A subtle encoding/json behavior, diagnosed in public by the reporter, and fixed with a fresh struct per payload.
Short answer
When two Newt targets shared a backend and a websocket reconnect re-sent the configuration, one target could inherit the other's expected status code and be reported unhealthy forever. Newt decoded each registration payload into its existing state, and Go's encoding/json can reuse existing slice elements while decoding an array. The fix decodes into a fresh WgData and swaps it in only on success. It was merged as PR #462 and closed issue #411.
The problem
Newt is the tunneling client for the Pangolin project. Issue #411 described two targets on the same host:port, one expecting 200 and the other 401. After a reconnect, the first target logged unexpected status code: 200 (expected: 401) on every check. A restart cleared it until the next reconnect.
Root cause
The reporter traced it in the issue thread. handleConnect decoded the payload into the persistent n.wgData:
json.Unmarshal(jsonData, &n.wgData) The relevant behavior is reuse of existing slice elements during decoding: a JSON null does not reset a non-pointer int. If targets arrived in a different order, or with a null status, old values survived into the wrong entries. The reporter linked a community pull request, #413, proposing the same approach. The supplied research records that PR as still open on Nov 3, 2026.
The fix
func (n *Newt) applyWgData(jsonData []byte) error { var wgData WgData if err := json.Unmarshal(jsonData, &wgData); err != nil { return err } n.wgData = wgData return nil } Three tests cover a reordered health-check payload with a null status, stale fields being cleared, and existing data being kept when the payload is malformed.
By the numbers
- Repository: fosrl/newt, 921 stars (Nov 3, 2026), Go, AGPL-3.0
- Files changed: 2 (1 source file, 1 test file)
- Lines: +77 / -1
- Commits: 1
- Opened: Oct 6, 2026. Merged: Oct 12, 2026. Time to merge: 6 days, 2 hours
- Review: approved and merged by a maintainer
How the review went
A maintainer approved, merged and replied "Thanks!"
The Nov 3, 2026 review notes record that the required CLA statement and AI Disclosure section were absent from this PR. Its approval and merge do not resolve those omissions.
FAQ
Why not reset fields individually?
Decoding into a fresh struct removes the whole class of slice-reuse and null-handling problems at once, and the error path leaves the previous state untouched.
Where can I read more?
See the pull request, the original issue, and the earlier community PR #413.
The goal
The hard part here was the diagnosis, and it happened in public, in the issue. The merged change is small because the cause was understood.
This fix was developed with Prilog. Read the original case study on the Prilog blog.

