Deferred send-reliability improvements: complexity versus operator benefit #8

Open
opened 2026-11-19 07:15:40 +01:00 by Dandaedre · 0 comments
Owner

Purpose

Keep the deferred send-reliability ideas for future reference. These are deliberately not scheduled: their complexity currently outweighs the benefit, and refresh/reconnect or explicit review is an acceptable recovery path.
Baseline: v0.10.0-rc.10, commit fa0c772. Bounded readback retries, pasted root HTTP URLs, shared-input-send button gating and clearer socket-write wording are already implemented and are not part of this backlog.

Deferred work

  • Automatic reconnect and preflight retry/backoff. Current choice: refresh/reconnect before writes rather than a new recovery state machine.
  • Automatic partial-write reconciliation, resumption, rollback and continuation of later screens. Requires reliable handling of uncertain writes, concurrent controllers and apply pulses; never blindly replay writes.
  • Persistent connections and active WebSocket response processing. Current choice: bounded sockets plus field readback; socket completion alone is not device acknowledgement.
  • Destination-scoped preflight/readback, snapshot reuse/caching and narrower stale-state comparisons. Could reduce broad-read costs, unrelated-state conflicts and Aquilon consistency-read failures, but needs adapter-specific dependency analysis and regression evidence before removing any safety checks.
  • Combining multiple screen sends under one final apply pulse. Global apply semantics and partially completed groups need investigation.
  • Send cancellation, an overall operation deadline and richer phase/per-screen progress. Requires explicit treatment of commands already sent and uncertain outcomes.
  • Hostname resolution, HTTPS and configurable Aquilon AWJ ports. Root HTTP URLs now work; the current deployments use IP literals and the normal AWJ port.
  • Verification of the entire intended geometry, including unchanged requested fields, and stronger apply-outcome evidence. Current changed-field readback cannot prove physical-output acceptance; avoid inventing acknowledgement semantics the device does not provide.

Accepted behavior and exclusions

The existing \u201cDevice state has changed\u201d dialog is acceptable, including after TAKE: \u201cSend my layout anyway\u201d rebuilds against current state and retains validation; cancel and Preview reload remain available. Do not automatically redirect to a new Preview bank without explicit confirmation.
Geometry approximation, source/output resolution conversion, an explicit \u201csend valid layers\u201d option, and expanded Aquilon fit/anchor/processed-input/source support are separate product decisions, not authorized implementation under this issue. Never silently skip invalid destinations or distort geometry.

Revisit only with evidence

Use a reproducible operator problem or measured latency/failure rate to justify individual changes. Preserve explicit writes, destination/bank safety, bounded reads, partial-outcome reporting, and no implicit TAKE, source routing or shared-input changes. Live device testing must preserve its authorized scope; emulator/readback results are not physical-output acceptance.

## Purpose Keep the deferred send-reliability ideas for future reference. These are deliberately not scheduled: their complexity currently outweighs the benefit, and refresh/reconnect or explicit review is an acceptable recovery path. Baseline: v0.10.0-rc.10, commit fa0c772. Bounded readback retries, pasted root HTTP URLs, shared-input-send button gating and clearer socket-write wording are already implemented and are not part of this backlog. ## Deferred work - Automatic reconnect and preflight retry/backoff. Current choice: refresh/reconnect before writes rather than a new recovery state machine. - Automatic partial-write reconciliation, resumption, rollback and continuation of later screens. Requires reliable handling of uncertain writes, concurrent controllers and apply pulses; never blindly replay writes. - Persistent connections and active WebSocket response processing. Current choice: bounded sockets plus field readback; socket completion alone is not device acknowledgement. - Destination-scoped preflight/readback, snapshot reuse/caching and narrower stale-state comparisons. Could reduce broad-read costs, unrelated-state conflicts and Aquilon consistency-read failures, but needs adapter-specific dependency analysis and regression evidence before removing any safety checks. - Combining multiple screen sends under one final apply pulse. Global apply semantics and partially completed groups need investigation. - Send cancellation, an overall operation deadline and richer phase/per-screen progress. Requires explicit treatment of commands already sent and uncertain outcomes. - Hostname resolution, HTTPS and configurable Aquilon AWJ ports. Root HTTP URLs now work; the current deployments use IP literals and the normal AWJ port. - Verification of the entire intended geometry, including unchanged requested fields, and stronger apply-outcome evidence. Current changed-field readback cannot prove physical-output acceptance; avoid inventing acknowledgement semantics the device does not provide. ## Accepted behavior and exclusions The existing \u201cDevice state has changed\u201d dialog is acceptable, including after TAKE: \u201cSend my layout anyway\u201d rebuilds against current state and retains validation; cancel and Preview reload remain available. Do not automatically redirect to a new Preview bank without explicit confirmation. Geometry approximation, source/output resolution conversion, an explicit \u201csend valid layers\u201d option, and expanded Aquilon fit/anchor/processed-input/source support are separate product decisions, not authorized implementation under this issue. Never silently skip invalid destinations or distort geometry. ## Revisit only with evidence Use a reproducible operator problem or measured latency/failure rate to justify individual changes. Preserve explicit writes, destination/bank safety, bounded reads, partial-outcome reporting, and no implicit TAKE, source routing or shared-input changes. Live device testing must preserve its authorized scope; emulator/readback results are not physical-output acceptance.
Dandaedre added the
enhancement
label 2026-11-19 07:15:48 +01:00
Sign in to join this conversation.
No Milestone
No project
No Assignees
1 Participants
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: idle-org/canvas-pilot#8
No description provided.