Deis Workflow 2.2 Release

25 Jul 2016

It may be the hottest July on record but we've got some new cool stuff in the latest release of Deis Workflow. Sit back, grab a cold drink, and check out the latest Workflow release.

Now Powered by Deployments

As part of Deis Workflow 2.2, developers can enable the use of Kubernetes Deployments! Running deis config:set DEIS_KUBERNETES_DEPLOYMENTS=1 will use start using Deployments under the hood. No need to re-create your application.

Previous to this release, Workflow would orchestrate rolling updates just like kubectl used to. Workflow's controller would increment and decrement replica counts over both old and new Replication Controllers. By using Deployments, all of this orchestration happens within Kubernetes control loops. This has some pretty sweet advantages:

  • deis/controller code is simplified
  • server-side execution means deployments will complete even when client connections are broken (internet drops, ctrl + c, etc)
  • release and rollbacks are also tracked in Kubernetes, giving operators total visibility
  • this opens the door for HorizontalPodAutoscaling in a future release

Cluster operators that would like Deployments as their default methodology today can set DEIS_KUBERNETES_DEPLOYMENTS=1 environment variable in the deis-controller component.

Changing deployment mechanisms can always reveal subtle timing behaviors that application developers may have inadvertently come to rely on. So we are rolling this out on an opt-in, per-application basis.

We hope to make this the default behavior in a few releases, so kick the tires and let us know if anything behaves out of the ordinary.

Log Shipping

We've continued to refine the log shipping system and now include an instance of Redis. This allows us to horizontally scale the number of logger processors while maintaining a consistent view via deis logs.

We've also added a new SumoLogic plugin to our fluentd setup. For more information on how to configure fluentd to ship to more backends see the project readme.

New Metric Dashboards

You may not know it, but Workflow ships a full metrics collection stack designed to keep an eye on all the Workflow components. The stack is backed by Fluentd, Telegraf, InfluxDB, and Grafana and monitors platform and component health. We will continue to add detailed component metrics over time. For information on the architecture and details on accessing the dashboard see the platform monitoring documentation:

Workflow Component Health

This dashboard aggregates memory and CPU usage for all Deis Workflow components.

Deis Workflow Health

Router Health

Router metrics gives you a high-level view of the routing subsystem including aggregate status codes, per-application request rate, per-application and aggregate response times.

Router Health

InfluxDB Health

The InfluxDB dashboard lets you keep an eye on the rate at which series are flowing into the database.

InfluxDB Health

Kubernetes APIServer Health

The Kubernetes dashboard brings together API server, kubelet, and Docker engine response times. Quickly see if any nodes are starting to act up.

Kubernetes APIServer Health

NSQ Health

NSQ dashboard shows queue depths for both logs and metrics. As well as a real-time count of metric and log consumers.

NSQ Health

Redis Health

Redis now holds a rolling set of logs per-application. This dashboard gives you overall redis health including CPU, memory and total operations per second.

Redis Health

Upgrading to Workflow 2.2

Upgrading is very easy if you've configured off-cluster object storage. Follow our upgrade instructions and enjoy Workflow 2.2.

Highlighted Changes in Workflow 2.2

Features

  • b59bbbc (fluentd) - fluentd: Adding sumologic plugin support
  • 424523c (logger) - storage: Add redis storage adapter
  • 2da72a5 (logger) - redis: Optimize with more aggresive pipelining
  • 0c82466 (logger) - storage: Make redis the default storage adapter
  • 2f92eca (monitor) - telegraf, grafana: Start collecting redis metrics
  • c9718e4 (charts) - logger: Add redis instance for use by logger
  • 7d40069 (charts) - swift: add support for swift storage
  • d6992e1 (charts) - telegraf: Configure telegraf to start fetching redis metrics
  • 0ae9d90 (workflow-cli) - deis: add deis shortcuts command
  • 2862f05 (workflow-cli) - colors: reserve magenta for controller log messages (#132)
  • 8a61e63 (controller) - scheduler: add support for set based requirement filtering via the kubernetes API
  • 47b5b08 (controller) - IDN: add support for international domains
  • 32be50a (controller) - scheduler: sort env vars and secrets by keys for easier hashing
  • c3c2494 (controller) - add Deployments support behind a feature flag
  • edb0383 (controller) - scheduler: feat(scheduler) prepend [namespace] to Scheduler log message for better traceability
  • 84b8080 (controller) - app: make deploy timeout configurable globally/per-app via DEISDEPLOYTIMEOUT, default is 2 minutes

Fixes

  • 65f8714 (dockerbuilder) - objectstore: set properly builder bucket file environment variable
  • 9179923 (dockerbuilder) - deploy.py: handle chunked output errors
  • 02112ed (dockerbuilder) - deploy.py: delay so stderr is logged before pod exits
  • 93b9c5c (logger) - redis: Pass style checks
  • 7144c4e (logger) - rc: Specify use of redis storage adapter
  • 9dae9cc (monitor) - telegraf: Create nsq topic at startup
  • a9b2275 (monitor) - telegraf: Update image default image tag in manifests
  • 8fd5ada (monitor) - grafana: Fix blank dashboards from appearing in dropdown
  • 7de06cc (monitor) - grafana: Redis cpu graph was not selecting right data
  • 0492bf8 (charts) - database: Fix logic for selecting off-cluster db
  • 6fd8407 (charts) - logger: Do not helm keep rc deis-logger-redis
  • c425baf (workflow-cli) - ps: restarting a single pod created by Deployments was not working
  • bcb8b01 (workflow-cli) - ps: give ps:restart a better understanding between RC and Deployment pods
  • 4b2c6d4 (workflow-cli) - scale: dont call controller if there is no valid scale pattern
  • 32177aa (workflow-cli) - settings: don't panic on empty settings file (#134)
  • 6214d96 (controller) - scheduler: if one RC fails to scale then ensure all other RCs are at the right level
  • 0bcea13 (controller) - scale: return error message in proper format
  • 7d24923 (controller) - restart: wait for the pods to be scheduled
  • 7647569 (controller) - scheduler: cast port to an int from environment (#857)
  • b27c816 (controller) - boot: change group ownership of docker socket to deis (#804)
  • d4415c9 (controller) - api: fail when rolling back to v1 (#762)
  • 1ac6b54 (controller) - release: return port from get_port for non-routable process types
  • a0567ea (controller) - api: remove command escaping from v1 (#822)
  • f403efb (controller) - tests: sort domains in tests to get past occasional ordering problems which cause test failures

Documentation

  • dd2b505 (workflow) - logging,monitoring: Update platform-logging and monitoring docs with recent changes
  • dfb6c8e (workflow) - install-workflow: note that k8s 1.2.x is required
  • aa40a49 (workflow) - swift: Add swift as an object storage
  • 3e7e865 (workflow) - deployments: add Deployments documentation
  • ee03d28 (workflow) - apps: document DEIS_DEPLOY_TIMEOUT and the nuances around that

Maintenance

  • 5c289c4 (slugbuilder) - buildpacks: update heroku-buildpack-go to v42
  • d5cdd0b (slugbuilder) - buildpacks: update heroku-buildpack-php to v108
  • 85c8292 (slugbuilder) - buildpacks: update heroku-buildpack-nodejs to v91
  • 165105f (slugbuilder) - buildpacks: update heroku-buildpack-python to v81
  • 9bbed87 (slugbuilder) - buildpacks: update heroku-buildpack-scala to v71

Posted in Workflow, Release, Announcement

triangle square circle

Did you enjoy this post?