2025-10-30 (Thursday)

[AkihiroSuda]

How to join

• Date: 2025-10-30 (Thursday) 05:30 UTC (Wednesday 22:30 PDT, 06:30 CET, 11:00 IST, 14:30 JST) - 06:30
  • Note: Summer time will end on 2025-10-26 in Europe, 2025-11-02 in the U.S.
• Zoom link: https://zoom-lfx.platform.linuxfoundation.org/meetings/lima

The meetings are open to anyone, not just for maintainers and contributors.

Attendees

@AkihiroSuda @jandubois @abiosoft @unsuman @balajiv113 @afbjorklund

Agenda

Note taker: @AkihiroSuda

Lima v2.0

• Now beta.0 https://github.com/lima-vm/lima/releases/tag/v2.0.0-beta.0

krunkit

• Merged in v2.0.0-beta.0 https://lima-vm.io/docs/config/vmtype/krunkit/

• llama.cpp (via ramalama) detects the GPU and makes some output, but the output seems broken. A bug of llama.cpp, libkrun, or one of their dependencies?

v2.0 ETA

• Sync with containerd v2.2 (2025-11-05) -- @AkihiroSuda

Remaining issues and PRs

• https://github.com/lima-vm/lima/milestone/59

Nothing seems a blocker for releasing v2.0 GA -- @AkihiroSuda

• autostart: Ensure instance is started/stopped by launchd or systemctl #4139
  • The amount is huge to review, also concerned about the maintenance cost with other possible combinations of limactl start CLI flags -- @AkihiroSuda
  • The regression risk seems low, as it only touched autostart stuffs -- @jandubois
  • We should try to review and merge in v2.0, but we can also postpone it to v2.1+ -- @AkihiroSuda
• Use the VM’s IP address on the same subnet as the host OS. #4175
  • How relates to gRPC and SSH forwarder? -- @jandubois
    • Basically it is the third forwarder that replaces gRPC and SSH -- @AkihiroSuda
  • Benefit compared to just use the IP, without using the portfwd ? -- @balajiv113
    • No need to remember the IP, that seems to be only the benefit I see -- @AkihiroSuda
  • Does it work with QEMU too? -- @jandubois
    • Probably yes with socket_vmnet -- @AkihiroSuda
  • Needs to consider how to specify the combination of the candidates of port forwarders, for each of protocols. Likely to need a YAML structure -- @jandubois
    • The YAML can be similar to abandoned PR Implement "Dual" port forwarder (SSH for TCP, GRPC for UDP) #4210 except the default value Use the VM’s IP address on the same subnet as the host OS. #4175 (comment) -- @AkihiroSuda
  • Better to postpone to v2.1+ -- @jandubois
• Minimize effects caused by restarting guestagent #4217
  • Relatively complex, but seem safe to merge -- @AkihiroSuda
• Allow all github:.../*.yaml references to be symlinks or redirects #4222
  • Will try to implement this weekend. Can be postponed to v2.1+ -- @jandubois
• Add global LIMA_BATS_ALL_TESTS_RETRIES variable #4224
  • Merged
• Add a GPU acceleration YAML field and a CLI flag #4257
  • The proper YAML should just have a single boolean. Implementation specific options should still be vmOpts -- @jandubois
  • "gpu", "gpuAccel", "gpuPassthru" ? -- @jandubois
    • "gpu" might be confusing as there is non-accelerated virtual GPU too -- @AkihiroSuda
    • "accel" is qemu's terminology, "passthru" is more descriptive -- @jandubois
    • "--gpu-passthru" might be too long? "--gpu" might be easier to use -- @AkihiroSuda
    • The simplest "--gpu" is fine we just have to have a doc to clarify that even without --gpu there is still non-accelerated GPU -- @AkihiroSuda
    • The YAML property should probably NOT go into the ".video" structure; as the GPU accel does not related to video display -- @jandubois

v1.2 EOL

• v2.0 GA + 3 months? Can be extended later on demand -- @AkihiroSuda

Tasks

• Maintainers: review the remaining PRs
• @jandubois : implement Allow all github:.../*.yaml references to be symlinks or redirects #4222
• @unsuman : implement Add a GPU acceleration YAML field and a CLI flag #4257
• @AkihiroSuda : bump up containerd to v2.2 when it is ready (2025-11-05) and release v2.0 GA with that
  • Probably have an RC with containerd v2.2 RC before that

Next meeting

• 2025-11-13 (Thursday) 10:15-14:00 @ KubeCon North America (Atlanta) - Building B | Level 1 | Exhibit Hall B3-B5 | Solutions Showcase

• 2025-11-20 (Thursday)

[AkihiroSuda]

AI-generated meeting note

Quick Recap

The team discussed the release of several software components including Airbnbread and Browser 2.2, with plans to address various integration issues and bugs before the scheduled release dates. They reviewed multiple open issues and agreed to merge some while postponing others to a later version, including discussions about auto-start functionality and port forwarding configurations. The team also covered GPU acceleration implementation in QEMU and related settings, with plans to release Browser 2.2 RC before November 5th and the final release on November 8th.

Next Steps

• Jan: Review the auto start PR, focusing on macOS side testing
• Jan: Implement additional functionality for the GitHub URL scheme
• Akihiro: Merge the auto start PR after confirming tests are passing
• Akihiro: Release Lima version 2.0 RC with Container D 2.2 RC before November 5th
• Akihiro: Decide on Python 1.2 EOL - maintain for 3 months after version 2.0 release
• Akihiro: Postpone direct IP port forwarding PR to version 2.1
• Akihiro: Add GPU YAML field and CRI flag for GPU support
• Ansuman: Work on adding GPU YAML field and flag
• Ansuman: Test GPU inferencing with CPU to verify GPU acceleration is working properly

Summary

Airbnbread and Browser Release Updates:
Akihiro discussed the release of Airbnbread and its integration with Tehran Teach, which now includes GPU acceleration. He noted some issues with the output from Damage CPP, which he believes may be a bug, and mentioned that Ramadama is aware of the problem. Akihiro also outlined plans for the release of Browser 2.2, which is scheduled for November 5th, along with Browser 2.0 and a minimal version. He emphasized the need to merge remaining issues before the 2.0 release and considered maintaining Browser 1.02 for three more months after the release of Browser 2.0.

Auto-Start Merge and Issue Review:
Jan and Akihiro discussed the status of several open issues and agreed to merge some that were ready, while postponing others to version 2.1. They reviewed the auto-start functionality, noting that while some tests were failing, they could proceed with merging as it was a low-risk change. Jan agreed to review the auto-start feature, particularly the macOS side, but noted he might not have time until the weekend.

Direct IP Port Forwarding Implementation:
The team discussed direct IP port forwarding functionality, particularly for BM's IP address with PCNAT. Akihiro explained that this feature would bypass gRPC and SSH tunneling when direct IP access is available, using a new dial function that connects directly to the host network. Jan expressed confusion about how this fits with existing SSH and gRPC port forwarder settings in Lima.yaml, and Balaji clarified that this implementation is intended to provide an alternative to gRPC tunnels when VZNAT or similar solutions provide direct IP access.

Port Forwarding Configuration Simplification:
Jan and Akihiro discussed the configuration of port forwarding settings in Lima.yaml, focusing on the need to make them less confusing for users. They agreed that different variables should be used for TCP and UDP, and that the settings should be moved from environment variables to Lima.yaml. Jan suggested postponing the implementation to a later release, as they were making decisions on the fly and wanted to mark it as experimental.

Rental Price and CEI Updates:
Akihiro discussed a rental price update and a strategy for fixing CEI failures, which Jan confirmed was safe to merge. Jan mentioned he needed to implement a part of a project but could not promise a timeline, indicating it might be moved to version 1 if not completed. The conversation ended with Jan adding Anders to the attendee list.

QEMU GPU Support Implementation:
The team discussed implementing GPU support in QEMU, with Akihiro proposing to add a Boolean option for GPU acceleration in the generic options. Jan suggested using "GPU pass-through" as it might be more self-explanatory, though "GPU Excel" was also considered. The team agreed to use a Boolean option for now, with the possibility of adding more specific GPU settings in the future.

GPU Settings and Release Planning:
The team discussed GPU-related settings and documentation, agreeing to use "GPU" as the CLI option name unless future use cases require a different name. Jan committed to reviewing the auto starts PR and implementing additional GitHub URL scheme functionality. Akihiro confirmed that the GPU acceleration is working, though there are some output issues with the current implementation. The team agreed to release the 2.2 RC before November 5th, with the final release scheduled for November 8th, ahead of Akihiro's departure for KubeCon.

[AkihiroSuda]

Corrections

The team discussed the release of several software components including Airbnbread and Browser 2.2

What are "Airbnbread" and "Browser" ?

Akihiro: Release Lima version 2.0 RC with Container D 2.2 RC before November 5th

Container D → containerd

Akihiro: Decide on Python 1.2 EOL - maintain for 3 months after version 2.0 release

I didn't talk about the 1995 edition of Python 🤷

Akihiro: Add GPU YAML field and CRI flag for GPU support

CRI → CLI

Akihiro discussed the release of Airbnbread and its integration with Tehran Teach, which now includes GPU acceleration.

What is "Airbnbread" ? "Tehran Teach" ?

He noted some issues with the output from Damage CPP, which he believes may be a bug, and mentioned that Ramadama is aware of the problem

• What is "Damage CPP" ?
• Ramadama → Ramalama

considered maintaining Browser 1.02 for three more months after the release of Browser 2.0.

• Browser 1.02 → Lima 1.2
• Browser 2.0 → Lima 2.0

The team discussed direct IP port forwarding functionality, particularly for BM's IP address with PCNAT.

• BM →　VM
• PCNAT → VZNAT

Rental Price and CEI Updates

???

GPU Excel

Excel → Accel