Let general-purpose instances start and route on a VPC network - #531
Merged
Merged
Conversation
A general-purpose instance attached to a VPC network never reached a running state: the container runtime rejected every sandbox because the network result did not describe the instance's own interface under the name the runtime asked for, so it could not find an address for it. The tap attachment now reports that interface under the requested name. The device still lives in the host namespace, which is what lets a microVM guest attach to it, and CHECK resolves the device from the attachment rather than from the reported name. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A guest that came up on a VPC network could not reach anything off-link: its kernel refused the default route, because a VPC gateway sits outside the prefix allocated to the attachment and nothing said the gateway was reachable on the link. The route handed to the guest is now marked on-link, the same treatment the host side of the attachment already gets. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
ecv
approved these changes
Sep 12, 2026
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Instances on the general-purpose runtime class could not start on a VPC network. Every sandbox was rejected and retried forever, so the instance never reached a running state, and the two defects behind that had to be cleared in order: the instance could not start, and once it started it had no route off its own link.
The network attachment now describes the instance's own interface under the name the runtime asks for, which is how a container runtime finds the instance's address and admits the sandbox. The device itself still lives in the host namespace, which is what lets a microVM guest attach to it. The default route handed to the guest is now marked on-link, so the guest kernel accepts it: a VPC gateway sits outside the prefix allocated to an attachment, and without that marking the guest comes up addressed but unable to reach anything.
Verified against the Kata release the cells run, which honours the on-link marking end to end. Unikernel instances and container workloads on the veth path are unaffected.
This unblocks the next stage of general-purpose VPC attachment rather than completing it: an instance has not yet been observed reaching the network end to end.
Follows #528. Related: datum-labs/kata-provider#13, datum-cloud/cloud#20.
Separately, the pod admission path writes the same network attachment as both a secondary and the default network, so the attachment is set up twice per sandbox. That belongs in datum-cloud/cloud and is not addressed here.
🤖 Generated with Claude Code