X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/e9e14834bc3ae398e1ac161281591bd99c9b458f..23369ce9704d3815faae22697338916a60ddf682:/docs/admin/Installation.md?ds=inline diff --git a/docs/admin/Installation.md b/docs/admin/Installation.md index f6692f58..78062908 100644 --- a/docs/admin/Installation.md +++ b/docs/admin/Installation.md @@ -4,8 +4,8 @@ This page contains generic installation instructions for Nominatim and its prerequisites. There are also step-by-step instructions available for the following operating systems: - * [Ubuntu 22.04](../appendix/Install-on-Ubuntu-22.md) - * [Ubuntu 20.04](../appendix/Install-on-Ubuntu-20.md) + * [Ubuntu 24.04](Install-on-Ubuntu-24.md) + * [Ubuntu 22.04](Install-on-Ubuntu-22.md) These OS-specific instructions can also be found in executable form in the `vagrant/` directory. @@ -27,48 +27,50 @@ and can't offer support. otherwise import and queries will be slow to the point of being unusable. Query performance has marked improvements with PostgreSQL 13+ and PostGIS 3.2+. -For compiling: - - * [cmake](https://cmake.org/) - * [expat](https://libexpat.github.io/) - * [proj](https://proj.org/) - * [bzip2](http://www.bzip.org/) - * [zlib](https://www.zlib.net/) - * [ICU](http://site.icu-project.org/) - * [Boost libraries](https://www.boost.org/), including system and filesystem - * PostgreSQL client libraries - * a recent C++ compiler (gcc 5+ or Clang 3.8+) - For running Nominatim: * [PostgreSQL](https://www.postgresql.org) (9.6+ will work, 11+ strongly recommended) * [PostGIS](https://postgis.net) (2.2+ will work, 3.0+ strongly recommended) + * [osm2pgsql](https://osm2pgsql.org) (1.8+, optional when building with CMake) * [Python 3](https://www.python.org/) (3.7+) - * [Psycopg2](https://www.psycopg.org) (2.7+) + +Furthermore the following Python libraries are required: + + * [Psycopg3](https://www.psycopg.org) * [Python Dotenv](https://github.com/theskumar/python-dotenv) * [psutil](https://github.com/giampaolo/psutil) * [Jinja2](https://palletsprojects.com/p/jinja/) - * [SQLAlchemy](https://www.sqlalchemy.org/) (1.4+ with greenlet support) - * [asyncpg](https://magicstack.github.io/asyncpg) (0.8+) * [PyICU](https://pypi.org/project/PyICU/) * [PyYaml](https://pyyaml.org/) (5.1+) * [datrie](https://github.com/pytries/datrie) - * [PHP](https://php.net) (7.3+) - * PHP-pgsql - * PHP-intl (bundled with PHP) - * PHP-cgi (for running queries from the command line) + +These will be installed automatically when using pip installation. + +When using legacy CMake-based installation: + + * [cmake](https://cmake.org/) + * [expat](https://libexpat.github.io/) + * [proj](https://proj.org/) + * [bzip2](http://www.bzip.org/) + * [zlib](https://www.zlib.net/) + * [ICU](http://site.icu-project.org/) + * [nlohmann/json](https://json.nlohmann.me/) + * [Boost libraries](https://www.boost.org/), including system and file system + * PostgreSQL client libraries + * a recent C++ compiler (gcc 5+ or Clang 3.8+) For running continuous updates: * [pyosmium](https://osmcode.org/pyosmium/) -For running the experimental Python frontend: +For running the Python frontend: + * [SQLAlchemy](https://www.sqlalchemy.org/) (1.4.31+ with greenlet support) + * [asyncpg](https://magicstack.github.io/asyncpg) (0.8+, only when using SQLAlchemy < 2.0) * one of the following web frameworks: * [falcon](https://falconframework.org/) (3.0+) - * [sanic](https://sanic.dev) * [starlette](https://www.starlette.io/) - * [uvicorn](https://www.uvicorn.org/) (only with falcon and starlette framworks) + * [uvicorn](https://www.uvicorn.org/) For dependencies for running tests and building documentation, see the [Development section](../develop/Development-Environment.md). @@ -84,7 +86,7 @@ Take into account that the OSM database is growing fast. Fast disks are essential. Using NVME disks is recommended. Even on a well configured machine the import of a full planet takes -around 2 days. On traditional spinning disks, 7-8 days are more realistic. +around 2.5 days. When using traditional SSDs, 4-5 days are more realistic. ## Tuning the PostgreSQL database @@ -96,14 +98,16 @@ your `postgresql.conf` file. maintenance_work_mem = (10GB) autovacuum_work_mem = 2GB work_mem = (50MB) - effective_cache_size = (24GB) synchronous_commit = off max_wal_size = 1GB - checkpoint_timeout = 10min + checkpoint_timeout = 60min checkpoint_completion_target = 0.9 + random_page_cost = 1.0 + wal_level = minimal + max_wal_senders = 0 The numbers in brackets behind some parameters seem to work fine for -64GB RAM machine. Adjust to your setup. A higher number for `max_wal_size` +128GB RAM machine. Adjust to your setup. A higher number for `max_wal_size` means that PostgreSQL needs to run checkpoints less often but it does require the additional space on your disk. @@ -116,15 +120,6 @@ you might consider setting: and even reduce `autovacuum_work_mem` further. This will reduce the amount of memory that autovacuum takes away from the import process. -For the initial import, you should also set: - - fsync = off - full_page_writes = off - -Don't forget to re-enable them after the initial import or you risk database -corruption. - - ## Downloading and building Nominatim ### Downloading the latest release @@ -149,6 +144,23 @@ wget -O Nominatim/data/country_osm_grid.sql.gz https://nominatim.org/data/countr ### Building Nominatim +#### Building the latest development version with pip + +Nominatim is easiest to run from its own virtual environment. To create one, run: + + sudo apt-get install virtualenv + virtualenv /srv/nominatim-venv + +To install Nominatim directly from the source tree into the virtual environment, run: + + /srv/nominatim-venv/bin/pip install packaging/nominatim-{db,api} + +#### Building in legacy CMake mode + +!!! warning + Installing Nominatim through CMake is now deprecated. The infrastructure + will be removed in Nominatim 5.0. Please switch to pip installation. + The code must be built in a separate directory. Create the directory and change into it. @@ -166,17 +178,6 @@ make sudo make install ``` -!!! warning - The default installation no longer compiles the PostgreSQL module that - is needed for the legacy tokenizer from older Nominatim versions. If you - are upgrading an older database or want to run the - [legacy tokenizer](../customize/Tokenizers.md#legacy-tokenizer) for - some other reason, you need to enable the PostgreSQL module via - cmake: `cmake -DBUILD_MODULE=on ../Nominatim`. To compile the module - you need to have the server development headers for PostgreSQL installed. - On Ubuntu/Debian run: `sudo apt install postgresql-server-dev-` - - Nominatim installs itself into `/usr/local` per default. To choose a different installation directory add `-DCMAKE_INSTALL_PREFIX=` to the cmake command. Make sure that the `bin` directory is available in your path