and performing flushes on each commit, use an
asynchronous flush instead of a synchronous flush (the default). This option
has no effect if
.I writemap
has not been set. It also has no effect if
.I nosync
is set.
.RE
.RS
.TP
.B nordahead
Turn off file readahead. Usually the OS performs readahead on every read
request. This usually boosts read performance but can be harmful to
random access read performance if the system's memory is full and the DB
is larger than RAM. This option is not implemented on Windows.
.RE

.TP
\fBindex \fR{\fI<attrlist>\fR|\fBdefault\fR} [\fBpres\fR,\fBeq\fR,\fBapprox\fR,\fBsub\fR,\fI<special>\fR]
Specify the indexes to maintain for the given attribute (or
list of attributes).
Some attributes only support a subset of indexes.
If only an \fI<attr>\fP is given, the indices specified for \fBdefault\fR
are maintained.
Note that setting a default does not imply that all attributes will be
indexed. Also, for best performance, an
.B eq
index should always be configured for the
.B objectClass
attribute.

A number of special index parameters may be specified.
The index type
.B sub
can be decomposed into
.BR subinitial ,
.BR subany ,\ and
.B subfinal
indices.
The special type
.B nolang
may be specified to disallow use of this index by language subtypes.
The special type
.B nosubtypes
may be specified to disallow use of this index by named subtypes.
Note: changing \fBindex\fP settings in 
.BR slapd.conf (5)
requires rebuilding indices, see
.BR slapindex (8);
changing \fBindex\fP settings
dynamically by LDAPModifying "cn=config" automatically causes rebuilding
of the indices online in a background task.
.TP
.BI maxentrysize \ <bytes>
Specify the maximum size of an entry in bytes. Attempts to store
an entry larger than this size will be rejected with the error
LDAP_ADMINLIMIT_EXCEEDED. The default is 0, which is unlimited.
.TP
.BI maxreaders \ <integer>
Specify the maximum number of threads that may have concurrent read access
to the database. Tools such as slapcat count as a single thread,
in addition to threads in any active slapd processes. The
default is 126.
.TP
.BI maxsize \ <bytes>
Specify the maximum size of the database in bytes. A memory map of this
size is allocated at startup time and the database will not be allowed
to grow beyond this size. The default is 10485760 bytes. This setting
may be changed upward if the configured limit needs to be increased.

Note: It is important to set this to as large a value as possible,
(relative to anticipated growth of the actual data over time) since
growing the size later may not be practical when the system is under
heavy load.
.TP
.BI mode \ <integer>
Specify the file protection mode that newly created database 
files should have.
The default is 0600.
.TP
\fBmultival \fR{\fI<attrlist>\fR|\fBdefault\fR} \fI<integer hi>\fR,\fI<integer lo>
Specify the number of values for which a multivalued attribute is
stored in a separate table. Normally entries are stored as a single
blob inside the database. When an entry gets very large or contains
attributes with a very large number of values, modifications on that
entry may get very slow. Splitting the large attributes out to a separate
table can improve the performance of modification operations.
The threshold is specified as a pair of integers. If the number of
values exceeds the hi threshold the values will be split out. If
a modification deletes enough values to bring an attribute below
the lo threshold the values will be removed from the separate
table and merged back into the main entry blob.
The threshold can be set for a specific list of attributes, or
the default can be configured for all other attributes.
The default value for both hi and lo thresholds is UINT_MAX, which keeps
all attributes in the main blob.
.TP
.BI rtxnsize \ <entries>
Specify the maximum number of entries to process in a single read
transaction when executing a large search. Long-lived read transactions
prevent old database pages from being reused in write transactions, and
so can cause significant growth of the database file when there is
heavy write traffic. This setting causes the read transaction in
large searches to be released and reacquired after the given number
of entries has been read, to give writers the opportunity to
reclaim old database pages. The default is 10000.
.TP
.BI searchstack \ <depth>
Specify the depth of the stack used for search filter evaluation.
Search filters are evaluated on a stack to accommodate nested AND / OR
clauses. An individual stack is assigned to each server thread.
The depth of the stack determines how complex a filter can be
evaluated without requiring any additional memory allocation. Filters that
are nested deeper than the search stack depth will cause a separate
stack to be allocated for that particular search operation. These
allocations can have a major negative impact on server performance,
but specifying too much stack will also consume a great deal of memory.
Each search stack uses 512K bytes per level. The default stack depth
is 16, thus 8MB per thread is used.
.SH ACCESS CONTROL
The 
.B mdb
backend honors access control semantics as indicated in
.BR slapd.access (5).
.SH FILES
.TP
.B ETCDIR/slapd.conf
default 
.B slapd 
configuration file
.SH SEE ALSO
.BR slapd.conf (5),
.BR slapd\-config (5),
.BR slapd (8),
.BR slapadd (8),
.BR slapcat (8),
.BR slapindex (8),
.BR slapmodify (8),
OpenLDAP LMDB documentation.
.SH ACKNOWLEDGEMENTS
.so ../Project
Written by Howard Chu.