Rebuild the client endpoint after repeated reconnect failures - #90
Merged
Merged
Conversation
A wedged iroh endpoint can be broken beyond what the network_change() nudge repairs: a relay link lost to a ping timeout that never re-establishes, stale cached paths for the server. Reconnect attempts then time out forever while a process restart connects instantly (seen on WSL2 with custom relays and a LAN server: heartbeat detected the drop in 33s, but every reconnect timed out at 30s until a manual restart). Introduce ClientEndpoint, a shared handle wrapping the live endpoint plus a rebuild recipe. Every third consecutive failed reconnect attempt the loop swaps in a freshly bound endpoint (new sockets, relay connections, discovery) and closes the old one in the background — the in-process equivalent of a restart, keeping the proxy listeners bound and held requests alive. Rebuilds skip the startup all-relays probe (a partial relay outage must not block recovery) and tolerate the online wait failing (mDNS can still reach a LAN server). Quick clients keep their fixed identity; keypair clients get a fresh ephemeral node id (the CLI status panel now tracks it). CLI, desktop, and FFI embedders all adopt the handle. Bridges deliberately do not rebuild: they dial on the server's own endpoint, and rebuilding it would sever every connected client. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UA6a2mvmPMsnymfox6nK7K
Review fixes on the endpoint-rebuild change: in ClientEndpoint creation a wait_online failure returned via `?` and dropped the just-bound endpoint without close(), which is fatal under the release profile's panic=abort. Also correct the architecture doc: network_change() applies only to retries that don't trigger a rebuild, not to every retry. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UA6a2mvmPMsnymfox6nK7K
Same pre-existing hazard as the client fix in the previous commit: create_server_endpoint propagated a wait_online failure via `?` and dropped the bound endpoint unclosed, fatal under panic=abort. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UA6a2mvmPMsnymfox6nK7K
This was referenced Sep 1, 2026
Merged
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
1 participant
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.
Summary
ClientEndpoint::rebuild) — fresh sockets, fresh relay connections, fresh discovery — the in-process equivalent of the manual restart that recovered a wedged client (network_change()alone couldn't repair a relay link lost to a ping timeout). Other retries keep thenetwork_change()nudge.ClientEndpointhandle (shared by CLI, desktop, and FFI) wraps the endpoint with a rebuild factory; the rebuild skips the startup per-relay probe and tolerates the online wait failing so a partial outage never blocks recovery. The old endpoint is closed gracefully on a background task (never dropped unclosed — fatal under panic=abort).Testing
reconnect_rebuilds_a_dead_endpoint(kills the client's endpoint, asserts reconnection lands on a freshly built one).ci/all.shgreen on all three host platforms (mac, linux, windows) for the feature commit; Linux clippy-D warnings+ 194 core/CLI tests green on the follow-up fixes.🤖 Generated with Claude Code
https://claude.ai/code/session_01UA6a2mvmPMsnymfox6nK7K