DC/OS 1.9 includes many new capabilities for Operators, and expands the collection of Data and Developer Services with a focus on:
- Tools for Production Operations - Monitoring and troubleshooting for distributed apps.
- Broader Workload Support - From traditional apps to machine learning.
- New data and developer services.
Contents
What's New
Apache Mesos 1.2 and Marathon 1.4 integrated
Container Orchestration
- Pods - Multiple co-located containers per instance, scheduled on the same host. For more information, see the documentation.
- GPU - Leverage GPUs to run novel algorithms. For more information, see the documentation.
- Significant scalability improvements.
DC/OS Monitoring and Operations
Remote Process Injection for Debugging
The new dcos task exec
command allows you to remotely execute a process inside the container of a deployed Mesos task, providing the following features.
- An optional
--interactive
flag for interactive sessions.
- Attach to a remote pseudoterminal (aka PTY) inside a container via the optional
--tty
flag.
- Combine the
--interactive
and --tty
flags to launch an interactive bash session or to run top
and see the resource usage of your container in real time.
For more information, see the documentation for the debugging documentation.
Logging
Stream task and system logs to journald by setting the mesos_container_log_sink
install-time parameter to journald
or journald+logrotate
. This allows you to:
- Include task metadata like container ID in your queries to more easily locate the logs that you want.
- Use new DC/OS CLI commands
dcos node log
and dcos task log
DC/OS CLI commands to query the logs. You can also make requests directly against the new Logging API.
- Set up log aggregation solutions such as Logstash to get logs into their aggregated storage solutions.
For more information, see the documentation.
Metrics
- Node-level HTTP API that returns metrics from frameworks, cgroup allocations per container, and host level metrics such as load and memory allocation.
- StatsD endpoint in every container for forwarding metrics to the DC/OS metrics service. This service is what exposes the HTTP API.
- Any metric sent to STATSD_UDP_HOST/PORT is available in the HTTP API
/container/<container_id>/app
endpoint.
For more information, see the documentation.
Improved GUI
Networking Services
- CNI support for 3rd party CNI plugins.
- Performance improvements across all networking features.
Other Improvements
DC/OS Internals
- Update DC/OS internal JDK to 8u112 for security fixes.
- Update DC/OS internal Python from 3.4 to 3.5.
- The
dcos_generate_config.sh --aws-cloudformation
command will now determine the region of the s3 bucket automatically, preventing region mistakes.
- Added the
dcos_add_user.py
script, which you can use to add or invites users to a DC/OS cluster from the command line. For more information, see the documentation.
- Added
dcos-shell
which activates the DC/OS environment for running other DC/OS command line tools.
Expanded OS Support
Expanded Docker Engine Support
- Docker 1.12 and 1.13 are now supported. Docker 1.13 is the default version.
Upgrades
Improved upgrade tooling and experience for on-premise installations. Upgrades now use internal DC/OS APIs to ensure nodes can be upgraded with minimal disruption to running DC/OS services on a node. The upgrade procedure has also been simplified to improve user experience.
For more information, see the documentation.
Known Issues and Limitations
- Marathon-LB does not support pods.
- This release candidate has task logging to journald disabled by default, so task logs will continue to be written to their sandboxes, and logrotated out. The
dcos task log
command will work as it did before.
- 4137 - Volumes do not persist.
Issues Fixed Since 1.9.0-rc1
- DCOS-9738 - CLI-only packages show up as installable in the UI.
- DCOS-14045 - Exhibitor requests hang/deadlock on a five master cluster.
- DCOS-14201 - Minuteman module minuteman_lb_mgr crashes.
- DCOS-14053 - Can't switch from Docker containerizer to Mesos containerizer.