[LFX Mentorship T-1 2026] Kubernetes Multi-Cluster Plugin Enhancement #6446

[mohammedfirdouss]

Overview

This issue tracks the LFX Mentorship project for Term 1 2026, focused on enhancing the PipeCD v1 plugin architecture. Specifically, this project aims to complete the multi-cluster deployment support for Kubernetes applications as outlined in RFC-0014.

Background

PipeCD v1 introduced a plugin-based architecture, and while a kubernetes_multicluster plugin exists, it currently only supports basic QuickSync and Rollback functions.

To support enterprise-grade reliability (multi-region availability, backup systems, and localized data residency), PipeCD needs to bridge the gap between single-cluster feature parity and multi-cluster flexibility. Currently, advanced deployment strategies like Canary and Baseline rollouts are missing for multi-cluster environments.

Current Status

Feature	Status	Notes
QuickSync (K8S_MULTI_SYNC)	Done	Deploys to multiple clusters at once
Rollback (K8S_MULTI_ROLLBACK)	Done	Handles partial failures
Aggregated Livestate	Partial	Functional, but has TODOs for ignore configs
Pipeline Sync stages	Not started	Canary, baseline, primary rollout missing
Traffic routing	Not started	Cross-cluster traffic management missing

---

Project Goals

Pipeline Sync Enhancements

• Implement missing stages from RFC-0014: K8S_CANARY_ROLLOUT, K8S_CANARY_CLEAN, K8S_BASELINE_ROLLOUT, K8S_BASELINE_CLEAN, and K8S_PRIMARY_ROLLOUT.
• Per-stage multi-target support: Allow different clusters to be targeted at specific stages of the pipeline.
• Failure isolation: Ensure that a deployment failure in one cluster does not unnecessarily halt the process for others.

Reliability & Drift Detection

• Livestate Improvements: Implement missing ignore configs for more accurate state reporting.
• Logic Refinement: Resolve DetermineStrategy TODOs in the codebase to allow for intelligent strategy selection across multiple targets.

Documentation & Testing

• User Docs: Update documentation to reflect the new multi-cluster capabilities and configuration options.
• Testing: Create comprehensive integration tests and example applications in the example/ directory.

---

Expected Outcomes

A fully functional multi-cluster Kubernetes plugin that supports advanced deployment strategies. Success is measured by the ability to execute a Canary rollout across multiple clusters simultaneously, promote to Primary, and handle drift detection correctly across all targets without "unimplemented" errors.

Ideal Mentee

While this project is highly technical, we are looking for mentees with:

• Go programming skills
• Good understanding of Kubernetes
• Understanding of CD strategies
• Familiarity with gRPC and asynchronous communication patterns.
• Familiarity with open source workflows and version control.

Mentors

• Eeshaan Sawant @eeshaanSA
• Khanh Tran @khanhtc1202
• Shinnosuke Sawada-Dazai @Warashi

Timeline

In accordance with the LFX Mentorship Term 1 2026 timeline.

Other Resources

• RFC-0014: Multi-cluster deployment for K8s
• Kubernetes Single-Cluster Plugin Implementation: Kubernetes plugin (single-cluster)
• Target Plugin: Kubernetes multicluster plugin
• Developer Framework: Plugin SDK

• CNCF Slack: Join the PipeCD Slack Channel here.

Interested in contributing as an LFX Mentee? Submit your application to the LFX Mentorship portal before the deadline!

[mohammedfirdouss]

What do you guys think about this? @khanhtc1202 @eeshaanSA @ffjlabo @Warashi

[eeshaanSA]

Hey @mohammedfirdouss, this might be a good idea. Although we are late, if approved, we might be able to submit this for Term 1. Let me escalate this.

@khanhtc1202

[Ayushmore1214]

@eeshaanSA @mohammedfirdouss @khanhtc1202 One possible concern is the overall scope, which is quite large and touches multiple core areas (pipeline stages, strategy, livestate, drift detection, and testing). It might help to clearly prioritize “pipeline sync stages + multi-target execution” as the core deliverable, and treat traffic routing and plan preview as stretch goals.Also, traffic routing in multi-cluster setups can vary a lot depending on networking, so keeping it minimal or experimental at first could reduce risk.
Since LFX is usually beginner-focused, this proposal may feel quite heavy and stressful due to the depth of changes involved. Narrowing the scope initially could make it more approachable while still having strong impact. Just my two bits on this

[khanhtc1202]

@mohammedfirdouss Thanks for the proposal, will add this to the LFX proposal of PipeCD 👍
Thanks @Ayushmore1214 for sharing your point, but I don't think LFX mentorship program limit the task to be begineer-focused only, so I think it's a valid one.

[mohammedfirdouss]

Hey @mohammedfirdouss, this might be a good idea. Although we are late, if approved, we might be able to submit this for Term 1. Let me escalate this.

@khanhtc1202

Thanks so much to the team for paying attention to this issue. Really excited about this @eeshaanSA @khanhtc1202

[mohammedfirdouss]

@eeshaanSA @mohammedfirdouss @khanhtc1202 One possible concern is the overall scope, which is quite large and touches multiple core areas (pipeline stages, strategy, livestate, drift detection, and testing). It might help to clearly prioritize “pipeline sync stages + multi-target execution” as the core deliverable, and treat traffic routing and plan preview as stretch goals.Also, traffic routing in multi-cluster setups can vary a lot depending on networking, so keeping it minimal or experimental at first could reduce risk. Since LFX is usually beginner-focused, this proposal may feel quite heavy and stressful due to the depth of changes involved. Narrowing the scope initially could make it more approachable while still having strong impact. Just my two bits on this

@eeshaanSA @mohammedfirdouss @khanhtc1202 One possible concern is the overall scope, which is quite large and touches multiple core areas (pipeline stages, strategy, livestate, drift detection, and testing). It might help to clearly prioritize “pipeline sync stages + multi-target execution” as the core deliverable, and treat traffic routing and plan preview as stretch goals.Also, traffic routing in multi-cluster setups can vary a lot depending on networking, so keeping it minimal or experimental at first could reduce risk. Since LFX is usually beginner-focused, this proposal may feel quite heavy and stressful due to the depth of changes involved. Narrowing the scope initially could make it more approachable while still having strong impact. Just my two bits on this

Awesome suggestion @Ayushmore1214, this would be discussed even further during meetings or on slack with the team, LFX is really not beginner focused, each project requires at least familiarity with some tools or the projects skillsets. But this is still all about learning.

[mohammedfirdouss]

@mohammedfirdouss Thanks for the proposal, will add this to the LFX proposal of PipeCD 👍 Thanks @Ayushmore1214 for sharing your point, but I don't think LFX mentorship program limit the task to be begineer-focused only, so I think it's a valid one.

Thanks so much for the review @khanhtc1202

[eeshaanSA]

@mohammedfirdouss, could you edit and make this issue inline with the templating I used in the community growth issue? Will help contributors understand everything better.

We need to update the title and body.

[mohammedfirdouss]

@mohammedfirdouss, could you edit and make this issue inline with the templating I used in the community growth issue? Will help contributors understand everything better.

We need to update the title and body.

Sure no problem, i would so shortly @eeshaanSA

[mohammedfirdouss]

@eeshaanSA Is this alright now?