Mesosphere DCOS, my production feedback.

Mesosphere DCOS release became one of the most noticeable industry events this year! 

As a result, I suspect large organizations will speed up containerisation adoption and DevOps practices in general.

Possibility of commercial support for DCOS was one of the turnkey factors to choose this platform for me also.

In this article, I would like to briefly highlight my experience with DCOS, which could be helpful for future migrations.

•      Excellent out of the box configuration for rollout of non-production environments. Production requires lots of extra tuning.
•      Great idea to use CoreOs as a host platform, however, I would appreciate this distribution to add all kernel modules supported by Redhat and Ubuntu.
•       Great DCOS dashboard, it’s easy to show business all the benefits.
•       Handy command line interface.
•       Platform upgrade requires creation of new AWS stack and migrating apps from one cluster to another. I would like to see rolling update of the nodes built in!
•       No multi region support out the box, I had to refactor whole AWS CloudFormation teamplate to have multi AZ setup. I see no options to configure multi region cluster with this release, since mesos masters are limited with one scalability group.
•       No AAA/SSL built in for admin endpoints, it should be developed by user and this is lots of time, however Mesosphere is going to fix this future releases.
•     Admin services (zookeeper, mesos, marathon etc. ) are not isolated from user workloads on network layer. Can be done by customizing slave nodes provisioning with iptables rules.
•      No support for multiple instance-store drives, which is important for “high-end” AWS instances, not difficult to fix manually.
•       No stateful containers support at the moment, but this is due to version of docker used. Support of storage drivers and logging has been recently added to Docker engine after the initial release of DCOS.
•       Mesosphere decided to use cloud specific rollout tools and provisioning, which will add complexity to the maintenance of multi cloud setups. I guess, this is because new tools like Terraform and Packer are not supporting all features currently.

In spite of highlighted “gaps”,  overall impression is VERY positive; 

I’m impressed with amount of work done by Mesosphere team, which allowed me to implement all I needed on current project much faster. 

I highly recommend it if you need highly available and scalable platform in minutes!