About
Sunday, November 8, 2026
Salt Lake City, UT
Salt Palace Convention Center
South Entrance (200)
Salt Lake City, UT 84101
#CNMaintainerSummit
MAINTAINER SUMMIT IS SOLD OUT.
ONSITE REGISTRATION IS NOT AVAILABLE.
The CNCF Maintainer Summit is an exclusive event for the people behind our projects to gather face-to-face, collaborate, and celebrate the projects that make “Cloud Native.” Programming will be focused on sharing best practices, diving into contributing processes, and solving common problems across projects to enrich our great community of maintainers. Interested in learning more? Check out the Maintainer Summit Europe 2026 post-event report.
The Schedule is Now Live!
Eligibility
Who can attend Maintainer Summit?
Additional Maintainer Summit Opportunities for Incubating + Graduated Project Maintainers, TAG Leads & TOC Members:
| Incubating Maintainers* + TAG Chairs, Technical Leads and Working Group Leads | Submit CFP Eligible for Program Committee |
| Graduated Maintainers* + Org Members + TOC | Submit CFP Eligible for Program Committee Eligible for Program Chair Project Photo |
*Your GitHub Handle MUST exist in:
https://github.com/cncf/foundation/blob/main/project-maintainers.csv
OR
https://github.com/cncf/foundation/blob/main/project-maintainers-.project.csv
Registration Details
This is not an event for general attendance. See the “Eligibility” section for guidelines. To attend, you must be registered for KubeCon + CloudNativeCon. The registration deadline for Maintainer Summit is 11:59 MST on Thursday, November 5, 2026.
Sponsor the Event
Contact [email protected] to secure your sponsorship today! Signed contracts must be received by September 25, 2026.
FAQ – Registration
When is the Maintainer Summit taking place?
Sunday, November 8, one day prior to KubeCon + CloudNativeCon North America 2026.
Where is the Maintainer Summit located?
Salt Palace Convention Center | 90 South West Temple, Salt Lake City, UT 84101
Badge Pick-Up is located at the South Entrance.
Who can attend the Maintainer Summit?
- Sandbox Project Maintainers (as listed in project-maintainers.csv)
- Incubating & Graduated Project GitHub Org Members
- CNCF TAG Leads & the CNCF TOC
Why do I need to be a CNCF Project GitHub Org member to attend?
We welcome everyone to start contributing to a project, but with 7,500+ CNCF Project Org members and limited space, we’ve decided to limit attendance to those who are already actively engaged a project.
I’m not listed as someone who could attend, but I contribute, what should I do?
We encourage you to continue to contribute and pursue your project’s path for becoming a GitHub Org member or climbing the contributor ladder to become a listed maintainer. Until then, there are plenty of other opportunities to get involved at KubeCon + CloudNativeCon. Check out the schedule and get involved.
What is a sponsored attendee?
- If you are a Maintainer, TAG Chair, Tech Lead, K8s SIG Lead, or WG Lead, and there is someone who is not yet a GH org member, you can sponsor them to attend the summit. If you are sponsoring an attendee, proactively contact [email protected] with the individual’s name, company, and community participation details.
- The sponsored attendee must indicate that they are a sponsored attendee during registration and share the name of the individual sponsoring them.
- Sponsored attendees will be reviewed weekly, note not all sponsored attendees will be granted registration.
How will I know if I’m registered for the Maintainer Summit?
After registering, your registration will be placed into a pending queue for review. Once a week, CNCF staff will validate that you meet the criteria to attend and will either approve or deny your registration. If you believe a mistake was made, please contact [email protected].
If I’m accepted as a Maintainer Summit speaker – do I also have to purchase registration for KubeCon + CloudNativeCon?
No, all CFP-submitted accepted maintainer summit speakers will be provided with an “All-Access” pass for CNCF-hosted co-located events and KubeCon + CloudNativeCon programming. (Note: This does not include proposals for Project Meetings)
I’m leading an unconference session at the summit. Do I need to purchase registration for KubeCon and CloudNativeCon?
Yes, due to the short time frame for planning unconference sessions, complimentary registration will not be provided.
Can I attend virtually?
Unfortunately, no, there is not an opportunity to attend virtually.
Do I need to register for KubeCon + CloudNativeCon to attend the summit?
Yes, to attend The Maintainer Summit, you must be registered for KubeCon + CloudNativeCon.
Is there maintainer support for attending?
Please apply for the Dan Kohn Maintainer Scholarship. Scholarship information is coming soon!
Will there be a registration option only to attend the Maintainer Summit and not KubeCon + CloudNativeCon?
No, there is NOT a single-day Maintainer Summit-only pass available.
FAQ – Sponsorships
Why are sponsors needed for a Maintainer Summit?
Community partners ensure that the Maintainer Summits are financially sustainable. Sponsor support helps cover event hard costs and provides opportunities for sponsor companies to engage as a thought-leader in the community.
Who can we reach out to about sponsorship opportunities?
Please email [email protected]
FAQ – Misc
Will the sessions be recorded? If so, where will they be posted?
All but the unconference sessions will be recorded. From previous experience, the more open and discussion-oriented unconference talks are, it is difficult to capture when recording. For those sessions, we encourage attendees to keep a shared notes doc and make those available after the event. For the sessions that are recorded, they should be posted within two weeks of the event.
I have a question! How do I contact the event staff?
Email us at [email protected] or ask a question on the #maintainer-summit Slack channel.
Schedule
Sessions will be recorded and available on the CNCF YouTube channel within two weeks.
Times shown in MST (UTC-7). Seating is first come, first served.
Plan your sessions and build your personal agenda.
Learn how to use the event app and sync favorites across devices.
Got your badge? Next, make sure to visit any Info Desk for an ID check and a 21+ wristband to redeem your drink tickets at event receptions.
Info Desk ID Check Hours:
Sunday, November 8 | 8:00 AM – 5:00 PM
Monday, November 9 | 7:00 AM – 7:00 PM*
Tuesday, November 10 | 7:30 AM – 6:15 PM*
*ID check stations will also be available during the reception.
When Crossplane graduated within the CNCF late last year, and Kyverno followed shortly after, the votes marked the culmination of years of work to mature each project and meet the expectations of the TOC. While both projects ultimately reached the same milestone, their journeys looked quite different. In this session, maintainers from Crossplane and Kyverno walk through what it really took to get there.
Crossplane’s first graduation application wasn't quite ready: the TOC raised concerns about vendor neutrality, long-term survivability, and the health of our extension ecosystem. We'll share how the project put this feedback into action by launching a vendor-neutral community registry, migrating release infrastructure to CNCF-owned accounts, and instituting new governance and lifecycle policies for our ecosystem.
Kyverno's first graduation push wasn’t ready either, however their path presented a different set of challenges. We'll discuss how the project addressed questions around maintainer diversity and long-term project sustainability, strengthened the contributor community across organizations, worked through technical, governance, and security due diligence, and partnered closely with the TOC throughout the process.
Together, these experiences highlight that graduation is not about following a single checklist. It's about demonstrating that a project has the technical maturity, governance, and community needed to thrive for years to come.
We'll also cover the follow-up commitments both projects made to the TOC after graduation, because the work of maturing a project never really ends.
Wherever your project sits on the maturity curve, come hear what we went through, the lessons we learned, and how you can make the CNCF maturity process work for your project and community.
Congratulations: you're officially an open source maintainer! But once the excitement fades, reality hits. You aren’t just writing code anymore; you’re triaging issues, managing community expectations, and protecting project stability. It’s an entirely different job.
This talk is a survival guide for making the leap from "PR author" to "project steward" without losing your mind along the way.
You'll learn how to navigate the psychological leap from writing code to reviewing, steering, and mentoring, how to graciously reject out-of-scope features while keeping contributors engaged, and practical tactics to protect your limited time.
Whether you just got maintainer rights or are aiming for them, you'll leave with actionable strategies to sustain your project and your sanity.
Menu:
Banbury Cross Assorted Donuts
Cinnamon Rolls (GF, VEG)
Envoy sits in the request path of much of the internet. A bug that slips through review can become a CVE with a coordinated disclosure, an embargo, and patch releases across supported versions. That math changed when AI-generated pull requests arrived. They look polished, they compile, they pass CI, and they are wrong in ways human contributions rarely are.
This talk shares the review playbook Envoy maintainers built in response. We cover the failure patterns we see in AI-written changes, how we adjusted reviewer expectations without banning AI outright, the signals that separate a careful contributor using AI from drive-by slop, and how our security team treats these changes in sensitive code paths like parsers and filters.
Attendees leave with a review checklist, contributor policy language they can copy, and real numbers on what this costs maintainer time. Every lesson applies to any project where a bad merge has consequences beyond a revert.
After 10 years of Kubernetes, maintainers across the ecosystem still independently answer the same question: is this resource healthy? Every CRD defines status, conditions, and readiness differently, so every tool – Argo CD, Flux, kubectl plugins, dashboards, CI/CD systems – ends up reimplementing health interpretation for the same resources, causing duplicated work, inconsistent behavior, and operational drift.
This session is a working discussion. Together, we'll explore what a shared specification for CRD health and readiness could look like, building on efforts such as kstatus. We'll discuss how conformance should be achieved: through a shared library, independent implementations against a common specification and conformance suite, or another approach. The goal is to reach consensus on whether this is worth pursuing formally, for example through a CNCF Sandbox project or working group, and who's willing to help drive it.
TAGs are where cross-project initiatives, shared practices, and maturity guidance get built, and that work is a collaboration with maintainers rather than something separate from them.
This panel brings one representative from each of the five TAGs onto one stage to show what each group is working on right now and, concretely, how a maintainer plugs in. This session treats the relationship as a two-way street: how maintainers join initiatives that shape industry-wide practices in their domain, how TAG output such as guidance, assessments, landscape analyses, and best practices helps a project mature as it moves from sandbox to incubating to graduated, and how a problem that several projects share can become a new initiative. Instead of five status updates, we want to share five real collaboration paths.
Attendees leave knowing which initiatives are relevant to their project, how to join or start one, and how TAG expertise feeds the project journey across maturity levels.
An open discussion with SIG leads on changes to the SRC triage process due to the influx of reports.
CNCF has a well-established model for source-code stewardship. But once a release is cut, many projects hand critical parts of their relationship with users to infrastructure they do not control.
Images, charts, and other artifacts are published through a patchwork of vendor-operated registries. Once registry addresses appear in manifests, policies, mirrors, and downstream products, changing them becomes an ecosystem-wide operation.
This talk examines distribution as an overlooked part of open source governance. The distribution channel defines an artifact’s canonical identity, availability, retention, and trust metadata. It also determines whether a project can change providers without disrupting users.
Drawing on our experience as Harbor maintainers, we will give maintainers a framework for deciding which parts of the distribution path must remain under project or foundation control, which can be delegated, and where CNCF may need a shared strategy.
AI coding agents are changing how software gets built — but what does that mean for open source maintainers? This talk shares practical lessons from integrating agentic AI into real CNCF project workflows: due diligence evaluation, project health monitoring, adopter interviews, and governance reviews.
We'll cover three things every maintainer needs to understand: skills (structured, composable units of agent work), evals (how you know an agent's output is correct before it ships), and observability (how you monitor agent behavior the same way you monitor your services).
Maintainers are already dealing with AI-generated PRs, bot-authored commits, and agent-driven reviews. We'll share patterns for integrating agents into your own project workflows: CI/CD, issue triage, release management, documentation, and the eval and observability infrastructure that makes them trustworthy rather than risky.
Concrete examples from both TOC governance workflows and CNCF project adoption patterns.
CNCF project reviews are essential for helping open source projects improve quality, sustainability, security, and overall maturity. Yet many community members are interested in contributing to these reviews and do not know where to start.
In this interactive session, we will explore the CNCF project review assets and discuss how contributors can participate in evaluating and strengthening projects across the cloud native ecosystem. We will cover Governance Reviews, Technical Reviews, General Technical Reviews, and Security Reviews, explaining their purpose, structure, and role in supporting projects through Sandbox, Incubation, and Graduation.
The session will also gather community feedback on existing documentation and processes, identify opportunities for improvement, and discuss how to make participation more accessible for reviewers and maintainers. Attendees will gain practical insights into one of the key mechanisms that supports CNCF project maturity and ecosystem health.
Menu:
Marinated Asian Noodle Salad: Baby Spinach, Napa Cabbage, Cashews, Mandarin Oranges, Honey Mustard Dressing (DF, VEG)
Mixed Green Salad: Hearts of Palm, Sprouts, Cherry Tomatoes, Ginger Soy Dressing (GF, DF, V, VEG)
Szechuan Beef, Peppers, Onions, Eggplant (GF, DF)
Soy Ginger Glazed Salmon, Braised Baby Bok Choy (GF, DF)
Hoisin-Grilled Chicken Thighs, Charred Scallions (GF)
Vegetable Fried Rice (GF, DF, V, VEG)
Stir-Fried Vegetables (GF, DF, V, VEG)
Mango Tapioca Shots, Coconut & Ginger Panna Cotta (GF)
Kubernetes is the second largest open source project in the world, and its future is driven by an elected committee of seven people serving two-year terms. Especially in a project of our size, it’s important to work publicly and in the open, so join us for an open-format interactive session with representatives of the Kubernetes Steering Committee. Bring your questions about project governance, the future of Kubernetes, how you can get involved (or more involved!), or anything else you can think of that we might be able to help you with. A form will be provided to ask questions anonymously.
Get your questions answered live by representatives from the Technical Oversight Committee in this open-format session. Come prepared for an engaging and interactive discussion!
2026 has been a banner year for vulnerability reports. The containerd project has received 2 orders of magnitude more vulnerability reports in 2026 than in any other year, and more than 3.5x as many reports in 2026 than all of 2020-2025 combined. There are needles (real findings) in the haystack too.
What are projects doing to cope? Are maintainers feeling burnout? Can we share any tools, agent skills, etc?
Session leads: Samuel Karp, Akhil Mohan
______________
SIG-Network Gateway API subproject meeting
Join us to discuss the future of Gateway API, next needs, what is going on, how you can help.
The Karmada maintainers and contributors will gather for a focused project meeting to discuss upcoming features, roadmap priorities, and community collaboration initiatives.
Meeting Agenda:
- Release Planning: Discuss the next major release timeline, feature priorities, and breaking changes
- Architecture Discussions: Review proposed enhancements to multi-cluster scheduling and workload distribution
- Community Growth: Strategies for expanding contributor base and improving documentation
- Technical Challenges: Address current issues in cluster federation and resource propagation
- Cross-Project Collaboration: Explore integration opportunities with other CNCF projects
This face-to-face meeting provides a valuable opportunity for the Karmada community to align on technical direction, resolve blockers, and strengthen collaboration among distributed team members.
Target Audience & Relevant Projects
Kubernetes Maintainers & Contributors: SIG Node, SIG Scheduling, SIG Architecture, SIG Storage, SIG Network
Container Runtimes & Sandboxing: containerd, CRI-O, Kata Containers, gVisor, Cloud Hypervisor
AI & Agent Platforms: Engineers building high-density, dynamic agent executors or sandbox environments
Problem Statement
Autonomous AI agents represent the fastest-growing workload category in cloud infrastructure, yet their operational profile inverts almost every fundamental assumption behind Kubernetes container orchestration:
Extreme Idleness: Agent sessions spend 90–95%+ of their lifespan idle waiting on model inference, tool I/O, or human approval, punctuated by high burstiness on wake-up.
Extreme Density: Production systems require packing 500–1000+ isolated micro-sandboxes per host rather than the traditional ~110 Pods per node.
Sub-Second Latencies: End users and RL agent loops demand sub-second sandbox resumption (P95 ≤ 500ms), far faster than traditional Pod scheduling cycles.
Untrusted Code Execution: Agents execute arbitrary, model-generated code next to sensitive tools and credentials, demanding lightweight hardware-level or kernel isolation (microVMs/kata/gVisor).
Because the Kubernetes Pod contract is fixed at creation time and every state change writes through etcd, emerging agentic frameworks (such as Agent Substrate, and many others) are increasingly routing around Kubernetes APIs on their hot paths, building custom out-of-tree control planes, running nested runtimes, and treating Kubernetes merely as an over-provisioned VM manager.
If Kubernetes does not evolve its process layer, the cloud-native ecosystem risks fracturing into proprietary sandbox fabrics.
Proposed Discussion Topic
This session will bring together maintainers across Kubernetes and the agentic ecosystem to align on how the core primitives should evolve:
Mapping the Frictions:
Where does the current Pod contract break down for agentic workloads?
Walkthrough of real-world workarounds: bypassing etcd, nested sandboxing, warm worker pools, and memory stranding.
Kubernetes as the Distributed Kernel:
Defining the boundary: What belongs in Kubernetes (Process/Kernel layer) vs. what belongs in agent frameworks (Userspace)?
The "Two-Level Control" model: How can the Kubelet delegate local actor/container management within a dynamic Pod envelope without overwhelming etcd?
Concrete Upstream Touchpoints:
SIG Node & CRI: Unprivileged sandboxing/microVMs (VM-as-a-container), dynamic container creation/destruction, and native suspend/resume/freeze lifecycle states.
Resource Management: Pod-level resource sharing, in-place resizing, workload-controlled swap/MGLRU, and node overcommit for idle states.
Networking & Identity: Dynamic endpoints and wake-on-traffic routing for suspended actors.
Action Items & Working Group Alignment:
Can we establish a shared cross-SIG/cross-project effort (e.g., under SIG Node / TAG Runtime) to formalize agentic requirements into KEPs?
Desired Outcomes
A shared understanding of agentic workload requirements across Kubernetes and sandbox runtime maintainers.
Agreement on architectural boundaries (what K8s should standardize vs. what runtimes/frameworks should handle).
Identification of concrete KEP targets and collaboration channels for 2027 development cycles.
Session leads: Dawn Chen and Antonio Ojea
Agents are starting to write and run their own CUDA code, and people want to sandbox it in environments where running a VM is not feasible. gVisor is a logical choice when virtualization isn't possible, but its GPU support lags behind. Each driver release has to be tracked by hand, only a few driver versions are qualified, MIG and multi-node NVLink aren't supported yet, and checkpoint/restore with a GPU doesn't work for every workload. This session is for comparing what issues people have run into when running workloads with GPUs on gVisor and working out what is worth fixing first. Anyone who maintains gVisor or runs sandboxes using it is welcome.
Questions to discuss
How can gVisor keep up with every GPU driver release, for example with a machine-readable description of the driver ABI?
How do we get new GPU generations and features, such as MIG and multi-node NVLink, supported before they ship?
What does GPU checkpoint/restore need to work for more workloads, including multi-GPU and ARM64?
What should a shared test and CI setup for GPU changes look like?
Session Lead: Kevin Klues
Sustaining trusted open source projects in this new environment raises several pressing questions, with two core challenges standing out:
Trust: Maintaining and deepening trust with both users and contributors.
Triage: Cutting through increased noise to identify valuable contributions and high-priority bugs as incoming volume accelerates.
I don't think 1) is solvable without some component of 2).
How do we build genuine trust when pull requests and review comments are increasingly automated? How do we safeguard against future supply-chain threats like the xz utils backdoor in this environment?
What does advancement on the contributor ladder mean when a maintainer spends hours reviewing code, only for the author to prompt an AI agent to address the feedback? Who actually performed the work, and who gained a deep understanding of the project's codebase and architecture?
How do we effectively filter incoming changes when individual developers can generate massive volumes of pull requests with minimal effort? What stops malicious actors from trivially gaming these workflows, and how do we ensure legitimate contributors remain visible through the noise?
While large-scale projects like Kubernetes have faced similar challenges before, the scale and velocity are escalating rapidly.
In my opinion these questions are far more critical than surface-level metrics like PR merge velocity or initial review response times. Throughput does not equate to value or quality.
We must maintain confidence among our community and downstream users that we are building a secure, high-quality infrastructure foundation.
We need strategies to cut through low-effort automated submissions, highlight high-value contributions, and adapt how we mentor and cultivate future maintainers.
The goal of this session is to brainstorm actionable solutions and guidance for CNCF project communities.
Session Leads: Benjamin Elder & Maciej Szulik
Live SIG-etcd community meeting, at which we will work on etcd 3.8 planning and deliverables, as well as subprojects like benchmarking.
What projects are involved?
Graceful Node Shutdown: what it should mean, and how workload controllers find out
What projects are involved?
Kubernetes: SIG Node, SIG Apps, SIG Cloud Provider, SIG Windows, WG Node Lifecycle
Target Audience & Relevant Projects
Kubernetes maintainers and contributors: SIG Node, SIG Apps, SIG Cloud Provider, SIG Windows, WG Node Lifecycle
Operators who run long shutdown windows (30 minutes or more) with per-priority termination, or who have seen DaemonSets not come back after a reboot
Anyone building drain or maintenance tooling on top of Node Lifecycle Conditions (Specialized Lifecycle Management kubernetes/enhancements#5683)
Problem Statement
Graceful Node Shutdown (kubernetes/enhancements#2000) has been beta and on by default since v1.21; priority-based shutdown (kubernetes/enhancements#2712) since v1.24. Neither has moved since. Writing the follow-up KEPs this year showed that the semantics were never settled and that nothing outside the node can see a shutdown happening:
The kubelet terminates by priority but admits by nothing. shutdownGracePeriodByPodPriority ends Pods in tiers, lowest priority first. Admit() rejects every new Pod for the whole window, on Linux and Windows. A system-critical Pod that needs to be created during the user-workload tier is rejected like everything else (Graceful Node Shutdown should allow admission of high priority pods kubernetes/kubernetes#142521).
The DaemonSet controller does not know. It creates a Pod on the node, the kubelet rejects it, the controller deletes it and creates another, until the node is gone (DaemonSet controller and Graceful Node Shutdown manager disagree when making workloads placement decision kubernetes/kubernetes#122912, EKS 1.34 DaemonSet Controller created a lot of pod during node deletion kubernetes/kubernetes#137895).
Nothing leaves the node. Node Lifecycle Conditions landed in v1.37 as admin-managed conditions. No kubelet writes GracefulNodeShutdownInProgress yet, so controllers, autoscalers and drain tooling have no signal to read.
The failure modes are not written down. A kubelet restarted mid-shutdown has no record that one was in progress, and operators have reported DaemonSets that did not restart after a GNS reboot.
Short window or long one? Is GNS a short window where nothing new runs, or a 30-plus-minute priority-ordered termination? The two answers want different APIs, and the same question applies to drain.
Proposed Discussion Topic
The plan today: tier-aware admission in the kubelet (v1.39, under the KEP-2000 update, with KEP-2712 updated alongside); the kubelet setting GracefulNodeShutdownInProgress and publishing the priority boundary it is terminating at; a DaemonSet controller reader that holds back Pods below that boundary (SLM: Publish Graceful Node Shutdown state for DaemonSet coordination kubernetes/enhancements#6249); GNS as an EvictionRequest requester (EvictionRequest API kubernetes/enhancements#4563) as the graduation step. The boundary needs a carrier, and there are two paths:
A) PodDeletionPreference. The kubelet writes a priorityBelow rule that any controller can read, so the published API is not GNS-specific and the same object serves scale-down scoring later. Depends on the PDP KEP landing with a priority-threshold rule and node scoping.
B) A GNS-specific Node status signal. The kubelet publishes the boundary on the Node object (for example a kubelet-owned HighestPriorityAllowed field). No new dependency, but SIG Node carries a GNS-only Node API and every consumer learns a GNS-specific field.
Which parts of that plan belong to the kubelet, which to workload controllers, and which to drain tooling
Who else should consume the signal: StatefulSet rollouts (SLM: StatefulSet rollouts respect Node lifecycle conditions kubernetes/enhancements#6265), Job placement (SLM: Job Pod replacement during node maintenance kubernetes/enhancements#6262), scale-down scoring
Several writers of the same Node condition (kubelet, kubectl drain, cluster autoscaler, admins): who sets, who clears, what happens when the node never returns (SLM: Coordinate Node lifecycle condition writers kubernetes/enhancements#6430)
What GA for KEP-2000 and KEP-2712 should require
Desired Outcomes
A written answer to the short-window versus priority-ordered question, or agreement on which SIG owns the answer
Agreement on the carrier for shutdown state, so the KEP-2000 update and KEP-6249 can go to review for v1.39 without reopening it
A list of consumers of the signal with a name next to each
A first cut of GA criteria for KEP-2000 and KEP-2712
Session Leads: Daniel Bruno, Ryan Hallisey, and Filip Křepinský
What projects are involved?
All CNCF
Description
DevStats has had data problems since February, and those problems are not likely to be solved. As such, we are looking towards what DevStats replacement will be — LFX Insights, CHAOSS, or something else, or some combination of tools.
If you use DevStats, or you care about project metrics, please join this discussion to share what community metrics are important to you in CNCF's future.
Session lead: Josh Berkus
What projects are involved?
Any CNCF project, including Kubernetes and its subprojects, e.g. agent-sandbox, Kueue, JobSet
Description
This is a follow-up to the Maintainer Summit EU 2026 session "How should maintainers navigate (and review) AI-based PRs?". Since March, projects have tried several approaches/tools:
stricter disclosure rules and closing low-effort PRs faster
AI pre-review bots (default-on or human-summoned) and its policy
growing reviewers and delegating ownership across code, docs, and releases
writing down conventions that used to live in maintainers' heads (CONTRIBUTING.md, AGENTS.md)
linters, validation checks, and release or CI flake automation
checked-in agent skills (either centrally in maintainer-tools repo or in each subproject repo directory)
Maintainers from Kubernetes subprojects (agent-sandbox, Kueue, JobSet) will share what we tried and how we measured it. We want to hear from other projects too, including release, docs, and security folks dealing with what happens after PRs merge.
We don't pitch or endorse any tool. We want to compare what happened in practice, where things broke, and how much depended on repo size or contributor mix. If your project tried any of this (whether you kept it or turned it off), tell us what you tried, how you could tell whether it helped, and the gotchas.
We'll post a summary back here afterward:
approaches/tools that helped maintainers, with gotchas, and when they apply (e.g. repo size, contributor mix)
how to tell if it's working, e.g. time to first maintainer response (contributor wait time), maintainer hours per release, checks that quality didn't drop (reverts, human review depth)
Out of scope: generic debates about whether AI is good or bad, or org-wide policy mandates.
Session leads: Janet Kuo and Kevin Hannon
This meeting is for OpenTelemetry Maintainers and the Governance and Technical Committees to attend, plan, and discuss OpenTelemetry initiatives, roadmap and community growth.
Menu:
Kettle Chips and French Onion Dip (GF, VEG)
Charcuterie – Assortment of sliced meats & cheese
What projects are involved?
wg-etcd-operator, sig-apps
Description
Kubernetes StatefulSet was created to support databases and cloud filesystems. The new etcd-operator project started out on StatefulSet, but found itself needing to move to creating a custom controller instead, as have a majority of database operators.
We will explain the roadblocks that compelled this change, and breifly review how database operators are dealing with pod management across the ecosystem. We'll discuss some what etcd-operator had to do to replace the StatefulSet functionality. And then we'll share the improvements that are coming to StatefulSet that might make creating a custom controller unnecessary in the future.
Session leads: Nont Thanonchai, Marciej Szulik, Josh Berkus
What projects are involved?
Kubernetes (SIG Architecture, SIG Node, SIG Apps, SIG Scheduling, SIG Auth, SIG Network, WG Device Management), and other CNCF AI related projects
Description
Kubernetes AI Conformance moved from a working group into a SIG Architecture subproject, reached 40+ certified platforms, and now has automated tests for all current MUST requirements.
Looking ahead to 2027, we're exploring what should come next, from inference and agentic workloads to any other gaps maintainers are seeing in practice. We'd like to hear from maintainers across SIGs and CNCF projects. What belongs in the platform contract next, what cross-project standards are missing, and how do we keep requirements vendor- and implementation-neutral?
Session leads: Janet Kuo, Rita Zhang, Federico Bongiovanni
What projects are involved?
OpenTelemetry
Description
The Green Software Foundation is working on the SCI for OpenTelemetry initiative, exploring how Software Carbon Intensity (SCI) can be represented through OpenTelemetry semantic conventions.
The intention is to eventually bring this work to the OpenTelemetry community, particularly the Semantic Conventions SIG, for discussion and potential adoption.
But before that happens, we want to hear from the maintainers and the community!
How should carbon and energy measurements fit into OpenTelemetry? Do we need new semantic conventions? What are the technical challenges, and how can we make sure this work aligns with the existing ecosystem?
Session lead: Diana Todea
KubeVirt Project Meetiing
Coat Check at the SPCC will close at 6:00 PM. Please collect all personal belongings before heading over to the Hyatt Regency.
If you need to store your items at the Hyatt, please visit the hotel's bell stand for assistance.
Menu:
Selection of Fresh Vegetables – Baby Carrots, Celery, Cauliflower, Cucumber, Radish, Heirloom Cherry Tomatoes, Endive, Baby Peppers VE, GF
Hummus VE, GF
Creamy Herb Onion V, GF
Green Goddess V, GF
Chef's Selection of Beehive and Park City Creamery Cheeses V, GF
Honeycomb, Dried and Fresh Fruit, Fig Jam V, GF, DF
Artisan Breads V
Lavosh V
Beef Wellington | Mushroom Duxelle
Bruschetta | Divina Roasted Tomato | Garlic | Basil | Balsamic Glaze (VE)