forked from kubernetes/community
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
MultiCluster 2019 unconference notes
- Loading branch information
Showing
1 changed file
with
84 additions
and
0 deletions.
There are no files selected for viewing
84 changes: 84 additions & 0 deletions
84
...nference-notes/Deploying_and_Scaling_Across_Multiple_Clusters-Vallery_Lancey.md
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,84 @@ | ||
| #### Presenter: | ||
| Vallery Lancey | ||
|
|
||
| #### Topic: | ||
| Unconference: Deploying and Scaling across Multiple Clsuters | ||
|
|
||
| #### Date & time: | ||
| 11/18/2019 9:30am | ||
|
|
||
| #### Notes | ||
|
|
||
| Getting started: what is multicluster? | ||
| * Federation / copy across otherwise independent clusters | ||
| * Deployment: rollout changes | ||
| - Hard w/o big CI/CD changes | ||
| - Granularity of rollouts is hard, if you have two clusters how do you launch less | ||
| than 50% at a time | ||
| - Non-native k8s apps often don’t deal with with partial rollout. | ||
| * Big variety of use cases & solutions, custom implementations everywhere | ||
|
|
||
| What is definition of multicluster? (vs just having a bunch of clusters) | ||
| * Deploy same app in multiple clusters | ||
|
|
||
| What are reasons to run? | ||
| * Big range, very customer-dependent | ||
| * Range from east-coast/west coast clusters to a cluster in every cell tower | ||
| * Different fault domains | ||
| * Avoid canary deployments | ||
| * Backup environment for rollbacks | ||
|
|
||
| Need tools above deployment layer, service meshes require committing to fearsome | ||
| stack (eg Istio) | ||
| * Example: how to generate Flagger to multicluster (Weaveworks incremental rollout | ||
| tool based on linkerd) | ||
| * Is there load balancing across clusters? | ||
| * Is the system aware of multiple clusters--basic things like how many clusters | ||
| to adjust rollout #s | ||
|
|
||
| Why multi-cluster? | ||
| * Technical reasons (5k node limit, etc) vs customer/user need | ||
| * User needs include | ||
| - CI/CD | ||
| - High availability | ||
| - Blast radius (or is this a technical detail?) | ||
| - Isolated environments | ||
| - Data locality/federal regulations | ||
| - Disaster recovery | ||
|
|
||
| Service meshes are an additional step for customers to migrate existing applications | ||
| to, but federation is needed now. This leads to custom solutions. | ||
|
|
||
| What are multi-cluster building blocks? | ||
| * Deployment | ||
| * Service mesh. But this is too big and meshes are too opinionated, should be | ||
| broken into layers | ||
| - Service discovery | ||
| - Connectivity | ||
| - Ingress | ||
| - Routing | ||
| - Signals: alerts/health/observability | ||
|
|
||
| Fundamental problems that define multicluster: | ||
| * Resource usage / technical limitations of data centers | ||
| * Administrative groups: independently run clusters that use common organizational | ||
| security policies | ||
| * Coordination across clusters | ||
| * Deployment | ||
| * Rollout new app | ||
| * Cascading triggers (service X launches which unblocks service Y to launch etc) | ||
|
|
||
|
|
||
| #### Key learnings / takeaways | ||
|
|
||
| Multi-cluster is being used today, but in ad-hoc and specific ways. Common usage | ||
| patterns need to be discovered and standardized in order to learn how k8s needs | ||
| to be extended to support multi-cluster well. | ||
|
|
||
|
|
||
| #### Action items | ||
| * Survey on customer needs | ||
| - Rather than current usage, as that can be driven by legacy accidents | ||
| * Start defining building block/concepts | ||
| * Actually we also need to survey current solutions in order to identify building blocks | ||
| * Focus on defining minimal set of building blocks to avoid the usual k8s explosion. |