OSM Database

The Maps OSM database is an important piece of the infrastructure, it's a PostgreSQL database that hosts the data source for tile generation and the Geoshapes service.

The OSM Database needs two processes that are fundamental in order to keep it available with the most updated data from OSM: initial import and replication.

Initial Importing of OSM data into PostgreSQL database

This process needs to be done in all clusters available, one at a time.

When you need to do initial import?

The following scenarios require a fresh install of the OSM database:

First setup of the cluster
Any upgrades or maintenance that require deleting the PostgreSQL data folder, for example:
- OS migration T198622
- Disk space in critical condition because of OSM replication
OSM replication is disabled for a long time range and the database is outdated
- Depending on the server condition, is safer to re-import OSM data instead of roll replication, because it can overload Tilerator and increase significantly the CPU usage

Importing

Checklist

Create a Phab task to track this work. Example.
Depool the maps master.
Isolate the maps master from the rest of the cluster. Example.
Disable Tilerator in puppet. Example.
Delete postgres data, recreate the empty DB.
Run `osm-initial-import` script to reimport data from OSM, needs root access.
Re-enable replication
Repool the maps master
Re-init all slaves (cookbook)
Re-enable tilerator

Depooling the maps master

From the puppetmaster, depooling maps1004 in eqiad:

sudo confctl select "dc=eqiad,cluster=maps,service=kartotherian,name=maps1004.eqiad.wmnet" set/pooled=no

sudo confctl select "dc=eqiad,cluster=maps,service=kartotherian-ssl,name=maps1004.eqiad.wmnet" set/pooled=no

Re-creating an empty Postgres DB

Simply remove the database and let puppet handle the rest:

rm -rf /srv/postgresql/9.6/main/* && run-puppet-agent && service postgresql restart

Re-creating Postgres DB from backup

Use the postgres resync cookbook

The initial-import-script

Initial data load of OSM into postgresql is done by running /usr/local/bin/osm-initial-import on the postgresql master node. This command can take over 12 hours or more, so running in screen or tmux is advised.

osm-initial-import \
    -d <date_of_import> \
    -s  <state_file_url> \
    -x webproxy.eqiad.wmnet:8080

date_of_import: find the latest dump at https://planet.osm.org/pbf/. Example: 160530. Do NOT use "latest", as that might change at any moment.
state_file_url: The URL to the state file corresponding to the dump, find the correct one at http://planet.openstreetmap.org/replication/ (the state file must be older than the dump). Example: http://planet.openstreetmap.org/replication/day/000/001/355.state.txt.

Troubleshouting

If the slave initialization timeouts in puppet, it needs to be run manually:

service postgresql@9.6-main stop

rm -rf /srv/postgresql/9.6/main

sudo -u postgres /usr/bin/pg_basebackup -X stream -D /srv/postgresql/9.6/main -h <maps_master fqdn> -U replication -w

# Run puppet to make sure recovery.conf file is created

service postgresql@9.6-main start

OSM replication

In order to keep the OSM Database updated, we use imposm. This tool periodically fetches data from OSM servers and applies the requested changesets in the PostgreSQL database. When it's complete, the replication script will trigger a request to Tilerator, sending a list of tiles that needs to be regenerated.

Monitoring

You can monitor PostgreSQL activities of the Maps clusters in the following dashboards:

General PostgreSQL metrics from prometheus postgres exporter
OSM sync lag: eqiad, codfw