Stac-server is an implementation of the STAC API specification for searching and serving metadata for geospatial data, including but not limited to satellite imagery). The STAC and STAC API versions supported by a given version of stac-server are shown in the table below. Additional information can be found in the CHANGELOG
stac-server Version | STAC Version | STAC API Version |
---|---|---|
0.1.x | 0.9.x | 0.9.x |
0.2.x | <1.0.0-rc.1 | 0.9.x |
0.3.x | 1.0.0 | 1.0.0-beta.2 |
0.4.x | 1.0.0 | 1.0.0-beta.5 |
The following APIs are deployed instances of stac-server:
| Name | STAC Version | STAC API Version | Description | | ----------------------------------------------------------------------------------------- | ------------ | ---------------- | | Earth Search | 1.0.0-beta.2 | 0.9.0 | Catalog of some AWS Public Datasets | | Landsat Look | 1.0.0 | 0.9.0 | |
flowchart LR
itemsForIngest[Items for ingest]
subgraph ingest[Ingest]
ingestSnsTopic[Ingest SNS Topic]
ingestQueue[Ingest SQS Queue]
ingestLambda[Ingest Lambda]
ingestDeadLetterQueue[Ingest Dead Letter Queue]
failedIngestLambda[Failed Ingest Lambda]
end
users[Users]
subgraph api[STAC API]
apiGateway[API Gateway]
apiLambda[API Lambda]
end
elasticsearch[(Elasticsearch)]
%% Ingest workflow
itemsForIngest --> ingestSnsTopic
ingestSnsTopic --> ingestQueue
ingestQueue --> ingestLambda
ingestLambda --> elasticsearch
ingestDeadLetterQueue --> failedIngestLambda
%% API workflow
users --> api
apiGateway --> apiLambda
apiLambda --> elasticsearch
Create a new deployment, copy the elasticsearch database, rename indexes,
The Serverless Framework supports provisioning AWS resources, but it does not support updating existing resources. In 0.4, the default Elasticsearch version has been updated from 7.9 to 7.10. Continuing to use 7.9 should not cause any problems, but it recommended that you manually upgrade to 7.10 by going to AWS Console - Amazon OpenSearch Service, choosing the Elasticsearch domain used by your stac-server deployment (e.g., stac-server-{stage}-es), choose Upgrade from the Actions menu, and then upgrade to Elasticsearch 7.10.
It is now recommended to disable automatic index creation.
Elasticsearch indices each have a mapping applied that determines how the data is indexed and searched over.
These mappings do not change the document data, but can change search behavior. One relevant mapping
behavior is that by default, string fields are analyzed for full-text search. In most cases with STAC Items,
values such as those in the id
and collection
fields should not be analyzed and should instead be searchable only
by exact matches. In Elasticsearch, this is known as a keyword
field type. Importantly, sorting may only be done over keyword
typed fields. As of 0.4.0, the default sort is now by properties.datetime
, then id
, then collection
, and results will not be returnd if any indicies have the id
or collection
fields mapped as text
instead of keyword
.
For each index (other than collections
), use GET to retrieve the endpoint GET /{collectionId}/_mapping
, and
validate that properties.datetime
type is date
, and id
and collection
mappings are keyword
(not text
with a keyword
subfield). For an AWS Opensearch Service instance, this can be done with a script similar to the one here.
The results should look simliar to this:
{
"my_collection_name": {
"mappings": {
"dynamic_templates": [
...
{
"strings": {
"match_mapping_type": "string",
"mapping": {
"type": "keyword"
}
}
},
...
],
"properties": {
....
"id": {
"type": "keyword"
},
"collection": {
"type": "keyword"
},
....
"properties": {
"properties": {
...
"datetime": {
"type": "date"
},
...
}
},
...
}
}
}
}
If this is not the case, the easiest solution to fix it is to:
- Deploy a 0.4.0 instance.
- Backup and restore the 0.3.0 instance's Elasticsearch indicies to the 0.4.0 instances's Elasticsearch database.
- Create a collection via ingest with a new collection name similar to the existing one (e.g., if index foo exists, create foo_new).
- Reindex from the the existing index (foo) to the the new one (foo_new).
- Delete the exiting index and rename the new one to the name of the formerly-existing one (e.g. foo_new -> foo).
Stac-server is a web API that returns JSON, see the documentation, or the /api endpoint which is a self-documenting OpenAPI document. Here are some additional tools that might prove useful:
- pystac-client: A Python client library and CLI for searching a STAC compliant API
This repository contains Node libraries for running the API, along with a serverless configuration file for deployment to AWS.
To create your own deployment of stac-server, first clone the repository:
git clone https://github.com/stac-utils/stac-server.git
cd stac-server
Copy the example serverless config file to a file named serverless.yml
:
cp serverless.yml.example serverless.yml
There are some settings that should be reviewed and updated as needeed in the serverless config file, under provider->environment:
Name | Description | Default Value |
---|---|---|
STAC_VERSION | STAC Version of this STAC API | 1.0.0 |
STAC_ID | ID of this catalog | stac-server |
STAC_TITLE | Title of this catalog | STAC API |
STAC_DESCRIPTION | Description of this catalog | A STAC API |
STAC_DOCS_URL | URL to documentation | https://stac-utils.github.io/stac-server |
ES_BATCH_SIZE | Number of records to ingest in single batch | 500 |
LOG_LEVEL | Level for logging (CRITICAL, ERROR, WARNING, INFO, DEBUG) | INFO |
STAC_API_URL | The root endpoint of this API | Inferred from request |
ENABLE_TRANSACTIONS_EXTENSION | Boolean specifying if the Transaction Extension should be activated | false |
After reviewing the settings, build and deploy:
npm install
npm run build
npm run deploy
This will create a CloudFormation stack in the us-west-2
region called stac-server-dev
.
To change the region or the stage name (from dev
) provide arguments to the deploy command
(note the additional --
in the command, required by npm
to provide arguments):
npm run deploy -- --stage mystage --region eu-central-1
Once deployed, there are a few steps to configure Elasticsearch.
It is recommended to disable the automatic index creation. This prevents the situation where a group of Items are bulk indexed before the Collection in which they are contained has been created, and an Elasticsearch index is created without the appropriate mappings.
This requires installing the requests, requests_aws4auth, and boto3 python libraries, for example, with:
pip install requests requests_aws4auth boto3
Then putting this code into a python file an running it:
from requests_aws4auth import AWS4Auth
import boto3
import requests
host = 'https://my-test-domain.us-east-1.es.amazonaws.com'
path = '/_cluster/settings'
region = 'us-west-2'
credentials = boto3.Session().get_credentials()
awsauth = AWS4Auth(credentials.access_key, credentials.secret_key, region, 'es', session_token=credentials.token)
r = requests.put(
f'{host}{path}',
auth=awsauth,
json={
"persistent": {
"action.auto_create_index": "false"
}
})
print(r.status_code)
print(r.text)
The collection
index must be created, which stores the metadata about each Collection.
Invoke the stac-server-<stage>-ingest
Lambda function with a payload of:
{
"create_indices": true
}
This can be done with the AWS CLI with (the final -
parameter pipes the output to stdout):
aws lambda invoke \
--function-name stac-server-dev-ingest \
--cli-binary-format raw-in-base64-out \
--payload '{ "create_indices": true }' \
-
Stac-server is now ready to ingest data!
STAC Collections and Items are ingested by the ingest
Lambda function, however this Lambda is not invoked directly by a user, it consumes records from the stac-server-<stage>-queue
SQS. To add STAC Items or Collections to the queue, publish them to the SNS Topic stac-server-<stage>-ingest
.
STAC Collections must be ingested before Items that belong to that Collection. Items should have the collection
field populated with the ID of an existing Collection. If an Item is ingested before ingestion of the Collection it contains,
ingestion will either fail (in the case of a single Item ingest) or if auto-creation of indexes is enabled (default) and multiple Items are ingested in bulk, the auto-created index will have incorrect mappings.
If a collection or item is ingested, and an item with that id already exists in STAC, the new item will completely replace the old item.
There is a 256 KB limit on the size of SQS messages. Larger items can by publishing a message to the stac-server-<stage>-ingest
SNS topic in with the format:
{
"href": "s3://source-bucket/source-key"
}
The s3://
, http://
, and https://
protocols are supported for remote ingest.
Stac-server can also be subscribed to SNS Topics that publish complete STAC Items as their message. This provides a way to keep stac-server up to date with new data. Use the AWS Lambda console for the function stac-server-<stage>-subscibe-to-sns
to subscribe to an SNS Topic for which you have the full ARN and permission to subscribe to. This could be an SNS Topic you created yourself to publish STAC records to, or a publicly available one, such as for Sentinel.
Note, that adding the subscription via the topic page does not seem to work. Instead, add a trigger on Lambda edit page.
Errors that occur during ingest will end up in the dead letter processing queue, where they are processed by the stac-server-<stage>-failed-ingest
Lambda function. Currently all the failed-ingest Lambda does is log the error, see the CloudWatch log /aws/lambda/stac-server-<stage>-failed-ingest
for errors.
Install NVM to manage your Node.js environment.
# uses version in .nvmrc
nvm install
nvm use
The package-lock.json was built with npm 8.5.0, so use at least this version.
There are several useful npm commands available to use locally:
# Install dependencies in package.json
npm install
# Run the build command in each of the packages (runs webpack)
npm run build
# Run ESLint
npm run lint
# To run tests for all packages
npm run test
# To build API docs from the api spec
npm run build-api-docs # TODO: this fails
Before the API can be run, Elasticsearch and Localstack need to be running. There is a docker-compose.yml
file to simplify running Elasticsearch locally:
docker-compose up -d
The API can then be run with:
npm run serve
Connect to the server on http://localhost:3000/
Other configurations can be passed as shell environment variables, e.g.,
export ENABLE_TRANSACTIONS_EXTENSION=true
export ES_HOST='https://search-stac-server-dev-es-7awl6h344qlpvly.us-west-2.es.amazonaws.com'
npm run serve
stac-server uses ava to execute tests.
# alias to run unit tests
npm test
# run unit tests in tests directory
npm run test:unit
# run unit tests with coverage
npm run test:coverage
# run tests from a single test file whose titles match 'foobar*'
npx ava tests/test-es.js --match='foobar*'
The System and Integration tests use an Elasticsearch server running in Docker and a local instance of the API.
When the system tests run, they:
- Wait for Elasticsearch to be available
- Delete all indices from Elasticsearch
- Start an instance of the API. That API will be available at http://localhost:3000/dev/
- Wait for the API to be available
- Run the system tests in
./tests/system/test-*.js
- Stop the API
Before running the system tests, make sure to start Elasticsearch using:
docker-compose up -d
Running these tests requires the timeout utility is installed. On Linux,
this is probably already installed, and on macOS it can be installed with brew install coreutils
.
Once Elasticsearch has been started, run the system tests:
npm run test:system
A subset of system tests may be run by providing a glob matching the test files to run:
npm run test:system test-api-item-*
Run the integration tests (Note: currently none exist):
npm run test:integration
The OpenAPI specification is served by the endpoint /api
.
This file is location in src/lambdas/api/openapi.yaml.
When the API is updated to a new STAC API release, this file must be updated. To update it, first install yq, then run:
bin/build-openapi.sh
This script combines all of the STAC API OpenAPI definitions for each conformance class into one file.
Next, edit that file to make it specific to this server. For example:
- edit to change the title from
STAC API - Item Search
to justSTAC API
- remove all of the Filter Extension references
- Fix each endpoint, especially the Landing Page defintion, which gets duplicated
- Add definitions for each tag
To validate the resulting OpenAPI file, run
npm run check-openapi
and fix any errors or warnings.
stac-server was forked from sat-api. Stac-server is for STAC versions 0.9.0+, while sat-api exists for versions of STAC prior to 0.9.0.