FlatHUB

Simple tabular data query interface

This is a web-based interface for querying large numeric tabular datasets, designed primarily for halo object catalogs from astrophysical simulations. It is built on Elasticsearch to allow efficient filtering and aggregation of datasets with billions of objects.

Python client

Server Installation

These are kubernetes definitions for elastic search and the web server as used in production.

There is also a docker-compose definition for installing everything you need on a single machine. To use it:

Edit the catalog definitions to define the object catalog(s) and fields. There are a number of examples there already.
Change ES_JAVA_OPTS and scale in docker-compose.yml as appropriate for your environment and needs
- The current settings start 5*20GB nodes on a single machine (requiring 100+GB memory). You can modify these to require lower values, e.g. if using a local system with limited memory.
- You can also setup your own elasticsearch cluster on multiple machines, in which case you only need the flathub service
- Update config to point to your elasticsearch cluster if it's not the default
Run docker-compose build to build the application
- Anytime you change the catalog files, you must re-run docker compose build to pick up the latest catalog files.
Run docker-compose up -d to bring up everything (the flathub service will exit with an error -- this is expected). Wait for a couple of minutes, until all the es containers are running.
Afterwards, run docker-compose run --rm flathub -s CATALOG for each defined catalog to create the databases.
Run docker-compose up -d (once more) to bring up the flathub webserver.
- At this point, the FlatHub UI and the Elastic Search endpoints will be available.
Run docker-compose run --rm -v /data/dir:/data flathub -i CATALOG /data/FILE ... to ingest data into each catalog.
- Or, alternatively, define your own script and execute it outside of the containers.
To bring the containers down, use docker compose down. To bring the containers down AND remove all stored data, use docker compose down -v.

Notes:

For macOS installations:
- Install hdf5 with brew install hdf5@1.10
- Run stack install with the --extra-include-dirs=/usr/local/opt/hdf5@1.10/include --extra-lib-dirs=/usr/local/opt/hdf5@1.10/lib
- If brew is not installed in the default location use the appropriate path for the above flags.
For internal FI use: elasticsearch { server = "http://astrodb:9200/" }
OpenAPI: http:///openapi.json

flatironinstitute/flathub

FlatHUB

Simple tabular data query interface

Python client

Server Installation