Run photon with opensearch service

[karussell]

I just wanted to note how to connect to an external opensearch service using photon. Let me know if there are other recommendations.

mkdir opensearch && cd opensearch

# example data extract which creates a photon_data folder
tar xjf ../photon-db-austria-0.7OS-latest.tar.bz2 

# Note that disabling security is only fine when we do not expose port 9200 (as we do due to binding to 127.0.0.1:9200).
# Also note that the command maps only the data directory and all different configs and plugins of the extract will be ignored.
# Remove the option `-d` to see the logs immediately, which important to simplify debugging if the container stops fast.
docker run -d --name opensearch-node \
  -p 127.0.0.1:9200:9200 \
  -e "OPENSEARCH_JAVA_OPTS=-Xms2g -Xmx2g" \
  -e "discovery.type=single-node" \
  -e "DISABLE_SECURITY_PLUGIN=true" \
  -v $(pwd)/photon_data/node_1/data:/usr/share/opensearch/data  opensearchproject/opensearch:latest

# get photon jar in this folder and connect via http port 9200
java -jar photon-opensearch-0.7.4.jar -transport-addresses localhost:9200

[lonvia]

One point to note: OpenSearch creates queries which are relatively complex and run into the configured limits. You want to add a:

 -e "indices.query.bool.max_clause_count=30000"

If you plan to create your own database from Nominatim or json dumps, you can save a bit of disk space by choosing a better compression method:

-e "index.codec=best_compression"

[karussell]

Once the index was create the index won't change anymore. Are there any tuning parameters that I can use (besides something like -e "index.refresh_interval=5m") to improve speed for reading?

And will the data extracts have only a single segment or should I do this after downloading?
(like POST /index/_forcemerge?max_num_segments=1)

[lonvia]

I haven't done any systematic tests yet regarding query performance and OpenSearch tuning. Help would be appreciated here.

I did play around with the force merge. It didn't do anything in terms of making the database smaller, so didn't follow up on that path. I would reconsider if it helps query performance-wise. Again, help with testing would be appreciated.