mirror of
https://github.com/open-metadata/OpenMetadata
synced 2026-05-24 09:39:11 +00:00
0ab31cb647
* fix(rdf): reclaim Fuseki disk via compaction + upgrade Jena 4.10 → 5.6.0 PR #28117's SPARQL cleanup converged the logical RDF state but never freed disk: TDB2 deletes only mark blocks free and the journal grows monotonically until /$/compact runs. RdfIndexApp.clearRdfData() now calls a new RdfStorageInterface.compactStorage() between clearAll() and reloadOntologies() so each recreate run reclaims to a fresh dataset directory. JenaFusekiStorage posts to /$/compact/{dataset}?deleteOld=true and polls /$/tasks/{id} until finished, with failures logged and swallowed (disk hygiene, not correctness). Also unifies the Jena classpath: openmetadata-service was on 4.10.0 and openmetadata-integration-tests on 5.0.0. Both now pin to 5.6.0 via a single root pom property, dropping the apache-jena-libs BOM in favour of explicit jena-core/arq/rdfconnection deps (we're a remote-Fuseki client and never embed TDB; pulling jena-tdb1/2 triggers a Jena 5/6 static-init regression). Picks up CVE-2025-49656 and CVE-2025-50151 (admin-side fixes shipped in Jena 5.5.0). Jena 6.x parked: both 6.0.0 and 6.1.0 hit a recursive clinit bug where TypeMapper.reset reads RDF.dtLangString before RDF.<clinit> completes. Fuseki server bumped 4.10/5.0 → 5.6.0 across all in-repo Dockerfiles; the unmaintained stain/jena-fuseki:* image references in dev compose files switched to building from docker/rdf-store/Dockerfile, and Testcontainers moved to secoresearch/fuseki:5.5.0 (maintained, CVE-fixed; the dataset is created by JenaFusekiStorage.ensureDatasetExists() so the stain-only FUSEKI_DATASET_1 env var is no longer needed).
49 lines
No EOL
1.6 KiB
YAML
49 lines
No EOL
1.6 KiB
YAML
# Standalone Apache Jena Fuseki for RDF/Knowledge Graph storage
|
|
services:
|
|
fuseki:
|
|
# Build from the in-repo Dockerfile (Fuseki 5.6.0). See
|
|
# ../development/docker-compose-fuseki.yml for the full rationale.
|
|
build:
|
|
context: ../rdf-store
|
|
dockerfile: Dockerfile
|
|
image: openmetadata-fuseki:5.6.0
|
|
container_name: fuseki-standalone
|
|
hostname: fuseki
|
|
ports:
|
|
- "3030:3030"
|
|
environment:
|
|
# Local-dev default — production deployments MUST override via
|
|
# FUSEKI_ADMIN_PASSWORD / FUSEKI_OPENMETADATA_PASSWORD env vars.
|
|
- FUSEKI_ADMIN_PASSWORD=${FUSEKI_ADMIN_PASSWORD:-admin}
|
|
- FUSEKI_OPENMETADATA_PASSWORD=${FUSEKI_OPENMETADATA_PASSWORD:-openmetadata-secret}
|
|
- JVM_ARGS=-Xmx4g -Xms2g
|
|
volumes:
|
|
# See docker-compose-fuseki-rosetta.yml — host bind path renamed so
|
|
# existing directories with the old stain layout aren't silently
|
|
# mounted at the new /fuseki-data path.
|
|
- ${DOCKER_VOLUMES_PATH:-./docker-volumes}/fuseki-tdb2-data:/fuseki-data
|
|
networks:
|
|
- fuseki-net
|
|
healthcheck:
|
|
test: ["CMD", "wget", "-q", "--spider", "http://localhost:3030/$/ping"]
|
|
interval: 15s
|
|
timeout: 10s
|
|
retries: 5
|
|
start_period: 30s
|
|
deploy:
|
|
resources:
|
|
limits:
|
|
memory: 4G
|
|
reservations:
|
|
memory: 2G
|
|
|
|
networks:
|
|
fuseki-net:
|
|
driver: bridge
|
|
|
|
# Usage:
|
|
# 1. Create the volume directory: mkdir -p fuseki-volume
|
|
# 2. Start: docker-compose -f docker-compose-fuseki-standalone.yml up -d
|
|
# 3. Access Fuseki UI: http://localhost:3030
|
|
# 4. Login: admin/admin
|
|
# 5. SPARQL endpoint: http://localhost:3030/openmetadata/sparql |