--- layout: default title: Command flags nav_order: 51 parent: Command reference --- # Command flags OpenSearch Benchmark uses command line flags to change Benchmark's behavior. Not all flags can be used with each command. To find out which flags are supported by a specific command, enter `opensearch-benchmark --h`. All command flags are added to a command using the following syntax: ```bash opensearch-benchmark -- ``` Flags that accept comma-separated values, such `--telemetry`, can also accept a JSON array. This can be defined by passing a file path ending in `.json` or inline as a JSON string. - Comma-seperated values: `opensearch-benchmark ... --test-procedure="ingest-only,search-aggregations"` - JSON file: `opensearch-benchmark ... --workload-params="params.json"` - JSON inline string: `opensearch-benchmark ... --telemetry='["node-stats", "recovery-stats"]'` ## workload-path This can be either a directory that contains a `workload.json` file or a `.json` file with an arbitrary name that contains a workload specification. `--workload-path` and `--workload-repository` as well as `--workload` are mutually exclusive. ## workload-repository This defines the repository from which OpenSearch Benchmark loads workloads. `--workload-path` and `--workload-repository` as well as `--workload` are mutually exclusive. ## workload-revision Defines a specific revision from the workload source tree that OpenSearch Benchmark should use. ## workload Defines the workload to use based on the workload's name. You can find a list of preloaded workloads using `opensearch-benchmark list workloads`. `--workload-path` and `--workload-repository` as well as `--workload` are mutually exclusive. ## workload-params Defines which variables to inject into the workload. Variables injected must be available in the workload. To see which parameters are valid in the official workloads, select the workload from [the workloads repository](https://github.com/opensearch-project/opensearch-benchmark-workloads). ## test-procedure Defines a test procedure to use. You can find a list of test procedures by using `opensearch-benchmark list test-procedures`. ## test-execution-id Defines a unique ID for the test run. ## include-tasks Defines a comma-separated list of test procedure tasks to run. By default, all tasks listed in a test procedure array are run. Tests are executed in the order they are defined in `test-procedure`---not in the order they are defined in the command. All task filters are case sensitive. ## exclude-tasks Defines a comma-separated list of test procedure tasks not to run. ## baseline The baseline TestExecution ID used to compare the contender TestExecution. ## contender The TestExecution ID for the contender being compared to the baseline. ## results-format Defines the output format for the command line results, either `markdown` or `csv`. Default is `markdown`. ## results-number-align Defines the column number alignment for when the `compare` command outputs results. Default is `right`. ## results-file When provided a file path, writes the compare results to the file indicated in the path. ## show-in-results Determines whether or not to include the comparison in the results file. ## provision-config-repository Defines the repository from which OpenSearch Benchmark loads `provision-configs` and `provision-config-instances`. ## provision-config-revision Defines the specific Git revision in the `provision-config` that OpenSearch Benchmark should use. ## provision-config-path Defines the path to the `--provision-config-instance` and any OpenSearch plugin configurations to use. ## distribution-version Downloads the specified OpenSearch distribution based on version number. For a list of released OpenSearch versions, see [Version history](https://opensearch.org/docs/version-history/). ## distribution-repository Defines the repository from which the OpenSearch distribution should be downloaded. Default is `release`. ## provision-config-instance Defines the `--provision-config-instance` to use. You can view possible configuration instances by using the command `opensearch-benchmark list provision-config-instances`. ## provision-config-instance-params A comma-separated list of key-value pairs injected verbatim as variables for the `provision-config-instance`. ## target-hosts Defines a comma-separated list of host-port pairs that should be targeted if using the pipeline `benchmark-only`. Default is `localhost:9200`. ## target-os The target operating system (OS) for which the OpenSearch artifact should be downloaded. Default is the current OS. ## target-arch The name of the CPU architecture for which an artifact should be downloaded. ## revision Defines the current source code revision to use for running a benchmark test. Default is `current`. This command flag can use the following options: - `current`: Uses the source tree's current revision based on your OpenSearch distribution. - `latest`: Fetches the latest revision from the main branch of the source tree. You can also use a timestamp or commit ID from the source tree. When using a timestamp, specify `@ts`, where "ts" is a valid ISO 8601 timestamp, for example, `@2013-07-27T10:37:00Z`. ## opensearch-plugins Defines which [OpenSearch plugins]({{site.url}}{{site.baseurl}}/install-and-configure/plugins/) to install. By default, no plugins are installed. ## plugin-params Defines a comma-separated list of key-value pairs that are injected verbatim into all plugins as variables. ## runtime-jdk The major version of JDK to use. ## client-options Defines a comma-separated list of clients to use. All options are passed to the OpenSearch Python client. Default is `timeout:60`. ## load-worker-coordinator-hosts Defines a comma-separated list of hosts that coordinate loads. Default is `localhost`. ## enable-worker-coordinator-profiling Enables a performance analysis of OpenSearch Benchmark's worker coordinator. Default is `false`. ## pipeline The `--pipeline` option selects a pipeline to run. You can find a list of pipelines supported by OpenSearch Benchmark by running `opensearch-benchmark list pipelines`. ## telemetry Enables the provided telemetry devices when the devices are provided using a comma-separated list. You can find a list of possible telemetry devices by using `opensearch-benchmark list telemetry`. ## telemetry-params Enables setting parameters for telemetry devices. Accepts a list of comma-separated key-value pairs, each of which are delimited by a colon or a JSON file name. ## on-error Controls how OpenSearch Benchmark responds to errors. Default is `continue`. You can use the following options with this command flag: - `continue`: Continues to run the test despite the error. - `abort`: Aborts the test when an error occurs. ## preserve-install Keeps the Benchmark candidate and its index. Default is `false`. ## kill-running-processes When set to `true`, stops any OpenSearch Benchmark processes currently running and allows Benchmark to continue to run. Default is `false`. ## chart-spec-path Sets the path to the JSON files containing chart specifications that can be used to generate charts. ## chart-type Generates the indicated chart type, either `time-series` or `bar`. Default is `time-series`. ## output-path The name and path used for the chart's output. Default is `stdout`. ## limit Limits the number of search results for recent test runs. Default is `10`.