Cloud Foundry Advisory Board Meeting, December 2015

Chip Childers from Cloud Foundry Foundation moderated the call. Dr. Max from IBM had managed the agenda, but was in China so wasn’t able to be on the call to lead it as he normally does. There were 31 people on the call.

Here is a recap, in chronological order:

Foundation Updates (Chip Childers, Stormy Peters, Cloud Foundry Foundation)

• Stormy discussed the Cloud Foundry Foundation having a focus on technical evangelism, to reach operators within the Cloud Foundry community and ecosystem. She noted that meetups serve as the primary vehicle, with about 170 scheduled worldwide annually. There is a mailing list for meetup organizers to view and a monthly report.
• She encourages meetup organizers to get on the list and mentioned she has travel funds to help speakers attend meetups. She also has boxes of t-shirts with the new logo!
• She encouraged people to contact her if they want to add more technical content on the CFF blog. Let her know and she can set people up with accounts.
• She discussed the idea of Cloud Foundry expert certification and a vendor-neutral trial of Cloud Foundry. There’s currently a group that meets every other Thursday to discuss certification. The idea is to create certs for dev and for ops, so people can say they are “Cloud Foundry-certified,” with advanced certifications to come later.
• The trial idea is to create a neutral place where potential customers can explore an instance of Cloud Foundry that is branded by the Foundation, just for people to try out. It would have limited resources and capability. If interested, the trial customers could be pointed to the available distros.

Chip discussed the Cloud Foundry Summit in Shanghai, which was held December 2-3. It was roughly the size of the Berlin summit in November (400 to 500 attendees) and featured some great sessions and training. Chip said there “was an amazing volume of activity related to Cloud Foundry in China” as well in some other locations in Asia. As the market continues to develop, he said the CFF will try to find ways to “fill the gaps in the geographic islands.”

 

MEGA (Amit Gupta, Pivotal)

• Amit discussed the rewritten consul release.
• Have unit-tested startup scripts, and is now time to add new features.
• Setting up consul as a recursor involves a number of known issues.
• Adding gte using get cpr.
• Upgrading 9.4.5 is a fair amount of work, given the upgrade path from previous verions of PostgreSQL.
• There’s a ton of work on cats around supporting Diego. For example, what back end to run to run the cats against? Working with Diego to import fitness tests into cats.
• Cut several releases of cf-release. Work on manifest generation tooling stalled out a bit due to some hairy pipeline issues. Will pick up work on this in the next week or so, depending on holiday schedules.

raised the issue of migrating to Spruce? He’s not a fan of Spiff, particularly relating to complex logins for manifest generation and static IP selection. Amit said many issues are being addressed with BOSH links and global networking. Dr. Nic said he simply wants to get rid of Spiff and migrate to Spruce, “because it’s so hard to use outside of Pivotal (CF).” Amit said he hasn’t received many complaints about Spiff and doesn’t see the same need articulated by Dr. Nic. The two agreed discuss this further offline.
 
Lattice ( , Pivotal)

• Pivotal will be withdrawing from Lattice. David said they don’t have the adoption participation to continue to move it forward.
• Perhaps everyone on the call has heard the news, but this should be news to a lot of people, Chip noted.
• Particulars of this announcement:
  • The next and last release has Windows support and some new CLI commands.
  • Windows support will land in v0.7.0 (this week)
  • vagrant up / terraform apply brings up Linux cluster from Windows
  • ltc works from Windows
  • ltc version and ltc sync—new CLI commands to check the cluster version and update the CLI
  • The upcoming release v0.7.0 will be the final release
  • v0.7.0 will not restore functionality to DigitalOcean, Google Compute or Openstack
  • Pivotal is taking the lessons from Lattice to create a local full CF instance

GitHub repo
 
Loggregator (Jim Campbell, Pivotal)

• Team back up to 2 pairs
• About to publish roadmap to cf-dev
• Focus on Metron->Doppler TCP, epic almost finished
• Parallel epic is Loggregator as a separate release, consumable by non-CF services. Almost finished
• Next epic is a performance characterization of the entire Loggregator chain

GitHub repo
 
CLI (Dies Koper, Fujitsu)

• Dies was not on the call, but submitted the following notes:
• Only a few updates due to PM and team members’ absence (CF Summit, vacations). Team should be at full strength again soon after New Year’s.
• Released v6.14.0, which allows Org and Space Mgrs to manage roles of their users (used to be admin-only)
• Making plans around having a CLI developer visit Sydney (Australia) in January to ramp up a new Sydney based CLI developer from Fujitsu
• Working on a number of chores to clean up the CLI code base in preparation for onboarding new team members
• Upcoming work is around exposing new Routing features through the CLI.
• Also about to start on extensions to hit the new v3 cc api endpoints. These will be hidden from users until feature set complete, which is at least several releases down the road.

GitHub repo
 
Routing (Shannon Coen, Pivotal)

• Continue to make progress for support of routing services.
• CLI work is all in queue.
• Work on support for TCP routing, and adding support for multiple app ports for apps running in Diego.
• You can run an app that listens for web traffic and debugs etc. It works for TCP and HTTP.
• Immediate question is with regard to integration with routing tier. Strongly considering changing the headers from a custom one to a Via header. He urges people to comment on this issue.

GitHub repo
 
Services Core (Marco Nicosia, Pivotal)

• cf-mysql-release v25 will include additional bugfixes and possibly a newer version of MariaDB, after superficial evaluation.
• Includes MariaDB 10.0.22
• Includes a manifest option to stop auto-publishing all service plans.
• Includes a new errand to auto-bootstrap a cluster. He noted “there are a million little edge cases and corner cases, so this was complex.”
• Useful in the event that all VMs are powered-off / terminated (say, to save money in a Lab env).
• Next up in cf-mysql: Reducing number of cluster VMs from three to two to save resources
• Leverages Galera arbitrator daemon to participate in leader elections, but not replicate data. Will maintain multi-AZ deploy.

 
BOSH ( , Pivotal)

• There’s been lots of testing. Lots of cases trying to break it. Looking for unexpected behaviors.
• Producing a few stem cells looks like there’s a lot of different security issues.
• Continue working on back up stuff with help from IBM.
• Finishing up BOSH store command, which is useful with the backup command. Very useful in recovery.
• Improving BOSH for not asking also sorts of credentials, especially for BOSH Lite.
• There a few stories left in epic about testing.
• Have also focused on improving candeling, and looking into edge cases.
• You can take a look at the CPI/API v 2 proposal in the repo.
• Team in New York is exploring BOSH Windows support.

GitHub repo
 
CAPI (Dieu Cao, Pivotal)

• Continuing work on v3 cc api and application process types support
• Elastic Clusters proposal—gathering feedback. Hope to incept on the CAPI part of the work in early January.
• Working on a proposal for support of Tasks. Plan to share on cf-dev soon when it’s a bit more baked.
• Finishing up last story on Private Brokers track (single space only)
• Removing support for v1 service brokers in the next cf-release

GitHub repo
 
Diego (Eric Malm, Pivotal)

• Working on bind-mounting downloads to improve buildpack staging times
• Mostly done with upgrade test suite, optimizing time to run, reliability on BOSH-Lite in CI
• Using BBS benchmark tests to determine scale limits of etcd‘s ability to store Diego data
• Simplified garden-container cache in executor to improve cell state response time
• Executor continually checks garden’s ability to create/run/destroy containers, rep opts of auction if it fails
• Route-emitter avoids destructive action on its in-memory DesiredLRP data if the DesiredLRP dataset not known to be up to date
• Fixed a bug that prevented staging implicit ‘library’ images from Docker Hub
• Reproducing observed slow evacuation at higher container densities to validate proposed introduction of native port-check action, changes to auction’s distribution of new containers

GitHub repo

from Huawei asked Eric if he would be able to share the “etcd” benchmarking result swith the community. “Capturing data from each run and figure out how to publish it what scale were able to hit with focused benchmarks,” he noted. He said this is a big problem and there’s the same issue with Kubernetes. He thought “the whole industry can leverage” the work Eric is doing. It didn’t seem to be a problem for Eric to respond positively to this request.

 
Garden-Linux (Dr. Julz, IBM)

• Reached feature parity with the last version to use btrfs (in other words, the switch to aufs is now truly complete). v0.329.0 is the version with all features on aufs.
• Improvements to the way host side ports are allocated for NetIn rules to reduce the chance of stale routes sending traffic to the wrong app. Now considering whether we need to improve this even further to handle stale routes which survive as long as VM recreation takes.
• Lots of work on performance monitoring and CI environments, our CI now deploys both isolated and full Diego environments and runs performance tests

GitHub repo
 
Guardian (Dr. Julz, IBM)

• Back up and running after a hiatus—full steam ahead!
• Currently aiming to ship v0.1.0 “MMVP” in Jan, which we hope will be deployable with Diego so that we can start integration testing
• No Docker image support yet
• No disk, network or cpu limits

Pivotal tracker – GitHub repo

from Stark & Wayne wanted to know how many cells have been deployed. Dr. Julz said there have been 100-cell deployments with 10,000 instances. The next magnitude will be 1,000 to 1,200 cells nodes with 200-250K instances. He noted that the bigger issue is instances, rather than cells.

 
Greenhouse/.NET (Mark Kropf, Steven Benario, Pivotal)

• Greenhouse team is actively doing .NET workshops with Pivotal customers and are encountering lots of unique windows configurations that are leading to improvements in the Windows 2012 R2 container support.
• Greenhouse team members met with our friends from Microsoft in Shanghai and received feedback from attending .NET devs.



• After shipping rfga to Pivotal customers got feedback and made dev team visits to learn how customers configure windows. More visits scheduled.

 

GitHub repo

 

 

Buildpacks / Stacks (Mike Dalessio, Pivotal)

• Made a bunch of releases last week and this week intro node 5 support.
• Statically linking open ssl 102, and finding it’s an example of why that’s not optimal. Now have a node binary with a vulnerability in it.
• nodejs-buildpack v1.5.4 addressed Node issues and OpenSSL CVEs, and adds Node 5 support
• java-buildpack v3.4 focused on improving developer diagnostic tools
• cflinuxfs2 1.19, 1.20, 1.21 addressing numerous CVEs
• Planned releases for later this week: ruby-buildpack v1.6.10, php-buildpack v4.3.1, go-buildpack v1.7.1, python-buildpack v1.5.2
• Danny Rosen, Pivotal PM, has begun a CF PM Dojo, shadowing Mike Dalessio, with the goal of being nominated as the PM for Buildpacks in early 2016.
• Investigating support for PHP 7

 
UAA (Sree Tummidi, Pivotal)

• Referred people to runtime pmc notes on github.

 

 

 

GitHub repo

The next regularly scheduled meeting will be at on Wednesday, January 13, 8am Pacific.