Skip to content

Changelog

All user-facing, notable changes will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[2.34.0] - 2024-11-12

Added

  • Internal calls of Pub service are cached for a short configurable time of 1-10 minutes. This improves performance and robustness in case of issues with Lifecycle, network or a database. (#213)

Changed

  • Upgraded third-party dependencies.
  • Enhanced the proxying of requests from Dashboard to Lifecycle in order to display detailed errors on the Dashboard (#540)

Fixed

  • Fixed Duplicate Prepared Statement in psycopg connector (#536)

[2.33.0] - 2024-10-01

Changed

  • Most database operations are now handled by the new database access layer, making direct SQL queries. It has a configurable connection pool that reduces the number of open connections. (#480)
  • Maintenance mode now also prohibits making calls to the Jobs. If turned on, it responds with "503 Service Unavailable". (#517)
  • Job's secrets are mounted when building the image. This hides the secret environment variables, so they can't be accessed after the build. (#474)

Removed

  • Support for OpenTelemetry exporters has been dropped.

Fixed

  • Fixed deployment error about missing warnings. (#523)
  • Added informative error about crash during Job initialization. (#469)
  • Resolved compatibility issues with the Racetrack client for Python 3.8 related to unsupported typing operands. (#527)

[2.32.1] - 2024-08-26

Added

  • sslmode and sslrootcert is supported in database connector for PostgreSQL (#509)
  • Deprecated fields in the manifest trigger a warning. (#482)

Changed

  • "404 Not Found" responses are treated as errors in Async Job calls. (#508)
  • Attempting to install multiple versions of the same infrastructure plugin causes an error. (#489)

[2.32.0] - 2024-08-19

Added

  • A new tab "External Consumers" has been added to the Dashboard. It allows to browse, read auth token and create a new ESC. The tab is only available for administrator users. (#484)
  • "External Consumers" can have multiple auth tokens to facilitate the process of rotating tokens. Tokens can be created and revoked from the Dashboard. Dashboard keeps track of the date when each token was in-use last time. (#487)

Changed

  • Reconciliation loop is now turned off by default. It can be enabled in configuration, if needed. Instead, reconciliation can be triggered manually through the Dashboard, at Administration tab. (#497)
  • Jobs are no longer deleted on unsuccessful, retried deployment. (#491)
  • Pub's metrics has been revised to detect overloaded jobs easier. (#493)

Fixed

  • Download button works on Kubernetes clusters. (#490)

[2.31.0] - 2024-07-09

Added

  • Lifecycle now provides more metrics to monitor its health. In particular, it tracks the number of established database connections and the duration of fetching Job data from the database. To improve logging, Lifecycle now displays more details about cancelled requests. (#472)
  • Support deployment warnings, such as #452. Internal racetrack components can now send "log" warnings for a deployment. (#473)

Changed

  • The Pub-Lifecycle client now has a timeout of 30 seconds.
  • The HTTP thread pool of Lifecycle has been increased to 60 threads. This setting is configurable through the config file.

Fixed

  • Fix portfolio UI sorting bug. (#479)

[2.30.0] - 2024-06-10

Added

  • Submitted manifest YAML is compared with the one in the job's repository. In case of incosistencies, the deployment is aborted. This only happens if the optional setting verify_manifest_consistency is turned on in the Image Builder. (#452)
  • Infrastructure plugins can now provide its own statistics for every job. These key-value statistics are displayed in Job details on a Dashboard. (#77)
  • Plugins added to racetrack can now be downloaded via the CLI or the admin interface. (#451)

Changed

  • The job_types method's format for plugins has been changed to support providing Dockerfiles directly by the Jobs. See Developing plugins guide. Previous format is still supported for backward compatibility. (#470)

[2.29.3] - 2024-05-21

Added

  • Racetrack is compatible with Python 3.12 (#463)

Changed

  • "Connection refused" error is now retriable in async job calls (will be retried automatically). (#459)

Fixed

  • Long deployments (over 20 minutes) no longer get stuck in the final stage. (#448)

[2.29.2] - 2024-04-30

Fixed

  • Fixed overwriting STARTING job status in a sync loop. (#457)

[2.29.1] - 2024-04-26

Fixed

  • Fixed typing error in racetrack CLI. (#455)

[2.29.0] - 2024-04-25

Added

  • Image builder keeps track of the duration of each phase of image building in its metrics, including: preparing job type builder, fetching the source code, building image and pushing image phases. (#453)

Changed

  • When building a Job, Racetrack only fetches a single commit from the selected branch in the repository containing the Job's source code. The image builder also performs "clone" and "checkout" in a single step. Downloading fewer data makes the build process faster. (#446)
  • .git folder is excluded from the job image when copying the source code.

Fixed

  • Fixed race condition bug in Async job calls. (#449)

[2.28.0] - 2024-04-16

Added

  • Status of an async job call can be checked at endpoint: /pub/async/task/{ID}/status. See Asynchronous calls to jobs guide for more details. (#434)
  • Maximum number of concurrent requests can be limited by max_concurrency field in a manifest: 
    jobtype_extra:
  max_concurrency: 1
    By default, concurrent requests are unlimited. Setting max_concurrency to 1 will make the job process requests one by one. Overdue requests will be queued and processed in order.

Having such concurrency limits may cause some requests to wait in a queue. If an average throughput is higher than the job can handle, the queue will grow indefinitely. To prevent that, you can also set jobtype_extra.max_concurrency_queue to limit the queue size. When the queue is full, the job will return 429 Too Many Requests status code.

See the Python job type reference - Docker command can be overwritten when running job locally. Check out racetrack run-local --help. (#431) - New flag in Racetrack CLI to avoid using cache when building the job's image: racetrack deploy --no-cache (#430) - Manifest input can be piped into the Racetrack CLI to deploy a job. (#240)

Changed

  • Asynchronous job calls are now resilient to restarts by automatically retrying the requests, The restart could be either due to an upgrade in Racetrack services or a job crash (like an Out of Memory kill). Regardless of the reason, the job call will keep retrying until it hits the maximum number of attempts. Async job calls are stored in a database for a short period. (#424)
  • If an error occurs while deploying a job to an infrastructure, the unsuccessful deployment is rolled back to clean up the mess. The timeout for intializing a job in the cluster (waiting until it's alive) has been increased to 15 minutes. (#436)
  • A new status STARTING has been introduced for a Job that is still initialising. This way, the ERROR state is reserved solely for the unexpected problems that arise after a job has been successfully initialised. It is possible to distinguish whether a job has ever run successfully. Thanks to this, version aliases, such as latest or wildcards, won't redirect to starting Jobs that are not yet ready. (#437)

Fixed

  • FastAPI dependency has been upgraded to solve memory leaks. (#442)
  • Fixed nullable extra_vars in racetrack CLI. (#440)

[2.27.0] - 2024-03-04

Added

  • Job type plugins can validate their part of the manifest. See validate_job_manifest hook in Developing plugins
  • Structured logging can be enabled by setting LOG_STRUCTURED environment variable to true. This will cause Racetrack services to produce logs in a JSON format. The default logging formatter can also be changed by jobs. For more details, refer to the documentation of the specifc job type plugin. Take a look at the python-logging-format sample to see how to configure your own logging formatter in jobs.
  • Introduced asynchronous calls to jobs. This feature lets you avoid timeouts on very long requests by switching to a two-step process: request a task and then check on the result. See Asynchronous calls to jobs guide for more details.

Changed

  • pydantic package has been upgraded to version 2 to prevent from conflicts with other up-to-date packages. pydantic is one of the dependencies of the racetrack-client package.

[2.26.0] - 2024-01-24

Changed

  • Job type plugins can now build job images from a single Dockerfile. Base Dockerfile has been merged with a job template. It gives more flexibility by allowing to parameterize all build commands of a base image with a user-defined configuration. See Developing plugins and Job type plugins documents. Old style is still supported to keep backwards compatibility. (#403)

[2.25.0] - 2024-01-16

Added

  • Administration tab in Dashboard shows number of jobs using the particular job type or infrastructure. This helps to indicate which job types are actually in use and which can be deleted. It also shows which job types and infrastructures are provided by which plugins in a tree view. (#392)

  • Activity tab on Dashboard tracks more kinds of events:

    • Plugin installed
    • Plugin uninstalled
    • Deployment attempt failed
    • Job moved (to other infrastructure)

    (#382)

Changed

  • Dashboard's "Audit Log" tab is now called "Activity".

Fixed

  • Dashboard's auto-update feature works with multiple Lifecycle replicas. Live updates can now be enabled on the jobs list. It is turned off by default. (#317)

[2.24.0] - 2023-12-14

Added

  • racetrack plugin bundle command has --out-filename parameter that allows you to overwrite the filename of the output ZIP file. (#391)

Changed

  • Infrastructure targets can now mount secret vars to a job regardless of environment variables. Secret variables are passed as a separate argument, they're no longer mixed with regular env vars. Infrastructure plugins should be updated accordingly due to interface change. Specifically, deploy_job function of JobDeployer class now has a new argument runtime_secret_vars: Dict[str, str] with secret env vars for a job. See Supported hooks. (#394)

[2.23.0] - 2023-12-04

Added

  • A command racetrack get remote -q (with flag -q or --quiet) prints only the current address of Lifecycle (without other logs), which makes it usable for scripts. Likewise, a command racetrack get pub -q (with flag -q or --quiet) prints the current address of Pub service. "Quiet" mode is automatically applied when not in a TTY.
  • Racetrack can be turned into a Maintenance mode. During that time, users are unable to make changes. Therefore, deploying, deleting and moving jobs is disabled. See how to enable maintenance mode (#370)
  • A job can be deployed through the Dashboard. To submit a job manifest, go to the "Deployments" tab and then "Deploy a new job" . Note that secret variables are not supported in this manner; instead, use the command line client. (#306)
  • A new tab called "Deployments" displays a list of recent deployment attempts. After selecting one, you may view deployment data such as status, failure cause, duration, and so on. In case of trouble with deployment, you can share this link with a support team.

Fixed

  • Fixed checking job condition in case of a missing Content-Type header of job's live endpoint. #376
  • Word wrapping applied to job logs in Dashboard. #372
  • SIGTERM signal triggers graceful shutdown. #379

[2.22.0] - 2023-11-07

Added

  • Manifest values can be overriden with key-value pairs coming from a command line. It doesn't modify actual file, but its one-time, in-memory version before submitting it. Racetrack client has --extra-vars KEY=VALUE parameter (or -e in short) that overwrites values found in YAML manifest.

    • KEY is the name of field and it can contain dots to refer to a nested field, for example git.branch=master.
    • VALUE can be any YAML or JSON object.

    Extra vars parameters can be used multiple times in one command. Example: racetrack deploy -e secret_runtime_env_file=.env.local -e git.branch=$(git rev-parse --abbrev-ref HEAD) It makes CLI commands more script-friendly, so you can overwrite manifest without tracking changes in job.yaml file. Tip: Use racetrack validate command beforehand to make sure your final manifest is what you expected. (#340)

  • You can install Racetrack to a standalone host (e.g. EC2 host or fresh VM instance) using the installer script that runs it on the Docker Engine infrastructure. (#308)

  • Command racetrack get auth-token prints out current auth token. It can be used in CLI scripts: curl -H "X-Racetrack-Auth: $(racetrack get auth-token)" (#357)

  • Command racetrack login --username <username> allows you to log in with your username and password (entered into the standard input) and saves the auth token without having to visit the Dashboard page.

Fixed

  • Manifest is validated after updating it on Dashboard. Changing primary keys (name or value) is forbidden. (#293)
  • Fixed reviving a dead job after deleting it. (#366)

[2.21.0] - 2023-10-16

Added

  • Pub can now be turned into a remote gateway mode. It allows to distribute services between clusters as 2 (or more) separate infrastructures:

    • main hub - hosting core Racetrack services: Lifecycle, image-builder, Dashboard, main Pub
    • remote jobs - infrastructure hosting jobs and remote Pub gateway that protects jobs from unauthorized access.

    It can cooperate with the following infrastructure plugins:

  • New administrative endpoint for cleaning up plugins mess. (#331)

Changed

  • Streaming live logs of a job can now work with multiple infrastructures. Logs streamer interface has been redesigned so infrastructure plugins has to be updated accordingly.
  • Prometheus, Grafana and PostgreSQL have been upgraded.

Fixed

  • Database connection is checked in background to avoid blocking metrics endpoint. It reduced number of open database connections. (#318)
  • Faulty plugins can be deleted.

[2.20.0] - 2023-09-27

Added

  • Lifecycle has more metrics and Grafana panels for monitoring: number of requests, websocket clients, server errors, open connections and active threads.
  • Improved error handling and logging in Lifecycle.

Fixed

  • Fixed metrics endpoint, which was making Lifecycle unresponsive to liveness probes in case of database connection errors. (#314)
  • Fixed TypeError in racetrack client on Python 3.8 (#320)

Removed

  • Auto-Update feature has been disabled temporarily due to generating too many requests on multiple replicas.

[2.19.0] - 2023-09-04

Added

  • Plugin's manifest can declare its category, which makes the label to be displayed on the plugins list. Category is a kind of the plugin and can be either "job-type", "infrastructure" or "core".
  • If you don't want the plugin to be loaded by Racetrack's Lifecycle or Image-builder (due to missing imports), you can enable it selectively for particular components. Declare components field in a plugin-manifest.yaml and add 'lifecycle' or 'image-builder' to the list. See Developing plugins. (#291)
  • Plugins can implement run_action method for calling a supplementary action with an endpoint. It can be used for debugging purposes or to extend plugin's functionality that is not covered by a plugin interface.

Changed

  • Installation to non-local cluster is done by utility script that automatically generates resources and unique passwords for your setup. (#298)

Fixed

  • Fixed racetrack run-local command. (#312)
  • Improved finalizing database connections. (#295)

[2.18.0] - 2023-08-04

Added

  • Job types can read the manifest file from the job's directory. It might be useful to configure the home page of a job (see example).
  • Racetrack-client has new flag --replace available when uploading a plugin: racetrack plugin install <file.zip> --replace. It will delete the existing versions (with the same name) on uploading a new one. Accordingly, there is a checkbox on the Dashboard's Administration tab for replacing a plugin on upload. By default, uploading a new plugin doesn't replace the older versions. (#270)

Changed

  • When a job is built from local files (with --build-context=local), files matching .gitignore are excluded. (#281)

[2.17.0] - 2023-07-17

Added

  • When a job is running a deprecated job type version (due to being no longer available), there is a notice displayed on a dashboard - "info" icon in a jobs tree and a "Notice" field after selecting a job. (#269)

Fixed

  • Missing schema.json is now distributed in PyPI packages. (#283)
  • Corrected aggregated metrics on Grafana Dashboards. (#273)
  • Fixed reloading local modules by plugins. (#275)

[2.16.1] - 2023-06-19

Changed

  • Dashboard displays original YAML for the job's manifest. (#262)
  • Dashboard now renders a table on the portfolio tab with QTable rather than TableFilter, which occasionally seemed to have rendering issues. Plus, it's more consistent with the current UI styling. (#258)

Fixed

  • Dashboard redirects to login page when session expires. (#266)

[2.16.0] - 2023-06-05

Added

  • Jobs tree on the Dashboard is refreshed in real time as soon as someone else's change is detected. You can turn it off on the "Jobs" page by clicking "3 dots menu", "Auto-update" toggle. (#239)

Changed

  • Chain calls to the jobs should be made by importing a dedicated function from a library provided by the job type plugin. This will keep the chain call function always up-to-date with the Racetrack version. See the example. (#20)

[2.15.0] - 2023-05-26

Changed

  • The name of the caller is recorded in the internal logs, giving the ability to track down who made the request based on its ID. Job types can also retrieve that information by extracting it from an HTTP header. See the python-job-type docs and a job for example. (#246)
  • racetrack-client shows more details in case of an HTTP error. (#245)
  • golang, python, and wrapper_properties fields are deprecated in Manifest schema, use jobtype_extra instead. This is backwards compatible. (#231)
  • Editing the manifest online triggers a redeployment of the job, keeping manifest up-to-date with running job. (#250)

[2.14.0] - 2023-05-18

Added

  • Dashboard links to job-related Grafana dashobards. issue #206
  • Manifest of a job can be edited online on the Dashboard after selecting a job. Keep in mind that your online changes can be overwritten by the next deployment, if you didn't reflect these temporary changes in git after all. issue #217

Changed

  • Minor UI improvements and styling for Dashboard, including:
  • Sorting jobs by status (starting from faulty ones)
  • Jobs ordering is persisted in local storage
  • Spinner indicators during loading the data
  • Status indicator (green/red/yellow) and number of jobs on a jobs tree
  • Panel takes up the whole space available - no white bars on the sides
  • Jobs tree has its own scrollbar
  • "10 seconds ago" labels are refreshed over time.

Fixed

  • Tree of jobs is refreshed after deleting a job. issue #211

[2.13.0] - 2023-05-10

Added

  • "Jobs" tab now shows the tree of jobs grouped by the family name. Jobs tree can be filtered by name, version, owner, job type version or infrastructure target. issue #212

Changed

  • Dashboard UI has been revamped and turned into Single Page Application using modern front-end frameworks, which made it more smooth and responsive. issue #212

[2.12.1] - 2023-04-21

Changed

  • Racetrack client returns non-zero exit code in case of error. issue #224
  • Dashboard is more responsive in case of a database malfunction (shows error rather than hanging indefinitely), since Postgres is no longer its hard dependency.

[2.12.0] - 2023-03-29

Added

  • New column in the list command of the Racetrack client displays a job type. Try it out by running racetrack list -c job_type. issue #207
  • New command racetrack call NAME ENDPOINT PAYLOAD [--version VERSION] [--remote REMOTE] [--curl] allows you to call an endpoint of a job. Provide the name of the job and the payload of the request in JSON or YAML format, for example racetrack call adder /api/v1/perform '{"numbers": [40,2]}'. Use --curl flag, if you want to generate a curl query instead of calling the job. Check out racetrack call --help for more details. issue #146
  • Name of the job can be autocompleted by hitting Tab while typing a CLI command. Remember to run racetrack --install-completion beforehand. Under the hood, it fetches the available jobs from the current remote.

Changed

  • The list command of the Racetrack client drops the fancy formatting and INFO/DEBUG logs when being piped into another command (ie. not connected to a terminal/tty device). Try racetrack list | cat. issue #207

Fixed

  • Performance of PUB component has been improved by reducing number of requests made to the Lifecycle and to the database. issue #155
  • Fixed saving plugin's config. issue #218
  • Racetrack's cookie can work even on non-HTTPS deployments. issue #225

[2.11.0] - 2023-03-17

Added

  • When specifying a job type version, wildcards can be used. For instance, mllab:1.3.*-3.9-bullseye to subscribe for the latest patches or mllab:1.*.*-3.9-bullseye if you feel more adventurous. The job could be upgraded without committing to manifest then. It's the extension of the existing latest tag, but it now supports multiple job type variants. Note: The release of a new job type version has no effect on existing jobs until they are redeployed. issue #183

Changed

  • Portfolio tab on Dashboard shows the exact version of a job type, even if its manifest specifies a wildcard version or "latest". issue #203

Fixed

  • Fixed opening a job after regenerating the token. issue #200
  • Fixed CSRF Trusted Origins protection. issue #197

[2.10.1] - 2023-03-10

Added

  • Grafana's "Lifecycle" dashboard now includes a new panel that tracks the number of jobs with the particular status.

Changed

  • "Orhpaned" jobs are no longer adopted by Racetrack. This was causing spurious "Lost" status after all. issue #163, issue #134
  • Third-party dependencies have been upgraded.

Fixed

  • Regenerating a user token via the button on your profile now no longer requires a refresh. issue #188

[2.10.0] - 2023-03-03

Added

  • Deletion pop-up contains the name of the job to confirm that you're about to delete the right one. issue #176
  • Button added to the profile page that lets you regenerate your authentication token. issue #139

Changed

  • Racetrack client keeps the local config file ~/.racetrack/config.yaml with mode 600 (not readable by others) and warns you if this file has insecure permissions since it may contain credentials. issue #179

[2.9.1] - 2023-02-20

Changed

  • Fatman has been renamed to Job in all database tables, columns and values. For instance, scope call_fatman is now call_job. issue #162

[2.9.0] - 2023-02-14

Added

  • Deployment phases, shown by a racetrack client, are more granular, including image building steps. issue #145
  • Racetrack comes with standalone Prometheus and Grafana. It contains dedicated dashboards for monitoring the jobs and internal services. issue #144
  • Database connection is being monitored closely by Lifecycle and displayed on a Grafana dashboard. issue #165

Changed

  • Fatman is renamed to Job. fatman.yaml is thus deprecated in favour of job.yaml file name. Although the endpoint for calling the Job has changed to /pub/job/NAME/VERSION/PATH, the old endpoint /pub/fatman/NAME/VERSION/PATH is still supported for backward compatibility. issue #136

[2.8.1] - 2023-01-27

Changed

  • Set User-agent header in all CLI requests to circumvent silly Cloudflare's protection.

[2.8.0] - 2023-01-26

Added

  • A job can be moved from one infrastructure target to another by using a new CLI command. Check out racetrack move --help.
  • A new command racetrack list allows to fetch the list of all deployed fatmen with the selected columns. See racetrack list --help for more details.

Changed

  • Syntax of Racetrack client has been rearranged. Check out racetrack --help for more details. Notable changes:

  • Racetrack URL is now called "remote" and it can be explicitly set with an optional parameter --remote alias in most of the commands. "Remote" can be a direct URL or an alias.

  • The current remote can be set once with racetrack set remote ALIAS_OR_URL and then you can omit --remote parameter in the next commands like racetrack list. Check your current remote with racetrack get remote.
  • Although racetrack deploy [WORKDIR] [REMOTE] syntax is still supported, it's deprecated now and racetrack deploy [WORKDIR] [--remote REMOTE] should be used instead.
  • Automatic completion can be activated by running racetrack --install-completion. Then, you'll see relevant prompts after hitting Tab.
  • To show the runtime logs, use racetrack logs NAME [--version VERSION] [--remote REMOTE]. No need to pass workdir with a manifest file any longer.
  • To show the build logs, use racetrack build-logs NAME [--version VERSION] [--remote REMOTE]. No need to pass workdir with a manifest file any longer.
  • To delete a job, use new syntax: racetrack delete NAME --version VERSION [--remote REMOTE] - job name and version is now required. No need to pass workdir with a manifest file any longer.
  • racetrack plugin install PLUGIN_URI [--remote REMOTE] - "remote" is now an optional parameter instead of a required argument.
  • To set up an alias for Racetrack's remote URL, use racetrack set alias ALIAS RACETRACK_URL (former racetrack config alias set).
  • To set read-access credentials for a git repository, use racetrack set credentials REPO_URL USERNAME TOKEN_PASSWORD (former racetrack config credentials set).
  • To save user's Racetrack Auth Token for Racetrack server, use racetrack login USER_TOKEN [--remote REMOTE] (former racetrack login RACETRACK_URL USER_TOKEN).

  • racetrack logout [--remote REMOTE] (former racetrack logout RACETRACK_URL).

  • In case of problems with reaching the Job, PUB shows the error page (JSON with the meaningful reason), instead of a white page of death.

  • The "Open" button gets deactivated on the Dashboard for jobs that are not Running.

[2.7.0] - 2023-01-05

Added

  • Every plugin can have its own configuration, stored in a YAML file. After uploading the plugin, the configuration can be edited on Dashboard's Administration tab, with a Edit Config button. Plugin can read the configuration from a file self.config_path: pathlib.Path.
  • Racetrack instance can operate with multiple infrastructure targets. By default, there is no infrastructure target. It has to be extended by installing a plugin. The infrastructure target is displayed next to the fatman name, on Dashboard's list. There is a new, optional field infrastructure_target in a Manifest. If given, it enforces using the specific target in case of many infrastructures are in use. Otherwise, the deafult infrastructure is applied.
  • Fatman can be composed of multiple Docker images provided by a plugin or user's manifest.

[2.6.0] - 2022-12-02

Added

  • "Portfolio" table in a Dashboard has a new column "Job type version".
  • List of available job types can be checked with: racetrack plugin list --job-types <racetrack_url>. It is also listed in the Administration tab on Dashboard.
  • Exact job type version can be checked at Fatman's /health endpoint.

Fixed

  • When uploading a faulty plugin, the errors are handled in a more reliable manner.
  • Plugin's requirements are installed with non-root user context.
  • Plugin directory is accessible when initializing the plugin in __init__ method.
  • Fixed conflict between newer protobuf version and opentelemetry package.

[2.5.1] - 2022-10-28

Added

  • Users can change their password on a Dashboard -> Profile tab -> Change Password
  • Fatman webviews can now serve ASGI applications (like FastAPI)

Fixed

  • Added missing __init__.py file in one of the racetrack-client modules.

[2.5.0] - 2022-10-18

Changed

  • One job type can be installed in multiple versions at the same time. Users have to pick one of these versions and specify it in the manifest of their fatman, eg. jobtype: python3:2.4.0. Version of a job type is now required in the jobtype field of Manifest in a manner name:version. latest version can be used (resolving to the highest semantic version), though it's discouraged.
  • Plugin bundler has been moved to racetrack-client. Now you can do racetrack plugin bundle instead of racetrack-plugin-bundler bundle.

[2.4.0] - 2022-10-11

Added

  • Plugins can be installed by means of racetrack-client command racetrack plugin install <plugin_uri> <racetrack_url>, where plugin_uri can be either:

    • local file path (eg. python3-job-type-2.4.0.zip),
    • URL to a remote HTTP file (eg. https://github.com/TheRacetrack/plugin/releases/download/2.4.0/python3-job-type-2.4.0.zip),
    • GitHub repository name (eg. github.com/TheRacetrack/plugin-python-job-type) - it takes the ZIP file from the latest release.
    • GitHub repository name with version (eg. github.com/TheRacetrack/plugin-python-job-type==2.4.0) - it takes the ZIP file from the specific release.

  • Plugins can be uninstalled with a new command: racetrack plugin uninstall <plugin_name> <plugin_version> <racetrack_url>.

  • List of currently installed plugins can be checked with: racetrack plugin list <racetrack_url>.

Changed

  • All job types are individual plugins now. Racetrack starts without any job type by default.
  • Base Fatman images are built inside Racetrack by image-builder so it's no longer needed to push images prior to the plugin release.

[2.3.0] - 2022-09-23

Added

  • Fatman access can be narrowed down to single endpoints. When adding Auth Resource Permission (in Admin panel), there is a new endpoint field, which narrows down the permission only to this particular Fatman's endpoint. If not set, it covers all endpoints (just as other resource filters do). For instance, ESC can have a permission with call_fatman scope to only call one endpoint.
  • OpenTelemetry can be turned on in Racetrack. Given the OTLP collector endpoint, the traces from lifecycle, pub and fatman will be sent there.

Changed

  • python3 job type is now based on 3.9-slim-bullseye image instead of 3.8-slim-buster.
  • Plugins are distributed as ZIP files. They can be installed and uninstalled in a Dashboard's Administration page. See using-plugins.md.

[2.2.1] - 2022-08-25

Changed

  • Maximum amount of memory for a fatman is reduced to 1GiB by default (if not set explicitly in a manifest). In general the maximum memory can't be more than 4 times higher than the minimum memory. Racetrack keeps an eye on this rule by automatically adjusting minimum memory amount, if needed. It is recommended to declare maximum memory amount explicitly in a manifest by defining resources.memory_max field.
  • Fatman versions in PUB's metrics are always resolved to the exact version instead of wildcards.

[2.2.0] - 2022-08-16

Added

  • Golang job types serve interactive Swagger UI with API documentation on the home page.

Changed

  • Function fatman_job_types of plugins changed its signature. Now the first parameter is a Docker Registry prefix to work well with different Docker registry namespaces. All job-type plugins has to be adjusted accordingly. See plugins-job-types.md for more details.
  • A namespace for docker images in a Docker Registry is configurable so that Racetrack can be released to any Docker Registry.
  • A namespace for fatman workloads running on Kubernetes is now configurable by means of FATMAN_K8S_NAMESPACE environment variable, by default it's set to racetrack.
  • Hostname subdomain of Racetrack services is now configurable by means of RACETRACK_SUBDOMAIN environment variable, by default it's set to racetrack.
  • Golang job type has been moved to Plugin - Go Job Type. Now it's not enabled by default in Racetrack.

[2.1.2] - 2022-07-29

Changed

  • Overall requests performance has been improved by switching from Flask & twisted (WSGI) to FastAPI & Uvicorn (ASGI) web server. Fatman server should now be less laggy under heavy load. Additionaly, FastAPI comes with better request validation and newer SwaggerUI page. Fatman redeployment is needed for the changes to take effect.

[2.1.1] - 2022-07-25

Added

Changed

  • Fatman's metrics now contain number of requests and duration for each endpoint (including auxiliary endpoints).

Fixed

  • Fixed redeploying fatmen from the dashboard
  • Making multiple attempts to deploy the same fatman at the same time is now disallowed.
  • "In progress" deployments are halted when Lifecycle restarts (due to upgrade or eviction)

[2.1.0] - 2022-07-08

Changed

  • Fatman secrets (git credentials and secret vars) are kept in Racetrack, so fatmen are fully reproducible again. In other words, it is technically possible to rebuild and reprovision fatman by anyone who is allowed to do so (anyone with permissions, not just the author) to apply latest Racetrack changes. You can click "Rebuild and provision" to do a complete redeployment of a fatman, or just "Reprovision" to apply changes without rebuilding the image (eg. useful when changing replicas count).

  • Racetrack client no longer depends on requests package in favour of using the built-in urllib module.

[2.0.0] - 2022-06-30

Added

  • Plugins can hook into the post_fatman_delete event to add their actions after deleting a fatman. See Teams Notifications Racetrack Plugin

  • Response errors are also displayed in the logs in addition to returning message in HTTP payload.

  • Racetrack client has new command racetrack run-local . <racetrack_url> allowing to run fatman locally based on docker image built by remote Racetrack. Docker has to be installed on your system.

Changed

  • Fatman permissions model is more restrictive. Now users, ESCs and fatman families can be granted fine-grained permissions. Users can have access to a subset of fatmen and they can see and manage only those that are allowed for them. Old permissions have been migrated (Fatman-to-Fatman and ESC-to-fatman access).

    By default, Users can now do the following:

    • read all fatman status (browse on a dashboard)
    • call endpoints of every fatman
    • deploy and redeploy fatmen
    • delete only fatmen that has been deployed by the user (starting from the newly deployed ones)

    Admin can grant or revoke permissions. Permisssion can cover either all fatmen, whole fatman family or a single fatman.

    Permission is related to one of the operation type (called "scope"):

    • read_fatman - list fatman, check fatman details
    • deploy_fatman - deploy fatman in a particular family, redeploy fatman
    • deploy_new_family - deploy new fatman family
    • delete_fatman - move to trash, dismantle from a cluster
    • call_fatman - call fatman endpoints
    • call_admin_api - not important for regular users
    • full_access - not important for regular users

  • Authentication tokens have been revamped and format changed. Users are required to use new tokens in order to use CLI racetrack client. Log in to Dashboard, copy the token and run racetrack login again. Old ESC tokens are still supported to keep backwards compatibility, but they should be regenerated if possible.

  • A method for calling fatman from inside of other fatman has been changed due to new auth model. See python-chain sample. All fatmen calling another fatman are required to be redeployed (to use new tokens).

  • Since there is one type of tokens, Auth header has been unified and renamed to X-Racetrack-Auth. It should be used no matter if it's ESC, User or a Fatman. Old headers are still supported but may be abandoned in the future: X-Racetrack-User-Auth, X-Racetrack-Esc-Auth, X-Racetrack-Fatman-Auth

  • Racetrack services has been adjusted to the new URL format. /ikp-rt prefix has been removed from all URLs. Instead, ikp-rt part may be included in the cluster hostname, making session cookies to work hassle-free and more secure. For instance, Racetrack Dashboard address is now: https://ikp-rt.<cluster.address>/dashboard. Lifecycle address is: https://ikp-rt.<cluster.address>/lifecycle. PUB runs on: https://ikp-rt.<cluster.address>/pub. Thus, Fatman instances can be accessed at: https://ikp-rt.<cluster.address>/pub/fatman/<name>/<version>/<path>.

[1.2.0] - 2022-05-31

Added

  • fatman.yaml manifest file supports overlays. Including extends field (with a path to base manifest) will make it to merge base and overlay layers. It may come in handy when working with multiple environments (dev/test/prod) referring to the same base manifest. See overlays sample.
  • Job types can be extended with plugins. Check out sample job-type plugin and developing-plugins.md to see a guide how to create and activate plugins.
  • Racetrack documentation is now served at "Docs" tab in Dashboard. Check out "Home" page to get started.
  • Plugins can extend Racetrack documentation with its own pages written in Markdown (it may be generated at runtime depending on plugin configuration).

Changed

  • racetrack validate PATH now evaluates the manifest outcome and prints it to verify the result of extending overlay manifest.
  • Critical actions like deleting fatman or redeploying it open a modal window with confirmation button to prevent accidental removal.
  • Racetrack environments are easier distinguishable between dev, test, preprod and prod due to different colors of the dashboard navbar and additional prefix [prod] in the title.

[1.1.0] - 2022-04-21

Added

  • New "Audit Log" tab in Dashboard shows activity events done by users, eg. "fatman F1 of user Alice has been deleted by Bob". It can be filtered by events related to a logged user, whole fatman family or a particular fatman.

Changed

  • Once a fatman name (and version) has been claimed, it can never be deployed again to prevent accidental/hostile reusing.
  • Login page has been revamped.

Fixed

  • Display error cause on dashboard in case of "Redeploy" failure.

[1.0.2] - 2022-04-06

Changed

  • Set default fatman maximum memory to 8GiB

[1.0.1] - 2022-04-06

Changed

  • Fatman dependency graph is interactive, nodes are movable. Clicking on a node filters out all its neighbours to see which fatmen can be accessed by particular ESC or which ESCs have access to a particular fatman.

Changed

  • Set default fatman minimum memory to 150MiB

[1.0.0] - 2022-03-29

Added

  • Fatman manifest can declare resources to be allocated to the Fatman with resources field, including minimum/maximum memory, minimum/maximum CPU cores. See Fatman Manifest File
  • Dashboard shows datetime when the last call was made to a fatman. Fatman redeployment is needed for the changes to take effect.
  • Dashboard has a new "Portfolio" tab for browsing fatmen freely with custom criteria and showing the candidates for removal.

[0.7.8] - 2022-03-18

[0.7.7] - 2022-03-18

Changed

  • Racetrack client validates given token when running racetrack login and complains immediately if it's wrong
  • Fatman returns 400 BAD REQUEST response in case of missing argument, unexpected argument or any other TypeError.

Fixed

  • Fixed Fatmen getting Lost. Fatmen shouldn't wrongly change its status to "Lost" any longer.

[0.7.6] - 2022-03-04

[0.7.5] - 2022-03-01

Added

  • Log messages contain Request Tracing ID, which allows to track down what was happening during the request processing by different services. PUB can read tracing ID from a specific HTTP request header. If not given, it will be generated. On IKP clusters this header is named traceparent (by default it's X-Request-Tracing-Id). ID is also returned in the response headers, so you can use it when reporting bugs. Fatman redeployment is needed for the changes to take effect.

[0.7.4] - 2022-03-01

Added

  • PUB collects histogram of fatman response times and metrics related to endless requests and response codes.

[0.7.3] - 2022-02-28

Changed

  • GET /metrics 200 requests are hidden from access logs.
  • Reading from stdin is prohibited at fatman initialization and it raises an error once detected, showing stack trace with the place where there was an attempt. Fatman doesn't go unresponsive any longer, awaiting input indefinitely.

Fixed

  • Fatman logs doesn't show misleading "Timing out client" errors anymore. It was related to HTTP sessions kept alive by browsers, not actual fatman request time-outs.

[0.7.2] - 2022-02-14

Changed

  • Access logs are not polluted with redundant /live, /ready & /health requests.

Fixed

  • Racetrack logs show non-ASCII characters

[0.7.1] - 2022-02-07

Added

  • Racetrack dashboard shows active plugins along with their versions.

Changed

  • Fatman manifest is checked against using HTTPS git remote only (SSH is not supported).
  • None values metrics are allowed. They're ignored instead of logging error as there is no null in Prometheus, only samples that don't exist.
  • Fatman access logs have shorter, concise format. Time, HTTP method, path and response code are the only ones displayed.

[0.7.0] - 2022-01-31

Added

  • When accessing a Fatman through a PUB, there can be used wildcard version format: 1.x or 1.2.x. This points to a highest stable version of the Fatman that matches the pattern. For instance, 1.x keeps track of all v1 updates, but skips next major versions 2.0.0 and above. Fatman redeployment is needed for the changes to take effect.

Changed

  • "latest" fatman version now points to the highest version disregarding versions with any label (eg. -dev). Thus deploying 2.0.0-alpha pre-release versions doesn't affect "latest" alias until 2.0.0 is deployed.

Fixed

  • Public endpoints are properly handled with version aliases (eg. "latest", "1.x")

[0.6.1] - 2022-01-24

Added

  • Phases of deployment are tracked by showing "building image", "creating cluster resources", "initializing Fatman entrypoint", etc. in RT client in place of bare "deployment in progress..." message.

Changed

  • Fatman versions are validated according to SemVer format. Version should match MAJOR.MINOR.PATCH[LABEL], eg.: 0.1.2 or 0.0.0-dev.

Fixed

  • Zero-value metrics are shown properly at /metrics endpoint of a Fatman.

[0.6.0] - 2022-01-17

Added

  • Racetrack functionality can be extended with plugins. Check out developing-plugins.md to see a guide how to create and activate plugins.

Changed

  • Whole directories are handled in static_endpoints to serve recursively all of its content.
  • Metrics from Postgres database and PUB are exported. It gives better monitoring of Fatman requests performance.
  • Registration form has been revamped to show errors in a more user-friendly way.
  • Pods resource requests have been adjusted to its actual needs. Cluster resource shortages shouldn't happen so prematurely anymore.

Fixed

  • Show real reason of internal exceptions, eg. Neo4j auth error

[0.5.0] - 2022-01-10

Changed

  • Overwriting fatmen is disabled. It's not allowed to deploy a fatman with the same name & version as an existing one, unless it's deployed with --force flag.
  • Duplicated env vars error is now more meaningful: "found illegal runtime env vars, which conflict with reserved names: RK_HOST"

[0.4.1] - 2022-01-05

Fixed

  • Fixed Manifests getting empty on Racetrack upgrade.

[0.4.0] - 2022-01-03

Added

  • Number of running instances of the Fatman can be configured with replicas field in a manifest.yaml. In case of multiple replicas, racetrack logs collects logs from all instances at once.
  • Deployed fatmen can be deleted through racetrack client using new command: racetrack delete <workdir> <racetrack_url> [--version X.Y.Z].

[0.3.5] - 2021-12-09

Added

  • Fatman endpoints that are supposed to be accessed publicly (without authentication) can be declared in public_endpoints field of manifest. This makes a request that needs to be approved by Racetrack admin. See python-ui-flask sample.

Fixed

  • Improved overall Fatman response time.

[0.3.4] - 2021-12-06

Changed

  • Fatman's /live, /ready and /health endpoints doesn't require authentication.

[0.3.3] - 2021-11-22

Changed

  • Fatman "dashboard" renamed to "webview". See Python Job Type docs: Custom Webview UI.
  • On local dev Racetrack there's no need to commit every job change prior to deployment (it gets the working directory snapshot).

Fixed

  • Default admin account is no longer recreated.

[0.3.1] - 2021-11-15

Added

  • New optional field in Manifest: labels - dictionary with metadata describing fatman for humans. Labels are shown on a dashboard page besides fatman name.
  • Example input data (shown on Swagger UI) can be defined for auxiliary endpoints by implementing docs_input_examples. See Python Job Type docs: Auxiliary endpoints section and python-auxiliary-endpoints sample.

Fixed

  • RT specific Dashboard cookie is removed upon logout
  • Error returned by Fatman contains more information (including Lifecycle response) and uses correct status code (instead of 500)
  • Fixed showing spurious success of Fatman redeployment (ensure new Fatman responds to health probes).

[0.3.0] - 2021-11-08

Added

  • racetrack logs command follows Fatman logs when called along with --follow flag.

Changed

  • Lifecycle API is fully authenticated, every request to Fatman needs to have either User, ESC or Fatman token.
  • X-Racetrack-Caller header has changed to X-Racetrack-Esc-Auth

[0.2.1] - 2021-10-25

Added

  • Multiple versions of the same fatman are allowed. Obsolete versions are not removed automatically. ESC permissions are assigned to whole fatmen family by its name.
  • Build logs are always captured and can be seen on Racetrack Dashboard by clicking "Logs" / "Build logs" or by issuing racetrack build-logs <workdir> <racetrack_url>.
  • Fatman logs are displayed on Dashboard under option "Logs" / "Runtime logs".

Changed

  • Fatman base URL is changed to /ikp-rt/pub/fatman/<fatman-name>/<version>/. Version can be e.g. 0.0.1 or latest.

[0.2.0] - 2021-10-11

Added

  • Environment vars can be configured in manifest (including build time env, runtime env and secrets). Therefore, pip dependencies can be installed from a private repository. See Python Job Type docs: Environment variables section.
  • Swagger UI allows to set X-Racetrack-Caller header for making authorized Fatman calls.
  • Dashboard profile shows a user token
  • Racetrack client deploy and logs commands enforces logging with token
  • racetrack login and logout commands
  • racetrack config show for viewing racetrack config

Changed

  • docs_input_example outcome is validated at deployment stage. It's checked against JSON-serializability and its size should not exceed 1 MB. This solves unresponsive Swagger UI pages, but it may enforce to use a brief example at the expense of having representative one.

[0.1.1] - 2021-10-04

Added

  • Customized Prometheus metrics can be exported by implementing metrics method.

Changed

  • Dashboard allows users to register, to be able to login and view list of Fatmen.

Fixed

  • Generating graph on RT Dashboard
  • Generating token for ESC in RT admin panel

[0.1.0] - 2021-09-27

Changed

  • Hide "Open" button for erroneous Fatman on Dashboard page. Instead, show a section containing error message and logs.
  • Racetrack client infers full lifecycle URL, namely https protocol and /ikp-rt/lifecycle path, if not given. For instance, ikp-dev-cluster.example.com is a valid URL to deploy fatmen.
  • Admin panel not available on Dashboard (moved to Lifecycle).

Fixed

  • Use DNS hostname of a cluster instead of raw IP address when redirecting to a Fatman URL (fixed in dashboard "Open" link as well as in racetrack deploy output).

[0.0.17] - 2021-09-20

Added

  • Fatman serves 3 version numbers at /health endpoint:
    • git_version - arising from the git history of a job source code,
    • fatman_version - taken from version field in manifest (this is also displayed at SwaggerUI page).
    • deployed_by_racetrack_version - version of the Racetrack the Fatman was deployed with.
  • Dashboard displays current Racetrack version in the footer.
  • Static endpoints for serving local files at particular path (eg: /xrai endpoint serving xrai.yaml)
  • Authorization to PUB (optional right now) for accessing Fatman /perform. Requires creating ESC and distributing its token to user.

Fixed

  • Show localized Dashboard dates on Firefox

[0.0.16] - 2021-09-13

Added

  • build-essential package is incorporated into python3 base image. gcc and g++ packages are there out-of-the-box, so there is no need to include them manually to your system_dependencies any longer.
  • Dashboard page shows both dates for each Fatman: "Created at" and "Last updated" (deployed at) along with its age (e.g. "5 minutes ago", "5 days ago").

Changed

  • In case of python3 initialization error, full traceback coming from fatman is displayed to a user (eg: it's easier to find out where is a bad import).
  • Deployment errors displayed by CLI client are shortened and yet more meaningful. Client's traceback is not displayed if not needed.
  • Datetimes shown on Dashboard get a localized timezone from a browser, so it's no longer UTC time (unless you work from UK).
  • Fatmen list on Dashboard is ordered by last deployment date.

Fixed

  • Fix retrieving logs right after deploying a Fatman.
  • Increased timeout for creating a pod. It takes into account longer image pulling.

[0.0.15] - 2021-09-07

Added

  • New required owner_email field (email address of the Fatman's owner to reach out) in manifest
  • List of Fatmen (in Dashboard) shows who deployed the fatman recently (username is taken from git credentials)
  • Auxiliary endpoints - custom endpoint paths handled by entrypoint methods (eg. /explain endpoint)

Changed

  • Racetrack client shows full logs from building the image in case of failure.

[0.0.14] - 2021-09-06

Changed

  • Fatman has read-write access to local working directory.
  • Deployment errors are more meaningful and less misleading due to revised health checks (ie. cluster errors are distinguished from job syntax errors or initialization errors).
  • One common versioning for all Racetrack components (backend and CLI client)

Fixed

  • Fix "Critical worker timeout" WSGI error
  • Increase memory limits for building images (up to 8Gi)
  • Fix Lifecycle resurrecting dead jobs

[0.0.5] - 2021-08-26

Added

  • Setting aliases for Racetrack servers

Changed

  • Syntax for configuring private registries credentials
  • Input payload is flattened (without "args" & "kwargs")

[0.0.4] - 2021-08-20

Added

  • Showing recent logs from Fatman

[0.0.2] - 2021-08-10

Added

  • Deploying Fatman