Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
271 lines (162 sloc) 13.5 KB

Developer's guide to Thoth

This document is from series of documents about project Thoth. The main goal of this document is to give a first touch on how to run, develop and use Thoth as a developer.

A prerequisite for this document are the following documents:

Preparing Developer's Environment

Use Ansible script git-clone-repos.yaml present in the Core repository and follow instructions present on the following page.

Once you finish cloning the GitHub repositories, the directory structure in your desired directory should state all the active repositories under the Thoth-Station organization on GitHub:

$ ls -A1 thoth-station/
adviser
amun-api
amun-client
amun-hwinfo
analyzer
...
user-api
workload-operator
zuul-test-config
zuul-test-jobs

These all are the repositories cloned to the most recent master branch (see also git-update-repos.yaml Ansible script to update repositories after some time).

Using Pipenv

All of the Thoth packages use Pipenv to create a separated and reproducible environment in which the given component can run. Almost every repository has its own Pipfile and Pipfile.lock file. The Pipfile file states direct dependencies for a project and Pipfile.lock file states all the dependencies (including the transitive ones) pinned to a specific version.

If you have cloned the repositories via the provided Ansible script, the Ansible scripts prepares the environment for you. It runs the following command to prepare a separate virtual environment with all the dependencies (including the transitive ones):

$ pipenv install --dev

As the environment is separated for each and every repository, you can now switch between environments that can have different versions of packages installed.

If you would like to install some additional libraries, just issue:

$ pipenv install <name-of-a-package>   # Add --dev if it is a devel dependency.

The Pipfile and Pipfile.lock file get updated.

If you would like to run a CLI provided by a repository, issue the following command:

# Run adviser CLI inside adviser/ repository:
$ cd adviser/
$ pipenv run ./thoth-adviser --help

The command above automatically activates separated virtual environment created for the thoth-adviser and uses packages from there.

To activate virtual environment permanently, issue:

$ pipenv shell
(adviser)$

Your shell prompt will change (showing that you are inside a virtual environment) and you can run for example Python interpret to run some of the Python code provided:

(adviser)$ python3
>>> from thoth.adviser import __version__
>>> print(__version__)

Developing cross-library features

As Thoth is created by multiple libraries which depend on each other, it is often desired to test some of the functionality provided by one library inside another.

Suppose you would like to run adviser with a different version of thoth-python package (present in the python/ directory). To do so, the only thing you need to perform is to run the thoth-adviser CLI (in adviser repo) in the following way:

$ cd adviser/
$ PYTHONPATH=../python pipenv run ./thoth-adviser provenance --requirements ./Pipfile --requirements-locked ./Pipfile.lock --files

The PYTHONPATH environment variable tells Python interpret to search for sources first in the ../python directory, this makes the following code:

from thoth.python import __version__

to first check sources present in ../python and run code from there (instead of running the installed thoth-python package from PyPI inside virtual environment).

If you would like to run multiple libraries this way, you need to delimit them using a colon:

$ cd adviser/
$ PYTHONPATH=../python:../common pipenv run ./thoth-adviser --help

Debugging application and logging

All Thoth components use logging that is implemented in the thoth-common package and is initialized in init_logging() function (defined in thoth-common library). This library setups all the routines needed for logging (also sending logs to external monitoring systems such as Sentry).

Besides the functionality stated above, the logging configuration can be adjusted based on environment variables. If you are debugging some parts of the Thoth application and you would like to get debug messages for a library, just set environment variable THOTH_LOG_<library name> to DEBUG (or any other log level you would like to see, so suppressing logs is also possible by setting log level to higher values like EXCEPTION or ERROR). An example of a run can be:

$ cd adviser/
$ THOTH_LOG_STORAGES=DEBUG THOTH_LOG_ADVISER=WARNING PYTHONPATH=../python pipenv run ./thoth-adviser provenance --requirements ./Pipfile --requirements-locked ./Pipfile.lock --files

The command above will suppress any debug and info messages in thoth-adviser (only warnings, errors and exceptions will be logged) and increases verbosity of thoth-storages package to DEBUG. Additionally, you can setup logging only for a specific file inside a package by using for example:

$ cd adviser/
$ THOTH_LOG_STORAGES_GRAPH_DGRAPH=DEBUG THOTH_LOG_ADVISER=WARNING PYTHONPATH=../python pipenv run ./thoth-adviser provenance --requirements ./Pipfile --requirements-locked ./Pipfile.lock --files

By exporting THOTH_LOG_STORAGES_GRAPH_DGRAPH environment variable, you set debug log level for file thoth/storages/graph/dgraph.py provided by thoth-storages package. This way you can debug and inspect behavior only for certain parts of application. If a file has underscore in its name, the environment variable has to have double underscores to explicitly escape it (not to look for a logger defined in a sub-package).

The default log level is set to INFO to all Thoth components.

Testing application against Ceph and graph database

If you would like test changes in your application against data stored inside Ceph, you can use the following command (if you have your gopass set up):

$ eval $(gopass show aicoe/thoth/ceph.sh)

This will inject into your environment Ceph configuration needed for adapters available in thoth-storages package and you can talk to Ceph instance.

In most cases you will need to set THOTH_DEPLOYMENT_NAME environment variable which distinguishes different deployments.

$ export THOTH_DEPLOYMENT_NAME=thoth-test-core

To browse data stored on Ceph, you can use awscli from PyPI utility which provides aws command (use aws s3 as Ceph exposes S3 compatible API).

To run applications against a graph database, export the following environment variables:

$ export DGRAPH_SERVICE_HOST=dgraph.thoth-station.ninja

If the Dgraph is serving requests on different port (not usual in Thoth deployments), you can specify also port by exporting:

# Default port on which Dgraph instances listen on:
$ export DGRAPH_SERVICE_PORT=9080

Deployments which are in stage/prod/test environment (or any other environment) are usually secured using TLS certificates which guard gRPC access to Dgraph instance. If you would like to communicate with sunch instance, you need to obtain TLS certificates and secure connection by exporting the following environment variable pointing to a directory which has all the certificates present:

$ export DGRAPH_TLS_PATH=tls/
$ ls tls/
ca.crt
ca.key
client.user.crt
client.user.key
node.crt
node.key

Running application inside OpenShift vs local development

All the libraries are designed to run locally (for fast developer's experience - iterating over features as fast as possible) as well as to run them inside a cluster.

If a library uses OpenShift's API (such as all the operators), the OpenShift class implemented in thoth-common library takes care of transparent discovery whether you run in the cluster or locally. If you would like to run applications against OpenShift cluster from your local development environment, use oc command to login into the cluster and change to project where you would like to operate in:

$ oc login <openshift-cluster-url>
...
$ oc project thoth-test-core

And run your applications (the configuration on how to talk to the cluster is picked from OpenShift's/Kubernetes config). You should see a courtesy warning by thoth-common that you are running your application locally.

To run an application from sources present in the local directory (for example with changes you have made), you can use the following command to upload sources to OpenShift and start a build:

$ cd adviser/
$ oc start-build adviser --from-dir=. -n <namespace>

You will see (for example in the OpenShift console) that the build was triggered from sources.

To see available builds (that match component name), issue the following once you are logged in and present in the right project:

$ oc get builds

If you would like to test application with unreleased packages inside OpenShift cluster, you can do so by installing package from a Git repo and running the oc build command above:

# To install thoth-common package from the master branch (you can adjust GitHub organization to point to your fork):
$ pipenv install 'git+https://github.com/thoth-station/common.git@master#egg=thoth-common'

After that, you can start build using oc start-build <build-name> --from-dir=. -n <namespace>. Note however that most of the Thoth's buildconfigs use Thoth to recommend application stacks. As you are using a Git version, this recommendation will fail with an error similar to this one:

thamos.swagger_client.rest.ApiException: (400)
Reason: BAD REQUEST
HTTP response headers: HTTPHeaderDict({'Server': 'gunicorn/19.9.0', 'Date': 'Tue, 13 Aug 2019 06:28:21 GMT', 'Content-Type': 'application/json', 'Content-Length': '45257', 'Set-Cookie': 'ae5b4faaab1fe6375d62dbc3b1efaf0d=3db7db180ab06210797424ca9ff3b586; path=/; HttpOnly'})
HTTP response body: {
  "error": "Invalid application stack supplied: Package thoth-storages uses a version control system instead of package index: {'git': 'https://github.com/thoth-station/storages' }",
}

To bypass this error you need to temporary turn off these recommendations by setting THOTH_ADVISE to 0 in the corresponding buildconfig:

oc edit bc <build-name> -n <namespace>

Please set the environment variable THOTH_ADVISE back to 1 after you test your changes.

Also not that files Pipfile and Pipfile.lock get updated. Please, do NOT commit such changes into repositories (we always rely on versioned packages).

Scheduling workload in the cluster

You can use your computer to directly talk to cluster and schedule workload there. An example case can be scheduling syncs of solver documents present on Ceph. To do so, you can go to user-api repo and run Python3 interpreter once your Python environment is set up:

$ # Go to a repo which has thoth-common and thoth-storages installed:
$ cd thoth-station/user-api
$ pipenv install --dev
$ # Log in to cluster - your credentials will be used to schedule workload:
$ oc login <cluster-url>
$ # Make sure you adjust secrets before running Python interpreter in storages environment - you can obtain them from gopass:
$ PYTHONPATH=. THOTH_MIDDLETIER_NAMESPACE=thoth-middletier-stage THOTH_INFRA_NAMESPACE=thoth-infra-stage KUBERNETES_VERIFY_TLS=0 THOTH_CEPH_SECRET_KEY="***" THOTH_CEPH_KEY_ID="***" THOTH_S3_ENDPOINT_URL=https://s3.url.redhat.com THOTH_CEPH_BUCKET_PREFIX=data/thoth THOTH_CEPH_BUCKET=thoth THOTH_DEPLOYMENT_NAME=thoth-core-upshift-stage pipenv run python3

After running the commands above, you should see Python interpreter's prompt:

>>> from thoth.storages import SolverResultsStore
>>> solver_store = SolverResultsStore()
>>> solver_store.connect()
>>> from thoth.common import OpenShift
>>> os = OpenShift()
Failed to load in cluster configuration, fallback to a local development setup: Service host/port is not set.
TLS verification when communicating with k8s/okd master is disabled
>>> all_solver_document_ids = solver_store.get_document_listing()
>>> (os.schedule_graph_sync_solver(solver_document_id, namespace="thoth-middletier-stage") for solver_document_id in all_solver_document_ids)

Once all the adapters get imported and instantiated, you can perform scheduling of workload using the OpenShift abstraction, which will directly talk to OpenShift's master to schedule workload in the cluster.

You can’t perform that action at this time.