We have devised our own test framework to test the spec against an OpenSearch cluster. We're still adding more features to the framework as the needs arise, and this document will be updated accordingly. This test framework has also been integrated into the repo's CI/CD pipeline. Checkout the test-spec workflow for more details.
Download and install the latest version of Node.js and npm from here and run npm install.
Install Docker Desktop from here.
Set up an OpenSearch cluster with Docker:
(Replace <<your_password>> with your desired password. If not provided, the default password inside the docker-compose.yml file will be used.)
export OPENSEARCH_PASSWORD=<<your_password>>
cd tests/default
docker compose up -dRun the tests (use --opensearch-insecure for a local cluster running in Docker that does not have a valid SSL certificate):
export OPENSEARCH_PASSWORD=<<your_password>>
npm run test:spec -- --opensearch-insecureRun a specific test story:
npm run test:spec -- --opensearch-insecure --tests tests/default/_core/info.yamlVerbose output:
npm run test:spec -- --opensearch-insecure --verboseWant to help with some missing tests? Choose from the remaining paths in the test coverage report:
npm run test:spec -- --opensearch-insecure --coverage-reportUse an Amazon OpenSearch service instance.
export AWS_ACCESS_KEY_ID=<<your AWS access key ID>>
export AWS_SECRET_ACCESS_KEY=<<your AWS secret access key>>
export AWS_SESSION_TOKEN=<<optional AWS session token>>
export AWS_REGION=us-west-2
export OPENSEARCH_URL=https://....us-west-2.es.amazonaws.com
npm run test:specRemember to set the OPENSEARCH_PASSWORD or AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables every time you start a new shell to run the tests.
The cluster is most likely hitting a disk watermark threshold. This example sets the disk watermark thresholds to 1500MB low, 100MB high, and 500MB flood stage, allowing the cluster to create indices even if the disk is almost full.
curl -k -X PUT --user "admin:${OPENSEARCH_PASSWORD}" https://localhost:9200/_cluster/settings -H 'Content-Type: application/json' -d'
{
"persistent": {
"cluster.routing.allocation.disk.watermark.low": "1500mb",
"cluster.routing.allocation.disk.watermark.high": "1000mb",
"cluster.routing.allocation.disk.watermark.flood_stage": "500mb",
"cluster.blocks.create_index" : null
}
}
'The cluster is not loading plugins correctly, maybe it was stopped using docker kill instead of docker stop. Recreating the cluster should fix the issue: docker compose up --force-recreate -d.
The spec tests reside in the tests/ directory. Tests are organized in suites (default, etc.), and subsequently in folders that match namespaces. For example, tests for APIs defined in spec/namespaces/indices.yaml can be found in tests/default/indices/index.yaml (for /{index}), and tests/default/indices/doc.yaml (for /{index}/_doc).
Additional suites require custom configuration that is defined in a separate docker-compose.yml. For example tests/plugins/index_state_management/docker-compose.yml uses a custom setting of plugins.index_state_management.job_interval=1 to cause the /_nodes API to return plugin information tested in tests/plugins/index_state_management/nodes/plugins/index_state_management.yaml.
Each yaml file in the tests directory represents a test story that tests a collection of related operations.
A test story has 3 main components:
- prologues: These are the operations that are executed before the test story is run. They are used to set up the environment for the test story.
- chapters: These are the operations that are being tested.
- epilogues: These are the operations that are executed after the test story is run. They are used to clean up the environment after the test story.
Check the test_story JSON Schema for the complete structure of a test story.
Below is the simplified version of the test story that tests the index operations:
$schema: ../../json_schemas/test_story.schema.yaml # The schema of the test story. Include this line so that your editor can validate the test story on the fly.
description: This story tests all endpoints relevant the lifecycle of an index, from creation to deletion.
prologues: [] # No prologues are needed for this story.
epilogues: # Clean up the environment by assuring that the `books` index is deleted afterward.
- path: /books
method: DELETE
status: [200, 404] # The index may not exist, so we accept 404 as a valid response. Default to [200, 201] if not specified.
chapters:
- synopsis: Create an index named `books` with mappings and settings.
path: /{index} # The test will fail if "PUT /{index}" operation is not found in the spec.
method: PUT
parameters: # All parameters are validated against their schemas in the spec.
index: books
request: # The request.
headers: # Optional headers.
user-agent: OpenSearch API Spec/1.0
payload: # The request body is validated against the schema of the requestBody in the spec.
mappings:
properties:
name:
type: keyword
age:
type: integer
settings:
number_of_shards: 5
number_of_replicas: 2
response: # The response.
payload: # Matching response payload. The entire payload is validated against the schema of the corresponding response in the spec.
status: 200 # This is the expected status code of the response. Any other status code will fail the test.
- synopsis: Retrieve the mappings and settings of the `books` index.
path: /{index}
method: GET
parameters:
index: books
flat_settings: true
- synopsis: Delete the `books` index.
path: /{index}
method: DELETE
parameters:
index: booksConsider the following chapters in ml/model_groups test story:
- synopsis: Create model group.
id: create_model_group # Only needed if you want to refer to this chapter in another chapter.
path: /_plugins/_ml/model_groups/_register
method: POST
request:
payload:
name: NLP_Group
description: Model group for NLP models.
response:
status: 200
output: # Save the model group id for later use.
test_model_group_id: "payload.model_group_id"
- synopsis: Query model group.
path: /_plugins/_ml/model_groups/{model_group_id}
method: GET
parameters:
# Use the output from the `create_model_group` chapter.
model_group_id: ${create_model_group.test_model_group_id}
response:
status: 200
- synopsis: Delete model group.
path: /_plugins/_ml/model_groups/{model_group_id}
method: DELETE
parameters:
# Use the output from the `create_model_group` chapter.
model_group_id: ${create_model_group.test_model_group_id}
response:
status: 200As you can see, the output field in the first chapter saves the model_group_id from the response body. This value is then used in the subsequent chapters to query and delete the model group.
You can also supply defaults for output values, e.g. for payload._version used in cluster/routing/awareness/weights.yaml.
version:
path: payload._version
default: -1
You can reuse output in payload expectations. See tests/plugins/index_state_management/nodes/plugins/index_state_management.yaml for an example.
It's common to add a feature to the next version of OpenSearch. When adding a new API in the spec, make sure to specify x-version-added, x-version-deprecated or x-version-removed. Finally, specify a semver or a semver range in your test stories or chapters as follows.
- synopsis: Search with `phase_took` added in OpenSearch 2.12 and removed in version 3.
version: '>=2.12 <3'
path: /{index}/_search
parameters:
index: movies
cancel_after_time_interval: 10s
method: POST
response:
status: 200The test tool will fetch the server version when it starts and use it automatically. The integration test workflow runs a matrix of OpenSearch versions, including the next version. Please check whether the workflow needs an update when adding version-specific tests.
OpenSearch consists of plugins that may or may not be present in various distributions. When adding a new API in the spec, you can specify x-distributions-included or x-distributions-excluded with a list of distributions that have a particular feature. For example, the Amazon Managed OpenSearch supports GET /, but Amazon Serverless OpenSearch does not.
/:
get:
operationId: info.0
x-distributions-included:
- opensearch.org
- amazon-managed
x-distributions-excluded:
- amazon-serverless
description: Returns basic information about the cluster.Similarly, skip tests that are not applicable to a distribution by listing the distributions that support or do not support it.
description: Test root endpoint.
distributions:
included:
- amazon-managed
- opensearch.org
excluded:
- amazon-serverless
chapters:
- synopsis: Get server info.
path: /
method: GET
response:
status: 200To test a particular distribution pass --opensearch-distribution to the test tool. For example, the following runs tests against an Amazon Managed OpenSearch instance.
export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...
export AWS_SESSION_TOKEN=...
export AWS_REGION=us-west-2
export OPENSEARCH_URL=https://....us-west-2.es.amazonaws.com
npm run test:spec -- --opensearch-distribution=amazon-managedThe output will visible skip APIs that are not available in the amazon-managed distribution.
PASSED _core/bulk.yaml (.../_core/bulk.yaml)
PASSED _core/info.yaml (.../_core/info.yaml)
SKIPPED indices/forcemerge.yaml (Skipped because distribution amazon-managed is not opensearch.org.)
Some APIs behave asynchronously and may require a test to wait for a task to complete. This can be achived with a combination of payload and retry.
For example, an ML task returns CREATED when created, and COMPLETED when it's done. The example below will retry 3 times with an interval of 30 seconds until the task is complete. The default wait time is 1s.
- synopsis: Wait for task.
path: /_plugins/_ml/tasks/{task_id}
method: GET
parameters:
task_id: ${create_model.task_id}
response:
status: 200
payload:
state: COMPLETED
retry:
count: 3
wait: 30000The test runner expects all tests in the same file to be variation of the same path in order to keep tests well-organized. Otherwise, a warning will be emitted.
WARNING Multiple paths detected, please group similar tests together and move paths not being tested to prologues or epilogues.
/_component_template/{name}
/_index_template/{name}
/{index}
The test runner may generate warnings that can be suppressed with warnings: at story or chapter level. For example, to suppress the multiple paths detected warning.
- synopsis: Create an index.
method: PUT
path: /{index}
parameters:
index: movies
- synopsis: Search the index to make sure it has been created.
method: POST
warnings:
multiple-paths-detected: false
path: /{index}/_search
parameters:
index: moviesThe test tool can generate a test coverage summary using --coverage <path> with the number of evaluated verb + path combinations, a total number of paths and the % of paths tested.
{
"evaluated_operations_count": 214,
"operations_count": 550,
"evaluated_paths_pct": 38.91
}The report is then used by the test-spec.yml workflow, uploaded with every run, combined across various versions of OpenSearch, and reported as a comment on each pull request.
The test tool can display detailed and hierarchal test coverage with --coverage-report. This is useful to identify untested paths. The report produces the following output with the missing ones.
/_alias (4)
GET /_alias
/{name} (3)
GET /_alias/{name}
POST /_alias/{name}
HEAD /_alias/{name}