Blazegraph performance: Add JVM heap memory configuration support

[WolfgangFahl]

Problem

Blazegraph containers started via pyomnigraph have poor SPARQL query performance due to insufficient JVM heap memory allocation. The default Docker container likely uses only 512MB-1GB heap, which is inadequate for complex queries on large datasets like GOV genealogy data.

Complex SPARQL queries (especially with property paths like (gp:isPartOf/gp:ref){1,10}) timeout or run very slowly on local dockerized Blazegraph instances.

Root Cause

The BlazegraphConfig.get_docker_run_command() method in omnigraph/servers/blazegraph.py (lines 37-44) does not set JVM heap parameters via the JAVA_OPTS environment variable.

Proposed Solution

Add JVM memory configuration support to the Blazegraph docker run command:

docker_run_command = (
    f"docker run -d --name {self.container_name} "
    f"-e BLAZEGRAPH_UID={os.getuid()} "
    f"-e BLAZEGRAPH_GID={os.getgid()} "
    f"-e JAVA_OPTS='-Xmx4g -Xms4g' "  # Add JVM heap settings
    f"-p {self.port}:8080 "
    f"-v {data_dir}/RWStore.properties:/RWStore.properties "
    f"-v {data_dir}:/data "
    f"{self.image}"
)

Recommendations

1. Make it configurable: Add optional java_opts or heap_size field to ServerConfig
2. Sensible defaults: Use at least 4GB heap by default for production use
3. Consider additional optimizations:
   • Add query hint settings: -Dcom.bigdata.btree.writeRetentionQueue.capacity=4000
   • Add branching factor: -Dcom.bigdata.btree.BTree.branchingFactor=128

Configuration example

Allow users to specify in servers.yaml:

blazegraph:
  server: "blazegraph"
  heap_size: "6g"  # or
  java_opts: "-Xmx6g -Xms6g -Dcom.bigdata.btree.writeRetentionQueue.capacity=4000"

Impact

This significantly improves query performance for users working with large RDF datasets and complex SPARQL queries.