adnapibar/custom-workloads-opensearch

Benchmark with custom workload with fake data

The repository contains some helper scripts to setup an EC2 machine and bulk index generate data.

Modify existing workload with additional operation

Copy and modify the nyc-taxis workload

mkdir ~/workloads
cd workloads
cp -r .benchmark/benchmarks/default/nyc_taxis .

modify operations/default.json directory to add the new operation e.g.

{
    "name": "date_histogram_subaggs",
    "operation-type": "search",
    "body": {
        "size": 0,
        "query": {
            "range": {
                "dropoff_datetime": {
                    "gte": "2015-01-01 00:00:00",
                    "lte": "2015-01-01 23:59:59"
                }
            }
        },
        "aggs": {
            "dropoffs_over_time": {
                "date_histogram": {
                    "field": "dropoff_datetime",
                    "interval": "60s"
                },
                "aggs": {
                    "payment_type": {
                        "terms": {
                            "field": "payment_type"
                        },
                        "aggs": {
                            "total_payment_amount": {
                                "sum": {
                                    "field": "total_amount"
                                }
                            }
                        }
                    }
                }
            }
        }
    }
}

Include the operation in the append-no-conflicts section in test_procedures/default.json

Downlaod OpenSearch 2.4 & run benchmark tests

Download a version of OpenSearch and start the instance

wget https://artifacts.opensearch.org/releases/core/opensearch/2.4.0/opensearch-min-2.4.0-linux-x64.tar.gz
tar -xf opensearch-min-2.4.0-linux-x64.tar.gz
cd opensearch-2.4.0
nohup ./bin/opensearch &>>os.log &

Run the benchmark

opensearch-benchmark execute_test --pipeline=benchmark-only --target-hosts=127.0.0.1:9200 --distribution-version=2.4.0 --workload-path=/home/ubuntu/workloads/nyc_taxis --test-procedure=append-no-conflicts

Generate fake data & run workload

The bulk_ingest.py script generates 10,000,000 documents of fake vehicle sales data and bulk indexes into the cluster.

Install Faker and opensearchpy for this.

To create a new workload which can be reused against other versions for comparison

opensearch-benchmark create-workload --workload=vehicles --target-hosts=127.0.0.1:9200 --indices="vehicles" --output-path=~/workloads

The workloads directory contains a workload for vehicles data, run the workload by running

opensearch-benchmark execute_test --pipeline=benchmark-only --target-hosts=127.0.0.1:9200 --distribution-version=1.0.0 --workload-path=~/workloads/vehicles --workload-params='number_of_replicas:0