Skip to content

Edge gateway starts without the forwarding sysctls its FIB lookup needs #586

Closed

Description

@privateip

Summary

The edge gateway needs IPv6 forwarding enabled on every interface it attaches to. Without it, the kernel FIB lookup the datapath relies on refuses every packet, and the datapath drops everything it is handed.

The gateway tries to set those sysctls at startup. The pod's /proc/sys is read-only, so the writes fail. The failure is logged as a warning, and the process carries on and reports healthy. This happens on every node running the datapath, including ones whose uplinks are otherwise fine.

The startup code looks like it treats the failure as fatal, but the helper it calls logs each failure and always returns success, so the error branch never runs. A comment calls the step best-effort, which matches what happens. The unreachable branch is what misleads.

Two things are wrong on their own: the DaemonSet does not get a writable /proc/sys, and a datapath that cannot forward starts anyway.

First reported in #583, which is about a separate defect in the same startup path.

Acceptance criteria

  • The gateway pod can write the forwarding sysctls it needs
  • A gateway that cannot enable forwarding fails visibly instead of running a datapath that drops every packet
  • The unreachable error branch and the comment disagreeing with it are reconciled

Phase 0 of #698.

Activity

  1. self-assigned this
    on Sep 21, 2026
  2. added
    needs triageIssues that need to be triaged before implementing
    and removed
    needs triageIssues that need to be triaged before implementing
    on Oct 1, 2026
  3. privateip commented on Oct 4, 2026

    @privateip
    CollaboratorAuthor

    Added as a sub-issue of datum-cloud/enhancements#884. The network load balancer runs on galactic-gateway's Maglev/DSR datapath, so a gateway that starts without forwarding drops all of a load balancer's traffic. Still reproduces on main: ConfigureFIBLookupUplinkSysctls still returns nil unconditionally, so the error branches in cmd/galactic-gateway/gateway.go can never be reached.

  4. added theissue type on Oct 4, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Type

Fields

Priority

None yet

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions