ORA

Observation

ORA-2026-0048 - Managed-feed peer mirrors need a separate timeout budget

ID
ORA-2026-0048
Date
2026-04-25
Status
surfaced
Maturity
M1
Source
docs/entries/observations/ORA-2026-0048_managed-feed-peer-mirror-timeout-budget.md

feed-protocolmanaged-feedmirror-timeout

ORA-2026-0048 - Managed-feed peer mirrors need a separate timeout budget

Type: observation Date: 2026-04-25 Source: FLT-0394 managed-feed iMac mirror warning diagnostic Observed by: CODEX-CLI-MacBook-Air-CAMBER-24

Observation

Managed-feed peer mirrors are best-effort replicas, but they were sharing the same 30 second hard timeout as primary append delegation and local readback. That budget is too tight for the current full-file mirror path once the fleet feed is several megabytes.

On 2026-04-25, repeated feed appends from the MacBook-Air primary emitted:

warning: managed-feed mirror to peer 'imac' failed for '/Users/chadbarlow/Desktop/FLEET_FEED.md' after 3 attempts; peer copy may be stale until a later sync.
warning: managed-feed mirror to peer 'imac' failed for '/Users/chadbarlow/Desktop/camber/CAMBER_FEED.md' after 3 attempts; peer copy may be stale until a later sync.

Direct diagnostics showed the peer was reachable and the remote paths were healthy:

  • ssh imac 'hostname; date -u ...' succeeded against ChadBarlowiMac.attlocal.net.
  • Remote Desktop lane entrypoints existed and resolved through iCloud symlinks.
  • Local and remote feed sizes matched: FLEET_FEED.md 6,729,939 bytes,

    • CAMBER_FEED.md 2,853,165 bytes.

  • Manual full-file SSH copy to remote /tmp took 26.85 seconds for

    • FLEET_FEED.md and 15.06 seconds for CAMBER_FEED.md.

The 26.85 second FLEET_FEED.md mirror is close enough to the 30 second cap that ordinary network variance can cause false mirror failures even while SSH, the peer alias, and the remote filesystem are correct.

Change Landed

scripts/fleet-runtime-helpers/feed-append now has a mirror-specific timeout:

SSH_MIRROR_HARD_TIMEOUT_SECONDS="${FEED_APPEND_MIRROR_HARD_TIMEOUT_SECONDS:-90}"

The default primary/delegation/readback timeout remains 30 seconds. Only the best-effort peer mirror copy path uses the longer 90 second budget. The warning now includes the per-attempt timeout value so future diagnostics can tell which budget fired.

Follow-up Shape

The larger structural problem is that each managed-feed append rewrites the entire feed file to each peer. At current sizes that costs roughly 42 seconds for a Fleet+Camber actionable append even when it succeeds. A separate ticket should evaluate a delta or async mirror path without changing authoritative queue semantics.