Date 18 Nov 2019
Minutes VM (Vicky) Brasseur (she/her) & Will Stevens Syed Ahmed Lisa Caywood Casey Cain Ian Rae Darien Hirotsu We're opening tickets as soon as they come up! See our Jira. Today's focus will be a discussion on contributing to TF.Docs needs a LOT of help. Let's focus on what we can do to make it suck less Leading question: How many of you have installed Tungsten Fabric?A number of "yes's" Tools used:tf-devstack Contrail Ansible & fab scripts CloudOps Automation Tooling Next leading question: Were you able to get it to work twice in a row? What should the process be?Submit a Blueprint to Jira Defend the Blueprint at the Technical Worksteam Meeting Once the Blueprint is accepted, create a Spec and land it in Gerrit The PTL for the project in question reviews the Spec and approves it, so it can be implementedNote that there are challenges with this since well... we don't have PTLs (also we need to determine what the projects are) Implement the Spec once the Technical WorksteamIf you are a first time contributor, submit an ICLA or CCLA via Gerrit (assumes that all repositories have been moved) Once the initial implementation is complete, test the implementation locallyThis is a challenge given the state of tf-devstack, but let's start here (see Submit a Gerrit patch once the implementation is complete and tested locallyNote that there are challenges with docs given the state of Zuul Once you have +2, the code is merged Note here, there is no developer guide to navigate this Many of us have tried to install tf-devstack to push a patchWS: I've tried to VERY hard to get tf-devstack in a place where I can test my patches to feel confident that my code worksLots of contorting and pain result from this exercise The best place we landed was building the artifacts and pointing Contrail Ansible Deployer to a ~private instance of Docker Hub with those artifacts WS: Also, there are two architectures that result from the deployers.Docker compose (Contrail Ansible Deployer)HA is an issue with this deployment No other CNI does it this way Our guess is that Kubernetes was not in the picture when the deployer was created Kubernetes manifest (One click guide)These manifests haven't been updated in a year They don't work any longer for Kubernetes 1.16 (Stateful and Daemon sets changed) This is the way all other CNIs are built (and how TF should do it) WS: Another challenge is building the manifest (see BluePrint) WS: If we can systematically build a manifest based on variables, then it's much easier to create a local dev environmentAs an artifact, it also helps us integrate with deployers that DevOps engineers use like RKE It was not clear that the BluePrint created needed to be discussed in the Technical Workstream Meeting (oops)Also, the TSC and the TWS members need better accountability to ensure that BluePrints don't die on the vine Blueprint process issuesNeed clarification on the mapping of Blueprints to specs and releases Specs are the expected implementation of a BlueprintPTLs will then have input on the spec PTLs have the responsibility of ensuring technical issues like backwards compatibility CLA Action items Casey Cain work with the TSC and TWS to drive the process with Prabhjot Singh Sethi to ensure BluePrints move through the process correctly (example issue: Will Stevens was not aware that he had to defend his blueprint with the TWS)