You've arrived at the repo for the backend of the Host Based Inventory (HBI).If you're looking for API, integration or user documentation for HBI please see the [Inventory section in our Platform Docs site] (https://platform-docs.cloud.paas.psi.redhat.com/backend/inventory.html).
This project uses pipenv to manage the development and deployment environments. To set the project up for development, we recommend using pyenv to install/manage the appropriate python (currently 3.6.x), pip and pipenv version. Once you have pipenv, do the following:
pipenv install --dev
Afterwards you can activate the virtual environment by running:
pipenv shell
Included is a docker-compose file dev.yml that will start a postgres database that is
useful for development.
docker-compose -f dev.yml up
Run the following commands to run the db migration scripts which will maintain the db tables.
The database migration scripts determine the DB location, username, password and db name from the INVENTORY_DB_HOST, INVENTORY_DB_USER, INVENTORY_DB_PASS and INVENTORY_DB_NAME environment variables.
python manage.py db upgrade
By default the database container will use a bit of local storage so that data you enter will be persisted across multiple starts of the container. If you want to destroy that data do the following:
docker-compose down
It is possible to run the tests using pytest:
pytest --cov=.
Or you can run the tests individually:
./test_api.py
pytest test_db_model.py
./test_unit.py
pytest test_json_validators.py
Depending on the environment, it might be necessary to set the DB related environment variables (INVENTORY_DB_NAME, INVENTORY_DB_HOST, etc).
This project uses Sonar for static code analysis. To set up sonar-scanner on your machine, download the CLI and set the SONAR_PATH env var like this:
export SONAR_PATH=/path/to/sonar-scanner/lib/sonar-scanner-cli-${SONAR_CLI_VERSION_NUMBER}.jarThen, run sh sonar_scan.sh to generate the .sonar/sonar-scanner.properties file. Replace the $token$ placeholder at the bottom of the file with your own personal Sonar token. Instructions on how to do this are located here.
After making that replacement, run sh sonar_scan.sh again. It should run and upload to Red Hat's SonarQube instance, providing you a link to the results.
This repository uses pre-commit to check and enforce code style. It uses Black to reformat the Python code and Flake8 to check it afterwards. Other formats and text files are linted as well.
Install pre-commit hooks to your local repository by running:
$ pre-commit installAfter that, all your commited files will be linted. If the checks don’t succeed, the commit will be rejected. Please make sure all checks pass before submitting a pull request. Thanks!
Prometheus was designed to run in a multi-threaded environment whereas gunicorn uses a multi-process architecture. As a result, there is some work to be done to make prometheus integrate with gunicorn.
A temp directory for prometheus needs to be created before the server starts. The prometheus_multiproc_dir environment needs to point to this directory. The contents of this directory need to be removed between runs.
A command to run the server in a cluster.
gunicorn -c gunicorn.conf.py run
Running the server locally for development. In this case it’s not necessary to care about the Prometheus temp directory or to set the prometheus_multiproc_dir environment variable. This is done automatically.
python run_gunicorn.py
prometheus_multiproc_dir=/path/to/prometheus_dir
APP_NAME="inventory"
PATH_PREFIX="/r/insights/platform"
INVENTORY_DB_USER="insights"
INVENTORY_DB_PASS="insights"
INVENTORY_DB_HOST="localhost"
INVENTORY_DB_NAME="insights"
INVENTORY_DB_POOL_TIMEOUT="5"
INVENTORY_DB_POOL_SIZE="5"
INVENTORY_LOGGING_CONFIG_FILE=logconf.ini
INVENTORY_DB_SSL_MODE=""
INVENTORY_DB_SSL_CERT=""
To force an ssl connection to the db set INVENTORY_DB_SSL_MODE to "verify-full" and provide the path to the certificate you'd like to use.
It is necessary to pass an authentication header along on each call to the service. For testing purposes, it is possible to set the required identity header to the following:
x-rh-identity: eyJpZGVudGl0eSI6IHsiYWNjb3VudF9udW1iZXIiOiAiMDAwMDAwMSIsICJ0eXBlIjogIlVzZXIiLCAiaW50ZXJuYWwiOiB7Im9yZ19pZCI6ICIwMDAwMDEifX19Cg==
This is the Base64 encoding of the following JSON document:
{"identity": {"account_number": "0000001", "type": "User", "internal": {"org_id": "000001"}}}The above header has the "User" identity type, but it's possible to use a "System" type header as well.
x-rh-identity: eyJpZGVudGl0eSI6IHsiYWNjb3VudF9udW1iZXIiOiAiMDAwMDAwMSIsICJ0eXBlIjogIlN5c3RlbSIsICJpbnRlcm5hbCI6IHsib3JnX2lkIjogIjAwMDAwMSJ9fX0K
This is the Base64 encoding of the following JSON document:
{"identity": {"account_number": "0000001", "type": "System", "internal": {"org_id": "000001"}}}If you want to encode other JSON documents, you can use the following command:
echo '{"identity": {"account_number": "0000001", "type": "System", "internal": {"org_id": "000001"}}}' | base64In case one needs to do this:
export INVENTORY_LEGACY_API_URL="/r/insights/platform/inventory/api/v1"
The inventory service has been integrated with the Payload Tracker service. The payload tracker integration can be configured using the following environment variables:
KAFKA_BOOTSTRAP_SERVERS=localhost:29092
PAYLOAD_TRACKER_KAFKA_TOPIC=platform.payload-status
PAYLOAD_TRACKER_SERVICE_NAME=inventory
PAYLOAD_TRACKER_ENABLED=true
The payload tracker can be disabled by setting the PAYLOAD_TRACKER_ENABLED environment variable to false. The payload tracker will also be disabled for add/delete operations that do not include a request_id. When the payload tracker is disabled, a NullObject implementation (NullPayloadTracker) of the PayloadTracker interface is used. The NullPayloadTracker implements the PayloadTracker interface but the methods are no-op methods.
The PayloadTracker purposefully eats all exceptions that it generates. The exceptions are logged. A failure/exception within the PayloadTracker should not cause a request to fail.
The payload status is a bit "different" due to each "payload" potentially containing multiple hosts. For example, the add_host operation will only log an error for the payload if the entire payload fails (catastrophic failure during processing...db down, etc). One or more of the hosts could fail during the add_host method. These will get logged as a "processing_error". If a host is successfully added/updated, then it will be logged as a "processing_success". Having one or more hosts get logged as "processing_error" will not cause the payload to be flagged as "error" overall.
The payload tracker status logging for the delete operation is similar. The overall status of the payload will only be logged as an "error" if the entire delete operation fails (a 404 due to the hosts not existing, db down, etc).
Run this command to generate a new revision in migrations/versions
python manage.py db revision -m "Description of revision"
Inventory uses Source-To-Image (S2I) for building of container images. The container image is built by running
s2i build . centos/python-36-centos7 inventory -e ENABLE_PIPENV=true
In addition, a Dockerfile is provided. The container image built this way is only intended for development purposes (e.g. orchestrating the container using docker-compose) and must not be used for production deployment.
docker build . -f dev.dockerfile -t inventory:dev
By default, the container runs the database migrations and then starts the inventory-mq service.
The application provides some management information about itself. These endpoints are exposed at the root path / and thus are accessible only from inside of the deployment cluster.
- /health responds with 200 to any GET requests, point your liveness or readiness probe here.
- /metrics offers metrics and monitoring intended to be pulled by Prometheus.
- /version responds with a json doc that contains the build version info (the value of the OPENSHIFT_BUILD_COMMIT environment variable)
Cron jobs push their metrics to a Prometheus Pushgateway instance running at PROMETHEUS_PUSHGATEWAY. Defaults to localhost:9091.
This section describes the process of getting a code change from a pull request all the way to production.
It all starts with a pull request. When a new pull request is opened the following checks are run automatically:
- database migrations
- code style checks
- unit tests
Should any of these fail this is indicated directly on the pull request.
When all of these checks pass and a reviewer approves the changes the pull request can be merged by someone from the @RedHatInsights/host-based-inventory-committers team.
When a pull request is merged to master a new container image is built and tagged as insights-inventory:latest. The latest image is deployed to the CI environment automatically. This is announced in #platform-inventory-standup slack channel as Inventory release pipeline (step 1)
Afterwards, the smoke tests are started automatically. These consist of a subset of IQE tests that cover inventory, advisor, engine and the ingress service.
Once the smoke test run finishes the results are reported in #platform-inventory-standup slack channel as Inventory release pipeline (step 2)
If the smoke tests pass, the container image is re-tagged as insights-inventory:stable
In certain situations it may be desired for an image to stay deployed in CI only for some time before proceeding with the QA deployment.
An example would be a possibly breaking change where we want to give integrators some time to test the integration in CI first.
This can be achieved by setting the promote-stable key to false in the inventory-pipeline config map.
This change needs to be made before the given commit is merged to master.
Vortex is a process that picks stable images of platform components and tests them together. The process is automated and runs periodically. By default the vortex job runs around every two hours.
Once the vortex tests are finished the stable inventory image is tagged as insights-inventory:qa and deployed to the QA environment. This is announced in #platform-inventory-standup slack channel as Inventory release pipeline (step 3).
Once the stable image lands in the QA environment, the QE testing can begin.
People in @platform-inventory-qe Slack group handle this part. They trigger the QE testing jobs [1][2]. Optionally, manual testing may be performed to validate certain features / bug fixes.
Once all of this is finished, a @platform-inventory-qe representative (manually) reports the results in the #platform-inventory-standup channel.
This step can be performed in parallel with step 4.
In order to promote a new image to the stage environment it is necessary to update the deploy.yml file.
The IMAGE_TAG parameter needs to be updated to the current PROMO_CODE (truncated to first 7 characters).
This applies to the following components:
- insights-inventory-reaper
- insights-inventory-mq-service
- insights-inventory
The IMAGE_TAG parameter should be updated for the stage namespace for each of the aforementioned components.
Note that insights-host-delete, which uses a different image, should not be updated.
There is a script utils/deploy.py that can be used to automatically update the IMAGE_TAG. Possible usage using
sponge from moreutils:
$ pipenv run python utils/deploy.py -s PROMO_CODE < ../app-interface/data/services/insights/host-inventory/deploy.yml | sponge ../app-interface/data/services/insights/host-inventory/deploy.ymlOnce the change has been made, submit a merge request to app-interface. For the CI pipeline to run tests on your fork, add @devtools-bot as a Maintainer. See this guide on how to do that.
After the MR has been opened, get somebody from AppSRE/insights-host-inventory to approve the MR by adding a /lgtm comment.
Afterwards, the MR is merged automatically and changes are deployed to the stage environment.
Once that happens, contact @platform-inventory-qe and request the image to be tested in the stage environment.
Once @platform-inventory-qe finishes testing of the image in both qa and stage environments (i.e. steps 4 and 5 are completed) they propose a release to production.
They do so by opening a MR in app-interface to update the IMAGE_TAG in the production namespace to the PROMO_CODE.
The same steps to create a MR to app-interface are followed as defined in step 5 except that the IMAGE_TAG parameter for the prod (instead of stage) namespace is changed.
$ pipenv run python utils/deploy.py -p PROMO_CODE < ../app-interface/data/services/insights/host-inventory/deploy.yml | sponge ../app-interface/data/services/insights/host-inventory/deploy.ymlOnce the MR has been created, an engineer, who is a member of AppSRE/insights-host-inventory and has been trained to perform inventory releases, approves the MR by adding a /lgtm comment.
The engineer is then responsible for monitoring of the rollout of the new image.
It is essential to monitor the health of the service during and after the production deployment. A non-exhaustive list of things to watch includes:
- Inventory deployments in platform-prod OSD namespace
- primarily ensure that the new pods are spun up properly
- #team-insights-alerts Slack channel for any inventory-related Prometheus alerts
- Inventory Dashboard for any anomalies such as error rate or consumer lag
- Kibana for any error-level log entries
Should unexpected problems occur during the deployment, it is possible to do a rollback.
This is done by updating the IMAGE_TAG parameters in deploy.yml to point to a previous PROMO_CODE.