anuna-research/elephant
1
0
Fork 0
Code Issues 6 Pull requests Projects Packages Wiki Activity Actions

continuous sync serial dial pass lets offline peers delay every theory #18

New issue
Closed
opened 2026-07-16 04:27:48 +00:00 by anuna-02 · 0 comments
anuna-02 commented 2026-07-16 04:27:48 +00:00 (Migrated from codeberg.org)
Copy link

Summary

Continuous sync awaits every theory/peer dial serially. An unreachable peer in one old theory therefore delays sync attempts for unrelated active theories. ELEPHANT_SYNC_INTERVAL is effectively "sleep after the entire serial pass", not the maximum retry interval for each peer.

Reproduction from the two-machine run

The daemon was configured with ELEPHANT_SYNC_INTERVAL=30. Relevant log times were:

03:53:00.775  constitution peer timed out
03:53:00.986  rc-live peer: connection lost
03:53:31.313  rootclaim peer timed out
03:53:34.929  another peer had no addressing information

03:54:36.193  constitution peer timed out
03:54:36.251  rc-live peer: connection lost

Subsequent rc-live attempts occurred at approximately 94-96 second intervals, despite the configured 30-second interval. The same pattern continued for the duration of the run.

Cause

src/p2p/sync.rs::dial_once loops over every local theory and roster peer and awaits dial_sync, open_bi, and the whole sync session inline. run_with_endpoint then sleeps for interval before invoking the next full pass. One or more transport timeouts are added directly to every peer's retry cadence.

Impact

  • A stale/offline member in an unrelated theory degrades convergence for active theories.
  • Delay grows with the number of unreachable roster peers.
  • The reconnect/discovery latency budget cannot be enforced per peer.
  • Repeated full-pass retries create avoidable DHT/relay traffic and noisy logs.

Expected / acceptance criteria

  • Schedule each (theory, peer) independently, or execute a snapshot of due peers with bounded concurrency.
  • Give each peer independent timeout, jittered backoff, last-attempt state, and cancellation on daemon shutdown.
  • An offline peer in theory A must not delay the first attempt or retry cadence for a reachable peer in theory B.
  • Add a deterministic scheduler regression with one never-completing dial and one immediately-ready dial.
  • Feed per-peer outcomes into the health reporting tracked by #15.
  • Keep concurrency bounded so a large roster cannot create unbounded tasks or sockets.

Environment

Elephant 0.1.2, macOS, two-machine FLF run. The serial structure is also present on current main (130a910).

### Summary Continuous sync awaits every theory/peer dial serially. An unreachable peer in one old theory therefore delays sync attempts for unrelated active theories. `ELEPHANT_SYNC_INTERVAL` is effectively "sleep after the entire serial pass", not the maximum retry interval for each peer. ### Reproduction from the two-machine run The daemon was configured with `ELEPHANT_SYNC_INTERVAL=30`. Relevant log times were: ```text 03:53:00.775 constitution peer timed out 03:53:00.986 rc-live peer: connection lost 03:53:31.313 rootclaim peer timed out 03:53:34.929 another peer had no addressing information 03:54:36.193 constitution peer timed out 03:54:36.251 rc-live peer: connection lost ``` Subsequent `rc-live` attempts occurred at approximately 94-96 second intervals, despite the configured 30-second interval. The same pattern continued for the duration of the run. ### Cause `src/p2p/sync.rs::dial_once` loops over every local theory and roster peer and awaits `dial_sync`, `open_bi`, and the whole sync session inline. `run_with_endpoint` then sleeps for `interval` before invoking the next full pass. One or more transport timeouts are added directly to every peer's retry cadence. ### Impact - A stale/offline member in an unrelated theory degrades convergence for active theories. - Delay grows with the number of unreachable roster peers. - The reconnect/discovery latency budget cannot be enforced per peer. - Repeated full-pass retries create avoidable DHT/relay traffic and noisy logs. ### Expected / acceptance criteria - Schedule each `(theory, peer)` independently, or execute a snapshot of due peers with bounded concurrency. - Give each peer independent timeout, jittered backoff, last-attempt state, and cancellation on daemon shutdown. - An offline peer in theory A must not delay the first attempt or retry cadence for a reachable peer in theory B. - Add a deterministic scheduler regression with one never-completing dial and one immediately-ready dial. - Feed per-peer outcomes into the health reporting tracked by #15. - Keep concurrency bounded so a large roster cannot create unbounded tasks or sockets. ### Environment Elephant 0.1.2, macOS, two-machine FLF run. The serial structure is also present on current `main` (`130a910`).
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
spec-005-vocab-reviewer-round
v0.1.6
v0.1.5
v0.1.4
v0.1.2
v0.1.1
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
anuna-research/elephant#18
No description provided.