OGR optionally supports spatial and non-spatial tables stored in SQLite 3.x database files. SQLite is a "light weight" single file based RDBMS engine with fairly complete SQL semantics and respectible performance.
The driver can handle "regular" SQLite databases, as well as Spatialite databases (spatial enabled SQLite databases).
The SQLite database is essentially typeless, but the SQLite driver will attempt to classify attributes field as text, integer or floating point based on the contents of the first record in a table. None of the list attribute field types existing in SQLite. Starting with OGR 1.10, datetime field types are also handled.
SQLite databases often due not work well over NFS, or some other networked file system protocols due to the poor support for locking. It is safest to operate only on SQLite files on a physical disk of the local system.
SQLite is an optionally compiled in driver. It is not compiled in by default.
By default, SQL statements are passed directly to the SQLite database engine. It's also possible to request the driver to handle SQL commands with OGR SQL engine, by passing "OGRSQL" string to the ExecuteSQL() method, as name of the SQL dialect.
Starting with OGR 1.8.0, the OGR_SQLITE_SYNCHRONOUS configuration option has been added. When set to OFF, this issues a 'PRAGMA synchronous = OFF' command to the SQLite database. This has the advantage of speeding-up some write operations (e.g. on EXT4 filesystems), but at the expense of data safety w.r.t system/OS crashes. So use it carefully in production environments and read the SQLite related documentation.
The driver looks for a geometry_columns table layed out as defined loosely according to OGC Simple Features standards, particularly as defined in FDO RFC 16. If found it is used to map tables to layers.
If geometry_columns is not found, each table is treated as a layer. Layers with a WKT_GEOMETRY field will be treated as spatial tables, and the WKT_GEOMETRY column will be read as Well Known Text geometry.
If geometry_columns is found, it will be used to lookup spatial reference systems in the spatial_ref_sys table.
While the SQLite driver supports reading spatial data from records, there is
no support for spatial indexing, so spatial queries will tend to be slow (use Spatialite for that).
Attributes queries may be fast, especially if indexes are built for
appropriate attribute columns using the "CREATE INDEX
Note: this support is limited to read-only operations.
The driver supports reading and writing to files managed by VSI Virtual File System API, which include "regular" files, as well as files in the /vsimem/ (read-write), /vsizip/ (read-only), /vsigzip/ (read-only), /vsicurl/ (read-only) domains.
Note: for regular files, the standard I/O operations provided by SQLite are used, in order to benefit from its integrity guarantees.
The SQLite driver can read and write SpatiaLite databases. Creating or updating a spatialite database requires explicit linking against SpatiaLite library (version >= 2.3.1). Explicit linking against SpatiaLite library also provides access to functions provided by this library, such as spatial indexes, spatial functions, etc...
A few examples :
# Duplicate the sample database provided with SpatiaLite ogr2ogr -f SQLite testspatialite.sqlite test-2.3.sqlite -dsco SPATIALITE=YES # Make a request with a spatial filter. Will work faster if spatial index has # been created and explicit linking against SpatiaLite library. ogrinfo testspatialite.sqlite Towns -spat 754000 4692000 770000 4924000
It is possible to open on-the-fly a shapefile as a VirtualShape with Spatialite. The syntax to use for the datasource is "VirtualShape:/path/to/shapefile.shp" (the shapefile must be a "real" file).
This gives the capability to use the spatial operations of Spatialite (note that spatial indexes on virtual tables are not available).
After the extension is loaded, a virtual table, corresponding to a OGR layer, can be created with one of the following SQL statement :
CREATE VIRTUAL TABLE table_name USING VirtualOGR(datasource_name); CREATE VIRTUAL TABLE table_name USING VirtualOGR(datasource_name, update_mode); CREATE VIRTUAL TABLE table_name USING VirtualOGR(datasource_name, update_mode, layer_name); CREATE VIRTUAL TABLE table_name USING VirtualOGR(datasource_name, update_mode, layer_name, expose_ogr_style);where :
From the sqlite3 console, a typical use case is :
sqlite> SELECT load_extension('libgdal.so'); sqlite> SELECT load_extension('libspatialite.so'); sqlite> CREATE VIRTUAL TABLE poly USING VirtualOGR('poly.shp'); sqlite> SELECT *, ST_Area(GEOMETRY) FROM POLY; 215229.266|168.0|35043411||215229.265625 247328.172|179.0|35043423||247328.171875 261752.781|171.0|35043414||261752.78125 547597.188|173.0|35043416||547597.2109375 15775.758|172.0|35043415||15775.7578125 101429.977|169.0|35043412||101429.9765625 268597.625|166.0|35043409||268597.625 1634833.375|158.0|35043369||1634833.390625 596610.313|165.0|35043408||596610.3359375 5268.813|170.0|35043413||5268.8125
Alternatively, you can use the ogr_datasource_load_layers(datasource_name[, update_mode[, prefix]]) function to automatically load all the layers of a datasource.
sqlite> SELECT load_extension('libgdal.so'); sqlite> SELECT load_extension('libspatialite.so'); sqlite> SELECT ogr_datasource_load_layers('poly.shp'); 1 sqlite> SELECT * FROM sqlite_master; table|poly|poly|0|CREATE VIRTUAL TABLE "poly" USING VirtualOGR('poly.shp', 0, 'poly')Refer to the SQLite SQL dialect for an overview of the capabilities of VirtualOGR tables.
The SQLite driver supports creating new SQLite database files, or adding tables to existing ones. Note that a new database file cannot be created over an existing file.
METADATA=yes/no: This can be used to avoid creating the geometry_columns and spatial_ref_sys tables in a new database. By default these metadata tables are created when a new database is created.
SPATIALITE=yes/no: (Starting with GDAL 1.7.0) Create the SpatiaLite flavour of the metadata
tables, which are a bit differ from the metadata used by this OGR driver and
from OGC specifications. Implies METADATA=yes.
Please note: (Starting with GDAL 1.9.0) OGR must be linked against libspatialite in order to support
insert/write on SpatiaLite; if not, read-only mode is enforced.
Attempting to perform any insert/write on SpatiaLite skipping the appropriate library support simply produces broken (corrupted) DB-files.
Important notice: when the underlaying libspatialite is v.2.3.1 (or any previous
version) any Geometry will be casted to 2D [XY], because earlier versions of this library
are simply able to support 2D [XY] dimensions. Version 2.4.0 (or any subsequent) is required in order to support 2.5D [XYZ].
INIT_WITH_EPSG=yes/no: (Starting with GDAL 1.8.0) Insert the content of the EPSG CSV files
into the spatial_ref_sys table. Defaults to NO for regular SQLite databases.
Please note: if SPATIALITE=yes and the underlaying libspatialite is v2.4 or v3.X,
INIT_WITH_EPSG is ignored; those library versions will unconditionally load the EPSG
dataset into the spatial_ref_sys table when creating a new DB (self-initialization). Starting
with libspatialite 4.0, INIT_WITH_EPSG defaults to YES, but can be set to NO.
FORMAT=WKB/WKT/SPATIALITE: Controls the format used for the geometry column. By default WKB (Well Known Binary) is used. This is generally more space and processing efficient, but harder to inspect or use in simple applications than WKT (Well Known Text). SpatiaLite extension uses its own binary format to store geometries and you can choose it as well. It will be selected automatically when SpatiaLite database is opened or created with SPATIALITE=yes option. SPATIALITE value is available starting with GDAL 1.7.0.
LAUNDER=yes/no: Controls whether layer and field names will be laundered for easier use in SQLite. Laundered names will be convered to lower case and some special characters(' - #) will be changed to underscores. Default to yes.
SPATIAL_INDEX=yes/no: (Starting with GDAL 1.7.0) If the database is of the SpatiaLite flavour, and if OGR is linked against libspatialite, this option can be used to control if a spatial index must be created. Default to yes.
COMPRESS_GEOM=yes/no: (Starting with GDAL 1.9.0) If the format of the geometry BLOB is of the SpatiaLite flavour, this option can be used to control if the compressed format for geometries (LINESTRINGs, POLYGONs) must be used. This format is understood by Spatialite v2.4 (or any subsequent version). Default to no. Note: when updating an existing Spatialite DB, the COMPRESS_GEOM configuration option can be set to produce similar results for appended/overwritten features.
SRID=srid: (Starting with GDAL 1.10) Used to force the SRID number of the SRS associated with the layer. When this option isn't specified and that a SRS is associated with the layer, a search is made in the spatial_ref_sys to find a match for the SRS, and, if there is no match, a new entry is inserted for the SRS in the spatial_ref_sys table. When the SRID option is specified, this search (and the eventual insertion of a new entry) will not be done : the specified SRID is used as such.
COMPRESS_COLUMNS=column_name1[,column_name2, ...]: (Starting with GDAL 1.10.0) A list of (String) columns that must be compressed with ZLib DEFLATE algorithm. This might be beneficial for databases that have big string blobs. However, use with care, since the value of such columns will be seen as compressed binary content with other SQLite utilities (or previous OGR versions). With OGR, when inserting, modifying or queryings compressed columns, compression/decompression is done transparently. However, such columns cannot be (easily) queried with an attribute filter or WHERE clause. Note: in table definition, such columns have the "VARCHAR_deflate" declaration type.
SQLite usually has a very minimal memory foot-print; just about 20MB of RAM are reserved to store the internal Page Cache [merely 2000 pages]. This value too may well be inappropriate under many circumstances, most notably when accessing some really huge DB-file containing many tables related to a corresponding Spatial Index. Explicitly setting a much more generously dimensioned internal Page Cache may often help to get a noticeably better performance. Starting since GDAL 1.9.0 you can explicitly set the internal Page Cache size using the configuration option OGR_SQLITE_CACHE value [value being measured in MB]; if your HW has enough available RAM, defining a Cache size as big as 512MB (or even 1024MB) may sometimes help a lot in order to get better performance.
Setting the OGR_SQLITE_SYNCHRONOUS configuration option to OFF might also increase performance when creating SQLite databases (altough at the expense of integrity in case of interruption/crash ).