secs-gem/docs/DAEMON_ROADMAP.md

# Vendor Daemon & gRPC API — Status, Known Issues, and Plan to Fab-Readiness

> **This is a forward-looking roadmap, not a description of shipped behaviour.**
> Every item carries a status marker. Do not read an item as "done" unless it
> says ✅. (Last full audit: 2026-06-10.)
>
> Status legend: ✅ done · 🚧 in progress · ⬜ planned · ⚠️ risk/unknown

## What this is

A vendor-facing **daemon** (`secs_gemd`) that runs the SECS/GEM engine as its
own process and exposes a small, name-based, language-agnostic API over gRPC,
so a tool's control software (in any language) can drive the equipment without
linking C++ or knowing SEMI. See `proto/secsgem/v1/equipment.proto`.

The point of the daemon model: it owns the durable HSMS relationship with the
host and stays conformant while the tool software restarts/upgrades/crashes.

## Current status (2026-06-10, end of day)

| Piece | Status | Notes |
|---|---|---|
| `proto/secsgem/v1/equipment.proto` | ✅ | v1 surface designed: universal + carrier/recipe/job tiers, `Subscribe` stream, health |
| `HostCommandRegistry::set_handler` behaviour hook | ✅ | the engine seam for command behaviour; tested |
| `EquipmentRuntime` (engine owner) | ✅ | tested (`test_runtime.cpp`); `secs_server` runs entirely on it (live GEM300 demo passes) |
| `register_default_handlers` (the 56 GEM handlers as a library fn) | ✅ | `src/gem/default_handlers.cpp`; tested (`test_default_handlers.cpp`) |
| gRPC/protobuf toolchain (Dockerfile + CMake codegen) | ✅ | grpc++ 1.51 / protoc 3.21; opt-in `SECSGEM_DAEMON`, graceful skip without grpc |
| `secs_gemd`: `SetVariables` / `FireEvent` / `GetControlState` / `GetVariables` | ✅ | **format-aware** both directions (declared SECS-II formats on write, `from_item` on read) and thread-safe (snapshot maps + posted writes + `read_sync` reads). In-process gRPC tests incl. run_async production mode (`test_daemon_service.cpp`, 61 assertions) |
| Daemon interop vs **secsgem-py** reference host | ✅ | `interop/daemon_interop.py` (via `gemd` compose service): gRPC `SetVariables(ChamberPressure=2.5)` + `FireEvent` → host receives `S6F11 CEID 300` carrying `<F4 2.5>` — value *and declared format* flow gRPC→engine→HSMS→host |
| Daemon interop vs **secs4j** (Java) | ✅ | `TARGET=gemd interop/secs4j_validate.sh` — 55/55 against the daemon's HSMS face; in CI |
| `Subscribe` host→tool command stream + `CompleteCommand` | ✅ | HCACK-4 contract implemented + tested in-process AND live vs secsgem-py (full loop: S2F41 → stream → complete → S6F11) |
| Universal RPC surface complete (vars/events/alarms/control-state/health) | ✅ | Phase A done; daemon tests 101 assertions, interop 15 checks |
| Python client package (the "beautiful API") | ✅ | `clients/python` (`secsgem-client`); 13-check interop green via the published API |

## Known issues (found in the 2026-06-10 audit; honest list)

- ✅ ~~**`GetControlState` cross-thread read.**~~ Fixed 2026-06-10: the runtime
  keeps an atomic control-state mirror updated via an `add_state_change_handler`
  observer (`HandlerSlot` primary+observers pattern), so the mirror survives
  `register_default_handlers` claiming the primary slot. `control_state()` is
  now safe from any thread.
- ✅ ~~**Alarms have no name key.**~~ Optional `name:` added to the alarm
  config (loader + validator + shipped equipment.yaml); daemon RPCs accept
  the name or the stringified ALID.
- ⬜ **`pvd_tool` predates the behaviour hook AND the runtime.** It still
  hard-codes START behaviour in a router handler and hand-wires its own
  main(). Migrate it to EquipmentRuntime + per-capability registration +
  `commands.set_handler` so the flagship example showcases the intended
  integration shape. (Phase C item 9.)
- ✅ ~~**Interop harnesses are manual.**~~ `tools/run_interop.sh` runs all nine
  validation steps with one command (verified green); CI lanes added, pending
  first-push verification (Phase 0 item 2).
- ✅ ~~**TSan lane doesn't cover the daemon.**~~ Covered locally + in CI
  with `tools/tsan.supp` (third-party-only suppressions). Caught + fixed a
  real test-side contract violation on its first run.
- ⚠️ **macOS bind-mount staleness can break Docker builds mid-edit** (a build
  reading a half-synced source file). Not a product bug; re-run the build.

## The `Subscribe` design (settled — implement to this)

`S2F42` is an *acknowledgement*, not a completion: SEMI separates "I accept
your command" from "the work finished". The conformant, non-blocking flow:

1. Host sends `S2F41 START`. The engine's `on_command` handler (registered by
   the daemon) runs on the io thread.
2. If no tool client is subscribed → fall back to the YAML declarative ack.
   If a tool is subscribed → push the command onto its `Subscribe` stream and
   **return `HCACK=4` (AcceptedWillFinishLater) immediately** — never block
   the io thread or the T3 window on the tool.
3. The tool does the work and reports the outcome via `FireEvent` (success
   event) / `SetAlarm` (failure) — exactly how secsgem-py applications and
   commercial gateways do it.
4. `CompleteCommand` therefore only correlates/audits the command lifecycle in
   v1. A *synchronous gating* mode (tool decides HCACK 0/2 before the S2F42
   goes out) requires a deferred-reply mechanism in the engine — explicitly a
   v2 refinement, not needed for conformance.

Sub-decisions (settled 2026-06-10, implemented + tested):
- v1 is a firehose: every subscriber receives every host request.
- NO buffering: with no subscriber a command takes its declarative YAML ack
  and is not replayed on reconnect — never "will finish later" for work no
  tool will do. Documented in the proto's Subscribe contract.

## Plan — ordered next steps

### Phase 0 — structural debts (from the 2026-06-10 design review; pay before sprinting)
The review's verdict: architecture and API bets are sound, but two structural
debts tax every later phase, and the most valuable tests aren't automated.

1. 🚧 **Multi-observer callbacks** (THE structural blocker — hit twice already).
   `HandlerSlot` (primary slot keeps legacy set_ semantics; append-only add_
   observers survive it) — done for `ControlStateMachine` + PJ/CJ stores, plus
   runtime atomic control-state mirror (race retired) and `add_link_observer`
   (WatchHealth foundation). ⬜ Remaining: roll the same 3-line pattern onto the
   other single-slot classes (comm-state, EPT, exceptions, substrates, modules,
   carriers, E84) as each phase needs them — mechanical now that the type exists.
2. 🚧 **CI the interop + conformance harnesses.** `tools/run_interop.sh` ✅ —
   one command runs ALL nine validation steps (build, unit, daemon-unit,
   py-host 31 checks, conformance 47, daemon bridge, spool restart, tshark,
   secs4j 55) with a PASS/FAIL summary; verified green end-to-end 2026-06-10.
   CI ✅ added but UNVERIFIED until pushed: grpc deps + `secs_gemd_tests` in
   the build job (fails loudly if the daemon silently drops out), and a new
   `python-interop` lane (py-host + conformance + daemon harness against
   localhost, no docker-in-docker). ⬜ Verify the lanes on the first push.
3. ✅ **Fix `CompleteCommand` proto comment** — it described the rejected
   blocking model; now states the HCACK-4 contract.
4. ✅ **Table-driven handler conformance test** — one ordered scenario
   drives 53 of the 56 handlers through `router.dispatch` (236 assertions).
   Golden frames: S1F13, S5F1, and a composed S6F11, all hand-computed from
   E5 rules (external pins, not codec-derived).
5. ✅ **Decomposed `register_default_handlers` into 15 per-capability
   functions** (identification, ECs, clock, event reports, remote commands,
   trace/limits, spooling, alarms, exceptions, material tracking, carriers,
   recipes, object services, jobs, terminal) — vendors register only what
   their equipment is; `register_default_handlers` = all 15. Magic constants
   replaced by YAML **role bindings** (`roles:` block — control_state_svid,
   clock_svid, cj_executing_ceid, cj_completed_ceid) parsed into the
   descriptor with historical defaults, validated (CEID roles must be
   declared). Tested: subset registration, role-driven SVID refresh, roles
   loader (present/custom/absent); full battery green (473/3087 core incl.
   the 53-handler sweep, live GEM300 demo, 20-check daemon interop).
6. ✅ **Standardize the mutable-read pattern** — `EquipmentRuntime::read_sync`
   (post-to-io + future with deadline; nullopt => UNAVAILABLE at the RPC edge).
   Precedent set by `GetVariables`; every future mutable read copies it.
7. ✅ `equipment_service.hpp` moved to `include/secsgem/daemon/` (apps/
   include-path hack removed). TSan daemon lane added locally + in CI
   (`tools/tsan.supp` suppresses UNinstrumented system libgrpc/libabsl
   internals only — our frames stay checked). The lane caught a real
   contract violation on its first run (a test reading the model from the
   test thread under run_async — fixed to read_sync); now TSan-clean with
   halt_on_error=1.
8. ✅ Identifier-safe name validation: `ConfigValidator` warns (not errors)
   on non-identifier variable/event/alarm/command names — bindings expose
   names as kwargs/attributes. Format-compliance property test ✅; unset-
   `Value` guard ✅.

### Phase A — finish the universal daemon surface (small, unblock vendors)
1. ✅ `GetVariables` — `from_item` reverse conversion (scalar for 1-element
   arrays, List otherwise; C2-as-text and U8>2^63 noted as TODOs) + reads via
   `read_sync`. Tested under **run_async (production threading)** — write
   through the API, read back through the API — plus empty-query-returns-all,
   INVALID_ARGUMENT on unknown names, and a live round-trip check in
   `daemon_interop.py`.
2. ✅ Alarm `name:` config field (optional local key; `name` appended LAST on
   the Alarm struct so existing brace-inits compile unchanged) + `SetAlarm`/
   `ClearAlarm` RPCs (addressable by config name AND stringified ALID).
   Validated end-to-end: gRPC `SetAlarm(chiller_temp_high)` -> secsgem-py host
   receives `S5F1 ALCD=0x84 ALID=1`.
3. ✅ `RequestControlState` — fires operator events on the io thread and
   reports what the E30 table actually did (ACCEPT iff landed in the requested
   state; the shipped table has NO operator path to EquipmentOffline and the
   test pins that honesty). ✅ `WatchHealth` — initial snapshot + push on
   link/control-state change (+ spool depth sampled at 500ms); unit-tested
   incl. the change push; link state still SELECTED/DISCONNECTED only
   (CONNECTED reserved, TODO in code). Interop covers RequestControlState;
   WatchHealth external check rides with Phase B.
4. ✅ Done per-item above (daemon suite at 101 assertions; interop at 15 checks).

### Phase B — the command stream (the big one)
5. ✅ `Subscribe`/`CompleteCommand` implemented per the HCACK-4 design.
   Reconnect decision settled and documented in the proto: **no buffering** —
   a command with no subscriber takes its declarative YAML ack (the honest
   pre-daemon behaviour) and is not replayed. Firehose fan-out; per-command
   forwarding handlers registered from the registry (new `names()`/`spec()`
   accessors); pending-id audit map. In-process tests drive a REAL S2F41
   through the default-handler router on the io thread: HCACK 4 with a
   subscriber (params arrive on the stream), declarative Accept without,
   CompleteCommand known/unknown ids, fallback restored after unsubscribe.
6. ✅ The full conformant loop runs against secsgem-py live: host `S2F41
   START` → `S2F42 HCACK=4` → tool receives Command(name=START, id) on the
   stream → `CompleteCommand` → tool fires the event → host receives `S6F11`.
   (interop now 20 checks.)
7. ✅ Java interop: `TARGET=gemd interop/secs4j_validate.sh` runs the full
   55-check secs4java8 suite against the DAEMON's HSMS face — 55/55 green
   (secs_gemd and secs_server sit on the same register_default_handlers, so
   byte-identical GEM is now proven, not assumed). CI step added. The
   command loop with a live subscriber is covered by the python harnesses
   (a Java *tool-side* gRPC client remains possible future work).

### Phase C — the beautiful Python client
8. ✅ `clients/python/` — pip-installable `secsgem-client`, pure Python,
   stubs pre-generated (relative-import fixed). The full agreed API:
   `eq.set(ChamberPressure=2.5)` kwargs + `eq["..."]` item syntax, `eq.get`,
   `eq.fire(event, **data)`, `eq.alarm`/`eq.clear`, `eq.control_state`,
   `eq.request_control_state`, `eq.health()`/`watch_health()`, and
   `@eq.on("START")` + `eq.listen(background=...)` with auto-CompleteCommand.
   Errors raise `SecsGemError` carrying the daemon's explanation.
   PROOF: `interop/pyclient_interop.py` drives the PUBLISHED package against
   a live daemon with secsgem-py as the host — 13 checks all green (S6F11/
   S5F1 set+clear on the wire, HCACK-4 command loop through the decorator,
   operator offline). Conversion layer unit-tested (bool-before-int etc).
   Wired into tools/run_interop.sh as the `pyclient` step.
9. ✅ `clients/python/examples/mini_tool.py` (~25 lines) and the C++
   `pvd_tool` migrated to EquipmentRuntime + register_default_handlers +
   `set_handler` (1093 -> 570 lines; now serves all 56 handlers; boots
   verified). Chapter 42 teaches the daemon path.
10. ✅ **C++ client** (`clients/cpp`): header-only twin of the Python
   client — `eq.set("ChamberPressure", 2.5)`, `eq.on("START", fn)` +
   `listen_async()`, alarms/health/control-state, SecsGemError. Tested
   end-to-end over loopback TCP against the real service inside
   `secs_gemd_tests` (141 assertions total), incl. the HCACK-4 loop.
   `cpp_mini_tool` is the worked example.

### Phase D — GEM300 in-the-loop (process/carrier tools)
10. 🚧 Semantics SETTLED: v1 is **observe-and-report** — the engine keeps
    acking S16/S3/S7/S2F15 from its FSM tables (what both references
    validated); the tool observes lifecycle events on the stream and reports
    physical progress back. Gating (tool decides the ack) = the documented
    v2 deferred-reply item. ✅ `ProcessJob` on the stream (PJ store observer:
    →Processing/Start=START, /Resume=RESUME, →Paused=PAUSE, →Stopping=STOP,
    →Aborting=ABORT; carries recipe + material bindings) +
    `ReportProcessJob` (SETTING_UP→SetupComplete, COMPLETE→ProcessComplete,
    ABORTED→AbortComplete; PROCESSING informational; INVALID_OBJECT /
    CANNOT_DO_NOW on unknown job / illegal transition). ✅ Carriers (E87):
    `CarrierStore` gained the HandlerSlot observer pattern; the daemon's
    id-observer forwards host S3F17 decisions as `CarrierAction` (PROCEED on
    Confirmed, CANCEL on CancelCarrier) and `ReportCarrier` drives the
    arrival/access flow (WAITING creates + slot map; IN_ACCESS/COMPLETE the
    access FSM). KNOWN EDGE: a host CancelCarrier on a still-NotConfirmed
    carrier is an FSM self-transition, so no observer fires and the tool
    isn't notified — fix needs an event-level (not state-change) hook;
    low priority (hosts cancel after a read, i.e. from Confirmed).
11. ✅ `ProcessProgram` on the stream when S7F3 lands (new RecipeStore
    added-observer) and `ConstantChange` when S2F15 is ACCEPTED (new
    EquipmentConstantStore changed-observer; rejected writes never fire).
    Python client: `on_process_job`/`on_recipe`/`on_constant_change` +
    `report_job`.
12. 🚧 In-process E2E covers the full job loop (S16F11→S16F5→stream→report→
    FSM, 175-assertion daemon suite); secs4j's 55 checks run against the
    daemon (declarative path). ⬜ A live host-driven job loop with a
    subscribed tool needs raw S16 frames host-side (secsgem-py 0.3.0 lacks
    S16 builders — see interop/raw_gem300_harness.py for the frame source).

### Phase E — hardening & operations
13. ✅ gRPC exposure: default flipped to `127.0.0.1`; Unix-domain-socket
    support verified (`--grpc unix:///...`); SECURITY.md documents the
    contract (unauthenticated API = localhost/UDS only; stunnel if remote).
    TLS creds remain optional future work (UDS removes the need same-host).
14. ✅ `tools/run_interop.sh` now 11 steps (added pyclient + daemon-ops);
    CI python-interop lane gained pyclient, spool-restart, and daemon-ops
    steps — every harness now runs in CI.
15. ✅ Graceful shutdown (SIGTERM/SIGINT -> gRPC drain with 2s stream-cancel
    deadline -> engine stop -> exit 0; journal-safe; the old in-code TODO is
    gone), Prometheus gauges (`secsgem_link_selected` / `_control_state` /
    `_spool_depth` via the Phase-0 observers + io-thread sampling),
    `--spool-dir` on the daemon, and `deploy/secs_gemd.service` (hardened:
    DynamicUser/ProtectSystem/StateDirectory/TimeoutStopSec). All enforced
    by `tools/check_daemon_ops.sh`. Single-session (HSMS-SS) assumption
    documented in ch42 §5.
16. 🚧 Surfaced: `Describe` (full name inventory), `FlushSpool` (purge/
    drain), `SendTerminalMessage` (S10F1 tool→host). ⬜ Still engine-only:
    traces (S2F23), limits (S2F45), substrate/E90 + module/E157 tracking —
    surface on demand.
17. ✅ Stream responsiveness: Subscribe/WatchHealth poll at 100ms (was
    500ms) so a cancelled stream frees its sync-server worker thread
    promptly — found via the randomized stress test below.
18. ✅ Test tiers added: a randomized concurrent RPC stress case (4 threads
    × 250 seeded ops incl. Subscribe churn; TSan target; prints its seed)
    and a **virtual fab** (`interop/virtual_fab.py` + the `fab` compose
    service / `tools/spawn_fab.sh`): N daemons, each with a secsgem-py host
    AND a secsgem_client tool, driven by seeded random traffic with
    end-to-end invariant checks (round-trips, S6F11/S5F1 delivery, the
    command loop). Wired into `run_interop.sh` (now 13 steps). Verified
    green at N=3.

### Phase F — fab acceptance (parallel track; the hard gate)
- ⚠️ **Standards correctness remains unverified against SEMI texts** (behaviour
  reconstructed without the standards; interop with secsgem-py/secs4j/Wireshark
  mitigates but does not prove). The #1 fab-readiness risk; needs real
  standards access and/or a fab's MES qualification run (`docs/MES_INTEROP.md`).
- ⬜ GEM compliance statement + manual matching the tool's data dictionary.
- ⬜ SECS-I serial driver (asio `serial_port` adapter; FSM done) — only if a
  target tool uses RS-232.