OSTRICH

Offset-enabled TRIple store for CHangesets

OSTRICH is an RDF triple store that allows multiple versions of a dataset to be stored and queried at the same time.

The store is a hybrid between snapshot, delta and timestamp-based storage, which provides a good trade-off between storage size and query time. It provides several built-in algorithms to enable efficient iterator-based queries at a certain version, between any two versions, and for versions. These queries support limits and offsets for any triple pattern.

Insertion is done by first inserting a dataset snapshot, which is encoded in HDT. After that, deltas can be inserted, which contain additions and deletions based on the last delta or snapshot.

Learn more about the internals of OSTRICH in the following articles:

• Triple Storage for Random-Access Versioned Querying of RDF Archives
• Scaling Large RDF Archives To Very Long Histories
• OSTRICH: Versioned Random-Access Triple Store
• GLENDA: Querying RDF Archives with full SPARQL

Building

OSTRICH requires ZLib, Kyoto Cabinet, Boost, Serd, Raptor2 and CMake (compilation only) to be installed. Inspect our CI workflow file to see how dependencies are installed on Ubuntu.

Compile:

$ mkdir build
$ cd build
$ cmake ..
$ make

Running

The OSTRICH dataset will always be loaded from the current directory.

Tests

build/ostrich_test

Query

build/ostrich-query-version-materialized patch_id s p o
build/ostrich-query-delta-materialized patch_id patch_id_end s p o
build/ostrich-query-version patch_id s p o

Insert

build/ostrich-insert [-v] patch_id [+|- file_1.nt [file_2.nt [...]]]*

Input deltas must be sorted in SPO-order.

Evaluate

Only load changesets from a path structured as path_to_patch_directory/patch_id/main.nt.additions.txt and path_to_patch_directory/patch_id/main.nt.deletions.txt.

build/ostrich-evaluate path_to_patch_directory patch_id patch_id_end

CSV-formatted insert data will be emitted: version,added,durationms,rate,accsize.

Load changesets AND query with triple patterns from the given file on separate lines, with the given number of replications.

build/ostrich-evaluate path_to_patch_directory patch_id patch_id_end patch_to_queries/queries.txt s|p|o nr_replications

CSV-formatted query data will be emitted (time in microseconds) for all versions for the three query types: patch,offset,limit,count-ms,lookup-mus,results.

Docker

Alternatively, OSTRICH can be built and run using Docker.

Build

docker build -t ostrich .

Instead of building the container yourself, you can use the pre-built image from DockerHub.

docker pull rdfostrich/ostrich

Test

docker run --rm -it --entrypoint /opt/patchstore/build/ostrich_test ostrich

Query

docker run --rm -it --entrypoint /opt/ostrich/build/ostrich-query-version-materialized ostrich patch_id s p o
docker run --rm -it --entrypoint /opt/ostrich/build/ostrich-query-delta-materialized ostrich patch_id patch_id_end s p o
docker run --rm -it --entrypoint /opt/ostrich/build/ostrich-query-version ostrich s p o

Insert

docker run --rm -it --entrypoint /opt/ostrich/build/ostrich-insert ostrich [-v] patch_id [+|- file_1.nt [file_2.nt [...]]]*

Evaluate

Only load changesets from a path structured as path_to_patch_directory/patch_id/main.nt.additions.txt and path_to_patch_directory/patch_id/main.nt.deletions.txt.

docker run --rm -it -v path_to_patch_directory:/var/patches ostrich /var/patches patch_id patch_id_end

Load changesets AND query with triple patterns from the given file on separate lines, with the given number of replications.

docker run --rm -it -v path_to_patch_directory:/var/patches -v patch_to_queries:/var/queries ostrich /var/patches patch_id patch_id_end /var/queries/queries.txt s|p|o nr_replications

Enable debug mode:

docker run --rm -it -v path_to_patch_directory:/var/patches -v patch_to_queries:/var/queries -v path_to_crash_dir:/crash --privileged=true ostrich --debug /var/patches patch_id patch_id_end /var/queries/queries.txt s|p|o nr_replications

Compiler variables

PATCH_INSERT_BUFFER_SIZE: The size of the triple parser buffer during patch insertion. (default 100)

FLUSH_POSITIONS_COUNT: The amount of triples after which the patch positions should be flushed to disk, to avoid memory issues. (default 500000)

FLUSH_TRIPLES_COUNT: The amount of triples after which the store should be flushed to disk, to avoid memory issues. (default 500000)

KC_MEMORY_MAP_SIZE: The KC memory map size per tree. (default 1LL << 27 = 128MB)

KC_PAGE_CACHE_SIZE: The KC page cache size per tree. (default 1LL << 25 = 32MB)

MIN_ADDITION_COUNT: The minimum addition triple count so that it will be stored in the db. Changing this value only has effect during insertion time. Lookups are compatible with any value. (default 200)

Cite

If you are using or extending OSTRICH as part of a scientific publication, we would appreciate a citation of our article.

@article{taelman_jws_ostrich_2018,
  author = {Taelman, Ruben and Vander Sande, Miel and Van Herwegen, Joachim and Mannens, Erik and Verborgh, Ruben},
  title = {Triple Storage for Random-Access Versioned Querying of RDF Archives},
  journal = {Journal of Web Semantics},
  year = {2018},
  month = aug,
  url = {https://rdfostrich.github.io/article-jws2018-ostrich/}
}

License

This software is written by Ruben Taelman, Olivier Pelgrin, and colleagues.

This code is copyrighted by Ghent University – imec and Aalborg University, and is released under the MIT license.