Drop the place_classtype tables on migration

[nominatim.git] / docs / admin / Migration.md

# Database Migrations

Nominatim offers automatic migrations for versions 4.3+. Please follow
the following steps:

* Stop any updates that are potentially running
* Update the backend: `pip install -U nominatim-db`
* Go to your project directory and run `nominatim admin --migrate`
* Update the frontend: `pip install -U nominatim-api`
* (optionally) Restart updates

Below you find additional migrations and hints about other structural and
breaking changes. **Please read them before running the migration.**

!!! note
    If you are migrating from a version <4.3, you need to install 4.3
    and migrate to 4.3 first. Then you can migrate to the current
    version. It is strongly recommended to do a reimport instead.

## 5.3.0 -> 5.4.0

### Removal of the place_classtype tables

Category search now runs against the `categories` column of `placex`, so the
per-category `place_classtype_*` tables are not used anymore. The migration
drops them. This frees a considerable amount of disk space on a planet
database but may take a moment as there is one table per category.

The `--min` option of `nominatim special-phrases` only restricted which of
those tables were created. It has no effect anymore and has been removed.

## 5.2.0 -> 5.3.0

#### Expensive table migrations

This version introduces new tables for postcodes, interpolations and
associatedStreet relations. Migration is possible but will take a bit
longer. It is recommended to take Nominatim offline while migrating to 5.3.

## 5.1.0 -> 5.2.0

### Lua import style: required extratags removed

Tags that are required by Nominatim as extratags are now always included
independent of what is defined in the style. The line

    flex.add_for_extratags('required')

is no longer required in custom styles and will throw an error. Simply
remove the line from your style.

## 4.5.0 -> 5.0.0

### PHP frontend removed

The PHP frontend has been completely removed. Please switch to the Python
frontend.

Without the PHP code, the `nominatim refresh --website` command is no longer
needed. It currently omits a warning and does otherwise nothing. It will be
removed in later versions of Nominatim. So make sure you remove it from your
scripts.

### CMake building removed

Nominatim can now only be installed via pip. Please follow the installation
instructions for the current version to change to pip.

### osm2pgsql no longer vendored in

Nominatim no longer ships its own version of osm2pgsql. Please install a
stock version of osm2pgsql from your distribution. See the
[installation instruction for osm2pgsql](https://osm2pgsql.org/doc/install.html)
for details. A minimum version of 1.8 is required. The current stable versions
of Ubuntu and Debian already ship with an appropriate versions. For older
installations, you may have to compile a newer osm2pgsql yourself.

### Legacy tokenizer removed

The `legacy` tokenizer is no longer enabled. This tokenizer has been superseded
by the `ICU` tokenizer a long time ago. In the unlikely case that your database
still uses the `legacy` tokenizer, you must reimport your database.

### osm2pgsql style overhauled

There are some fundamental changes to how customized osm2pgsql styles should
be written. The changes are mostly backwards compatible, i.e. custom styles
should still work with the new implementation. The only exception is a
customization of the `process_tags()` function. This function is no longer
considered public and neither are the helper functions used in it.
They currently still work but will be removed at some point. If you have
been making changes to `process_tags`, please review your style and try
to switch to the new convenience functions.

For more information on the changes, see the
[pull request](https://github.com/osm-search/Nominatim/pull/3615)
and read the new
[customization documentation](https://nominatim.org/release-docs/latest/customize/Import-Styles/).

## 4.4.0 -> 4.5.0

### New structure for Python packages

The nominatim Python package has been split into `nominatim-db` and `nominatim-api`.
Any imports need to be adapted accordingly.

If you are running the Python frontend, change the server module from
`nominatim.server.falcon.server` to `nominatim_api.server.falcon.server`.

If you are using the Nominatim library, all imports need to be changed
from `nominatim.api.<module>` to `nominatim_api.<module>`.

If you have written custom tokenizers or sanitizers, the appropriate modules
are now found in `nominatim_db`.

## 4.2.0 -> 4.3.0

### New indexes for reverse lookup

The reverse lookup algorithm has changed slightly to improve performance.
This change needs a different index in the database. The required index
will be automatically build during migration. Until the new index is available
performance of the /reverse endpoint is significantly reduced. You should
therefore either remove traffic from the machine before attempting a
version update or create the index manually **before** starting the update
using the following SQL:

```sql
CREATE INDEX IF NOT EXISTS idx_placex_geometry_reverse_lookupPlaceNode
  ON placex USING gist (ST_Buffer(geometry, reverse_place_diameter(rank_search)))
  WHERE rank_address between 4 and 25 AND type != 'postcode'
    AND name is not null AND linked_place_id is null AND osm_type = 'N';
```

## 4.0.0 -> 4.1.0

### ICU tokenizer is the new default

Nominatim now installs the [ICU tokenizer](../customize/Tokenizers.md#icu-tokenizer)
by default. This only has an effect on newly installed databases. When
updating older databases, it keeps its installed tokenizer. If you still
run with the legacy tokenizer, make sure to compile Nominatim with the
PostgreSQL module, see [Installation](Installation.md#building-nominatim).

### geocodejson output changed

The `type` field of the geocodejson output has changed. It now contains
the address class of the object instead of the value of the OSM tag. If
your client has used the `type` field, switch them to read `osm_value`
instead.

## 3.7.0 -> 4.0.0

### NOMINATIM_PHRASE_CONFIG removed

Custom blacklist configurations for special phrases now need to be handed
with the `--config` parameter to `nominatim special-phrases`. Alternatively
you can put your custom configuration in the project directory in a file
named `phrase-settings.json`.

Version 3.8 also removes the automatic converter for the php format of
the configuration in older versions. If you are updating from Nominatim < 3.7
and still work with a custom `phrase-settings.php`, you need to manually
convert it into a json format.

### PHP utils removed

The old PHP utils have now been removed completely. You need to switch to
the appropriate functions of the nominatim  command line tool. See
[Introducing `nominatim` command line tool](#introducing-nominatim-command-line-tool)
below.

## 3.6.0 -> 3.7.0

### New format and name of configuration file

The configuration for an import is now saved in a `.env` file in the project
directory. This file follows the dotenv format. For more information, see
the [installation chapter](Import.md#configuration-setup-in-env).

To migrate to the new system, create a new project directory, add the `.env`
file and port your custom configuration from `settings/local.php`. Most
settings are named similar and only have received a `NOMINATIM_` prefix.
Use the default settings in `settings/env.defaults` as a reference.

### New location for data files

External data files for Wikipedia importance, postcodes etc. are no longer
expected to reside in the source tree by default. Instead they will be searched
in the project directory. If you have an automated setup script you must
either adapt the download location or explicitly set the location of the
files to the old place in your `.env`.

### Introducing `nominatim` command line tool

The various php utilities have been replaced with a single `nominatim`
command line tool. Make sure to adapt any scripts. There is no direct 1:1
matching between the old utilities and the commands of nominatim CLI. The
following list gives you a list of nominatim sub-commands that contain
functionality of each script:

* ./utils/setup.php: `import`, `freeze`, `refresh`
* ./utils/update.php: `replication`, `add-data`, `index`, `refresh`
* ./utils/specialphrases.php: `special-phrases`
* ./utils/check_import_finished.php: `admin`
* ./utils/warm.php: `admin`
* ./utils/export.php: `export`

Try `nominatim <command> --help` for more information about each subcommand.

`./utils/query.php` no longer exists in its old form. `nominatim search`
provides a replacement but returns different output.

### Switch to normalized house numbers

The housenumber column in the placex table uses now normalized version.
The automatic migration step will convert the column but this may take a
very long time. It is advisable to take the machine offline while doing that.

## 3.5.0 -> 3.6.0

### Change of layout of search_name_* tables

The table need a different index for nearest place lookup. Recreate the
indexes using the following shell script:

```bash
for table in `psql -d nominatim -c "SELECT tablename FROM pg_tables WHERE tablename LIKE 'search_name_%'" -tA | grep -v search_name_blank`;
do
    psql -d nominatim -c "DROP INDEX idx_${table}_centroid_place; CREATE INDEX idx_${table}_centroid_place ON ${table} USING gist (centroid) WHERE ((address_rank >= 2) AND (address_rank <= 25)); DROP INDEX idx_${table}_centroid_street; CREATE INDEX idx_${table}_centroid_street ON ${table} USING gist (centroid) WHERE ((address_rank >= 26) AND (address_rank <= 27))";
done
```

### Removal of html output

The debugging UI is no longer directly provided with Nominatim. Instead we
now provide a simple Javascript application. Please refer to
[Setting up the Nominatim UI](Setup-Nominatim-UI.md) for details on how to
set up the UI.

The icons served together with the API responses have been moved to the
nominatim-ui project as well. If you want to keep the `icon` field in the
response, you need to set `CONST_MapIcon_URL` to the URL of the `/mapicon`
directory of nominatim-ui.

### Change order during indexing

When reindexing places during updates, there is now a different order used
which needs a different database index. Create it with the following SQL command:

```sql
CREATE INDEX idx_placex_pendingsector_rank_address
  ON placex
  USING BTREE (rank_address, geometry_sector)
  WHERE indexed_status > 0;
```

You can then drop the old index with:

```sql
DROP INDEX idx_placex_pendingsector;
```

### Unused index

This index has been unused ever since the query using it was changed two years ago. Saves about 12GB on a planet installation.

```sql
DROP INDEX idx_placex_geometry_reverse_lookupPoint;
```

### Switching to dotenv

As part of the work changing the configuration format, the configuration for
the website is now using a separate configuration file. To create the
configuration file, run the following command after updating:

```sh
./utils/setup.php --setup-website
```

### Update SQL code

To update the SQL code to the leatest version run:

```
./utils/setup.php --create-functions --enable-diff-updates --create-partition-functions
```

## 3.4.0 -> 3.5.0

### New Wikipedia/Wikidata importance tables

The `wikipedia_*` tables have a new format that also includes references to
Wikidata. You need to update the computation functions and the tables as
follows:

  * download the new Wikipedia tables as described in the import section
  * reimport the tables: `./utils/setup.php --import-wikipedia-articles`
  * update the functions: `./utils/setup.php --create-functions --enable-diff-updates`
  * create a new lookup index:
```sql
CREATE INDEX idx_placex_wikidata
  ON placex
  USING BTREE ((extratags -> 'wikidata'))
  WHERE extratags ? 'wikidata'
    AND class = 'place'
    AND osm_type = 'N'
    AND rank_search < 26;
```
  * compute importance: `./utils/update.php --recompute-importance`

The last step takes about 10 hours on the full planet.

Remove one function (it will be recreated in the next step):

```sql
DROP FUNCTION create_country(hstore,character varying);
```

Finally, update all SQL functions:

```sh
./utils/setup.php --create-functions --enable-diff-updates --create-partition-functions
```

## 3.3.0 -> 3.4.0

### Reorganisation of location_area_country table

The table `location_area_country` has been optimized. You need to switch to the
new format when you run updates. While updates are disabled, run the following
SQL commands:

```sql
CREATE TABLE location_area_country_new AS
  SELECT place_id, country_code, geometry FROM location_area_country;
DROP TABLE location_area_country;
ALTER TABLE location_area_country_new RENAME TO location_area_country;
CREATE INDEX idx_location_area_country_geometry ON location_area_country USING GIST (geometry);
CREATE INDEX idx_location_area_country_place_id ON location_area_country USING BTREE (place_id);
```

Finally, update all SQL functions:

```sh
./utils/setup.php --create-functions --enable-diff-updates --create-partition-functions
```

## 3.2.0 -> 3.3.0

### New database connection string (DSN) format

Previously database connection setting (`CONST_Database_DSN` in `settings/*.php`) had the format

   * (simple) `pgsql://@/nominatim`
   * (complex) `pgsql://johndoe:secret@machine1.domain.com:1234/db1`

The new format is

   * (simple) `pgsql:dbname=nominatim`
   * (complex) `pgsql:dbname=db1;host=machine1.domain.com;port=1234;user=johndoe;password=secret`

### Natural Earth country boundaries no longer needed as fallback

```sql
DROP TABLE country_naturalearthdata;
```

Finally, update all SQL functions:

```sh
./utils/setup.php --create-functions --enable-diff-updates --create-partition-functions
```

### Configurable Address Levels

The new configurable address levels require a new table. Create it with the
following command:

```sh
./utils/update.php --update-address-levels
```

## 3.1.0 -> 3.2.0

### New reverse algorithm

The reverse algorithm has changed and requires new indexes. Run the following
SQL statements to create the indexes:

```sql
CREATE INDEX idx_placex_geometry_reverse_lookupPoint
  ON placex
  USING gist (geometry)
  WHERE (name IS NOT null or housenumber IS NOT null or rank_address BETWEEN 26 AND 27)
    AND class NOT IN ('railway','tunnel','bridge','man_made')
    AND rank_address >= 26
    AND indexed_status = 0
    AND linked_place_id IS null;
CREATE INDEX idx_placex_geometry_reverse_lookupPolygon
  ON placex USING gist (geometry)
  WHERE St_GeometryType(geometry) in ('ST_Polygon', 'ST_MultiPolygon')
    AND rank_address between 4 and 25
    AND type != 'postcode'
    AND name is not null
    AND indexed_status = 0
    AND linked_place_id is null;
CREATE INDEX idx_placex_geometry_reverse_placeNode
  ON placex USING gist (geometry)
  WHERE osm_type = 'N'
    AND rank_search between 5 and 25
    AND class = 'place'
    AND type != 'postcode'
    AND name is not null
    AND indexed_status = 0
    AND linked_place_id is null;
```

You also need to grant the website user access to the `country_osm_grid` table:

```sql
GRANT SELECT ON table country_osm_grid to "www-user";
```

Replace the `www-user` with the user name of your website server if necessary.

You can now drop the unused indexes:

```sql
DROP INDEX idx_placex_reverse_geometry;
```

Finally, update all SQL functions:

```sh
./utils/setup.php --create-functions --enable-diff-updates --create-partition-functions
```

## 3.0.0 -> 3.1.0

### Postcode Table

A new separate table for artificially computed postcode centroids was introduced.
Migration to the new format is possible but **not recommended**.

Create postcode table and indexes, running the following SQL statements:

```sql
CREATE TABLE location_postcode
  (place_id BIGINT, parent_place_id BIGINT, rank_search SMALLINT,
   rank_address SMALLINT, indexed_status SMALLINT, indexed_date TIMESTAMP,
   country_code varchar(2), postcode TEXT,
   geometry GEOMETRY(Geometry, 4326));
CREATE INDEX idx_postcode_geometry ON location_postcode USING GIST (geometry);
CREATE UNIQUE INDEX idx_postcode_id ON location_postcode USING BTREE (place_id);
CREATE INDEX idx_postcode_postcode ON location_postcode USING BTREE (postcode);
GRANT SELECT ON location_postcode TO "www-data";
DROP TYPE IF EXISTS nearfeaturecentr CASCADE;
CREATE TYPE nearfeaturecentr AS (
  place_id BIGINT,
  keywords int[],
  rank_address smallint,
  rank_search smallint,
  distance float,
  isguess boolean,
  postcode TEXT,
  centroid GEOMETRY
);
```

Add postcode column to `location_area` tables with SQL statement:

```sql
ALTER TABLE location_area ADD COLUMN postcode TEXT;
```

Then reimport the functions:

```sh
./utils/setup.php --create-functions --enable-diff-updates --create-partition-functions
```

Create appropriate triggers with SQL:

```sql
CREATE TRIGGER location_postcode_before_update BEFORE UPDATE ON location_postcode
    FOR EACH ROW EXECUTE PROCEDURE postcode_update();
```

Finally populate the postcode table (will take a while):

```sh
./utils/setup.php --calculate-postcodes --index --index-noanalyse
```

This will create a working database. You may also delete the old artificial
postcodes now. Note that this may be expensive and is not absolutely necessary.
The following SQL statement will remove them:

```sql
DELETE FROM place_addressline a USING placex p
 WHERE a.address_place_id = p.place_id and p.osm_type = 'P';
ALTER TABLE placex DISABLE TRIGGER USER;
DELETE FROM placex WHERE osm_type = 'P';
ALTER TABLE placex ENABLE TRIGGER USER;
```