Running Distributed Load

By default, Apache Solr Benchmark generates all load from the machine where you run the solr-benchmark command. For large clusters or high-throughput benchmarks, a single machine may become the bottleneck before Solr does. Distributed load generation lets you spread the workload across multiple machines.

Architecture

In distributed mode, one machine acts as the coordinator and one or more additional machines act as workers. The coordinator:

  • Parses and distributes the workload
  • Sends task assignments to each worker
  • Collects and aggregates metrics from all workers

Each worker:

  • Executes its assigned portion of the load against the Solr cluster
  • Sends metrics back to the coordinator

The coordinator and workers communicate over the network. All machines must be able to reach each other and the Solr cluster.

Prerequisites

  • The same version of Apache Solr Benchmark must be installed on all coordinator and worker machines
  • All machines must have network access to the Solr cluster
  • Workers must have the workload data files available at the same path as the coordinator (or accessible via a shared filesystem)

Configuration

Start the benchmark daemon on each worker

On each worker machine, start the benchmark daemon:

solr-benchmark daemon --worker-ip WORKER_IP

Replace WORKER_IP with the IP address of that worker machine.

Run the benchmark from the coordinator

On the coordinator machine, pass the worker IPs via --worker-ips:

solr-benchmark run \
  --workload nyc_taxis \
  --pipeline benchmark-only \
  --target-hosts solr1:8983,solr2:8983,solr3:8983 \
  --worker-ips 192.168.1.10,192.168.1.11,192.168.1.12

The coordinator automatically divides the corpus and task schedule across the specified workers.

How load is divided

The corpus is partitioned by line ranges in the NDJSON data files. Each worker receives a non-overlapping slice of documents to index. For query tasks, each worker runs the full query schedule with the specified clients count, so the effective query rate is clients × number_of_workers.

Example: 3-worker setup 

# On worker-1 (192.168.1.10):
solr-benchmark daemon --worker-ip 192.168.1.10

# On worker-2 (192.168.1.11):
solr-benchmark daemon --worker-ip 192.168.1.11

# On worker-3 (192.168.1.12):
solr-benchmark daemon --worker-ip 192.168.1.12

# On the coordinator:
solr-benchmark run \
  --workload nyc_taxis \
  --pipeline benchmark-only \
  --target-hosts solr1:8983,solr2:8983 \
  --worker-ips 192.168.1.10,192.168.1.11,192.168.1.12 \
  --user-tag "workers:3"

When to use distributed load

Consider distributed load generation when:

  • A single machine cannot sustain the target query rate (CPU or network saturated on the load generator, not on Solr)
  • You want to simulate many independent clients connecting from different source IPs
  • Your indexing throughput is limited by the speed of reading and sending documents from the coordinator