Supervisor filtered only on enabled, so disabled-mode segments (e.g.
the OPS_SEED_TEMPLATES skeleton, which seeds mode='disabled' until an
operator finishes wiring) got a task spawned every 10 s only to exit
immediately on the task's own enable-check, spamming the log with
"segment X disabled or removed, task exiting" every supervisor tick.
Aligning the supervisor's filter with the task's exit condition lets
disabled-mode segments stay quiescent. Flipping mode away from
'disabled' via the config UI now reaches the engine within one
supervisor cycle (≤10 s), unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Extends ensure_default_templates to also seed the 5 公共段 (前端码车 /
前端放车 / 前端摆渡 / 窑尾摆渡 / 卸砖 / 回车) and their shared resource
declarations (transfer_front / transfer_tail / unload_position /
return_line / robot_arm) so operators have the full skeleton to bind
equipment + signals against.
Engine now runs a station signal-conflict check during the run-halt
phase: any station whose presence and vacancy are both true with Good
quality emits ops.alarm.signal_conflict + segment.fault_locked and
transitions the segment to Faulted. Closes the final P8 alarm type
from design doc §8.1.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EventInsert + EventRecord + record_event now carry the subject_type /
subject_id columns added by the P1 migration. Ops events populate
"segment" / "station" subjects so the timeline can be filtered without
parsing event_type strings. Platform SourceCreated / Updated / Deleted
attribute themselves to subject_type="source". Adds get_events_*_filtered
in core and exposes GET /api/event on ops with event_type /
event_type_prefix / subject_type / subject_id query params, closing
design doc §14 "event 表能按 ops.* 和 subject_type/subject_id 查到全链路事件".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds ResourceRegistry::sweep_stale and runs it on each supervisor tick
so a panicked or stuck segment task can't keep a shared resource
locked indefinitely. The per-segment task refreshes heartbeat on every
iteration for each key in runtime.held_resources, distinguishing live
owners from dead ones.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ensure_kiln_templates idempotently inserts 6 dry-kiln stations and 6
segments (infeed/step/outfeed × 2 kilns) with their canonical step
sequences from §10.1. Equipment and station-signal bindings stay
operator-owned through the CRUD APIs. Startup runs the seed only when
OPS_SEED_TEMPLATES=true|1, so production deployments don't accidentally
mutate config.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
step_executor gains three dispatch modes: pulse (default), hold
(hold_until_confirm), and value (transfer_move_to writes the target
station's code). The engine now sends step.stop_command_role whenever
cancel_on_fault is true on Faulted entry, and threads a target-station
lookup ahead of dispatch. A new simulate module patches the resolved
confirm signal after a short delay when SIMULATE_PLC is set, so
segments can be driven end-to-end without a real PLC.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Land design doc §12 P1 + P2: ops business tables plus station/segment
configuration endpoints. The engine (P3) consumes these as its inputs.
- Migration: six ops tables (station, station_signal, process_segment,
segment_step, segment_interlock, segment_resource) plus event attribution
columns (subject_type, subject_id).
- model.rs: FromRow structs and string-backed enum helpers
(StationType, StationSignalRole, SegmentMode, ActionKind, OnTimeout,
InterlockAppliesTo, RuleKind).
- service: station CRUD with signal-binding upsert; segment CRUD with
nested step/interlock CRUD and transactional resource replacement.
- handler: 13 endpoints covering design doc §9.1 config routes with
validator-based input checks and enum allowlists.
- router: wires the new routes; smoke tests cover station and segment
collection routes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Wires AppState with EventManager, SegmentRuntimeStore, ResourceRegistry
and an engine supervisor that idles until P1 lands the segment schema.
The run() bootstrap connects enabled sources, installs a Ctrl+C handler,
and disconnects on shutdown, matching the feeder app lifecycle. The
router exposes /ws/public, /ws/client/{id}, simple_logger middleware
and a permissive CORS layer.
AppEvent covers the full ops.* taxonomy from the spec; resource lease
tracking includes heartbeat timestamps for the §7 recovery strategy and
has two unit tests for acquire/release semantics.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Lifts ControlUnit, UnitRuntime/State, ControlRuntimeStore and unit CRUD
into app_feeder_distributor; their feeder-specific semantics
(DistributorRunning state, run_time/stop_time/acc_time/bl_time pacing)
violated the shared-core invariant.
WsMessage drops the UnitRuntimeChanged variant in favor of a generic
AppEvent(AppWsEvent) carrying {app, event_type, data}. Feeder emits
"feeder"/"unit_runtime_changed"; the web client dispatches by app
namespace, leaving core free of business types. The equipment handler
keeps its friendly unit-exists check by issuing an inline EXISTS query
instead of pulling the unit service.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bring back crates/app_feeder_distributor/src/control/simulate.rs which
was dropped when feeder's local connection/telemetry/websocket modules
moved into core; the restored module reuses core's equivalents via
plc_platform_core::{connection, service, telemetry, websocket}.
Also restore the call sites that were lost alongside it:
- app.rs starts the chaos task when SIMULATE_PLC=true
- engine.rs fires simulate_run_feedback after each pulse command so
the auto-control state machine sees the RUN bit transition it
would get from a real PLC
- handler/control.rs does the same after manual start/stop commands
The SIMULATE_PLC flag is now read via simulate::enabled() from the
environment rather than state.config.simulate_plc (the old config
struct was removed with the module migration). To expose equipment
ids by kind (used for run feedback), build_equipment_maps now also
returns HashMap<kind, Uuid>.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The split existed only so handle_control_event could log
UnitStateChanged and then delegate; that's no longer worth its own
function. Co-locating the UnitStateChanged special case with the other
match arms makes its 'log-only, no persist' treatment self-evident,
and the call chain drops from handle → persist → record to handle →
record.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
After record_event became the primitive for INSERT + broadcast + tracing,
persist_and_broadcast no longer persists or broadcasts directly — it
translates a PlatformEvent into an EventInsert and delegates. The new
name makes the layering explicit (record_event is the primitive,
record_platform_event is the PlatformEvent translator).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add MetadataCache to PlatformContext — a lazy-loaded, cross-app cache
of code fields used when formatting event messages. Each persisted
AppEvent previously did 1-2 extra SELECTs to look up the code for its
human-readable message; after this change the same id hits the cache
on all subsequent events.
Invalidation: the platform-owned equipment handler invalidates its
entry on update/delete; feeder's unit handler does the same for
units. Deletes are invalidated for hygiene only — no further events
should target a deleted id.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
record_event now logs at the matching tracing level (error/warn/info)
using the persisted message — giving every app uniform event logs for
free. Feeder's handle_control_event collapses from a 60-line match
(which just duplicated the persisted message with less-readable UUIDs)
to a single if-let for UnitStateChanged, which is the only AppEvent
that is intentionally not persisted.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Move the INSERT + WebSocket broadcast mechanism out of the feeder app
and into plc_platform_core as pub record_event(pool, ws, EventInsert).
The event table schema is owned by core, so writing to it is a platform
capability — apps (feeder, future ops) should only decide what to emit,
not how to persist it. Also replaces the 7-tuple in core's
persist_and_broadcast with the named EventInsert struct for readability.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- engine.rs: use is_none_or instead of map_or, remove redundant closures
- service/control.rs: move get_equipment_role_points before test module
- event.rs: replace complex 7-element tuple with PersistableEvent struct
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace verbose fully-qualified paths with top-level use imports
for plc_platform_core types and services throughout control.rs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The feeder router uses platform_routes() directly from core,
making these re-export modules unused. The only actual dependency
(SignalRolePoint) now references plc_platform_core directly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move TelemetryProcessor (PointNewValue batching/dedup/broadcast) from feeder to
plc_platform_core. PlatformBuilder.build() now auto-wires telemetry processing and
reconnect task. PlatformContext.emit_event() handles connection management side effects
(connect/reconnect/disconnect/subscribe/unsubscribe) directly. Simplify PlatformEventSink
trait from 6 methods to single on_event(). Feeder's AppEvent now only contains business
events; FeederPlatformEventSink only handles UnitsChanged for control runtime.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Move source/point/equipment/tag/page handlers from feeder to plc_platform_core
using State<PlatformContext>; feeder re-exports via handler modules
- Keep batch_set_point_value in feeder (requires app-specific write key auth)
- Add PlatformEvent enum and persist_and_broadcast() in core for platform event
persistence to DB + WebSocket broadcast
- Add PlatformContext::emit_event() that handles both sink notification and
async persistence in one call
- Add platform_routes<S>() in core for centralized route registration;
both feeder and ops merge it instead of duplicating route definitions
- Implement FromRef<AppState> for PlatformContext in both apps
- Add FeederPlatformEventSink adapter bridging core events to feeder's
EventManager + ControlRuntimeStore
- Add event namespace prefixes: platform.source.created, feeder.unit.fault_locked, etc.
- Register full platform CRUD routes in ops app
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- PlatformContext now holds pool, connection_manager, ws_manager
- bootstrap_platform returns PlatformBuilder for pre-Arc setup
- Feeder AppState embeds PlatformContext (state.platform.pool etc.)
- Ops AppState embeds PlatformContext with real DB connection
- Remove WebSocket type duplication: feeder re-exports from core
- Add subscribe_room/send_to_room/remove_room_if_empty to WebSocketManager
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
caoqianming