rusty_agent_server/lib.rs
1//! # rusty-agent-server
2//!
3//! The network face of [`rusty_agent_runtime`]: an axum-based HTTP + SSE server
4//! implementing a pragmatic Agent-Protocol subset (see
5//! `docs/rusty-server-design.md`). The server ships as a **library** —
6//! users build their graphs, register them in a [`GraphRegistry`], and call
7//! [`serve`] (or [`router`] to embed the routes in a larger application):
8//!
9//! ```no_run
10//! use rusty_agent_runtime::prelude::*;
11//! use rusty_agent_server::{serve, GraphRegistry, ServerConfig};
12//!
13//! # async fn demo(graph: Graph, spec: StateSpec) -> std::io::Result<()> {
14//! let mut registry = GraphRegistry::new();
15//! registry.register("my_agent", graph, spec);
16//!
17//! let config = ServerConfig::new(
18//! "127.0.0.1:8080".parse().unwrap(),
19//! "./data/checkpoints",
20//! );
21//! serve(registry, config).await
22//! # }
23//! ```
24//!
25//! ## Endpoint inventory (v0.5)
26//!
27//! | Endpoint | Purpose |
28//! |---|---|
29//! | `GET /ok` | liveness |
30//! | `GET /info` | service version + registered graphs and their channels |
31//! | `POST /threads` | create a thread bound to a registered graph |
32//! | `POST /threads/{id}/fork` | time travel: copy the thread's checkpoint history (full or up to `checkpoint_id`) into a new thread |
33//! | `GET /threads/{id}/state` | latest checkpoint as `{values, next, checkpoint}` (the checkpoint ref carries its pinned `policy_version`, R0.8 wave 4) |
34//! | `POST /threads/{id}/state` | write a new checkpoint (`update_state` analog; `as_node` is accepted for LangGraph compatibility but not recorded; optional `enqueue: [...]` submits tasks atomically with the checkpoint through the R0.6 wave-2b transactional outbox — one transaction on Postgres, outbox-first best-effort ordering on the JSON-file backend) |
35//! | `POST /threads/{id}/history` | checkpoint list, newest first, `limit`/`before` |
36//! | `POST /threads/{id}/runs` | background run: `202 + run_id` |
37//! | `POST /threads/{id}/runs/wait` | blocking run: terminal result as JSON |
38//! | `POST /threads/{id}/runs/stream` | run with SSE streaming (`updates`/`values`/`messages`/`metadata`/`error`/`end`); a fresh run starts a new frame sequence, so `Last-Event-ID` is ignored here |
39//! | `GET /runs/{id}/stream` | attach to an existing run's SSE stream: replay honoring `Last-Event-ID`, then live frames |
40//! | `GET /runs/{id}/events` | Flight Recorder: the run's journaled `RunEvent`s as `{run_id, events, complete}` (snapshot flushed per checkpoint boundary and at run completion; persisted under `{store_path}/journals/` or the `server_journals` table; fetchable by run id even after the live run record is evicted or the process restarts) |
41//! | `GET /runs/{id}/fixture` | Flight Recorder: download the run as a portable `ReplayFixture` bundle (journal + graph topology hash + final checkpoint) for CI replay |
42//! | `POST /runs/replay` | Flight Recorder: re-drive a journaled run against its registered graph and verify the replayed evidence → `{run_id, verified, expected_events, actual_events, first_divergence}` (`422` when the graph is not registered in this process or the journal carries recorded effect calls) |
43//! | `GET /runs/diff?base=&branch=` | Flight Recorder: structural diff of two runs' journals (core's `BranchDiff` shape: `first_divergent_seq`, `added`/`removed` events, per-step channel diffs, token/cost totals) |
44//! | `DELETE /threads/{id}/runs/{run_id}` | rollback: delete a finished run's checkpoints (JSON-file checkpointer only; `409` on Postgres) |
45//! | `GET /runs/{run_id}` | run status polling (plus `output`/`error`/`interrupt` once terminal) |
46//! | `POST /assistants` | create a named graph alias with config metadata |
47//! | `GET /assistants` / `GET /assistants/{id}` | list / fetch assistants |
48//! | `POST /crons` | schedule recurring runs (interval secs or 5-field cron expr) |
49//! | `GET /crons` / `DELETE /crons/{id}` | list / delete crons |
50//! | `POST /triggers` | create an event-driven trigger: name, target (assistant or thread), action (`start_run` / `resume_thread` / `send_message`), `{{event.*}}` input template, per-trigger webhook secret, optional debounce window |
51//! | `GET /triggers` / `GET /triggers/{id}` / `PATCH /triggers/{id}` / `DELETE /triggers/{id}` | trigger registry (tenant-scoped) |
52//! | `POST /triggers/{id}/webhook` | signed event ingress: HMAC-SHA256 over the raw body, `X-Rusty-Signature: sha256=<hex>`, constant-time compare; `401` unsigned/invalid. Not behind the API-key layer — the signature is the credential and resolves the owning tenant |
53//! | `GET /triggers/{id}/events` | the trigger's event log (payload hash, action, run id, status), newest first |
54//! | `GET /triggers/{id}/dead-letter` | failed events, inspectable and replayable |
55//! | `POST /triggers/{id}/events/{event_id}/replay` | re-execute a logged event immediately (the replay is itself logged) |
56//! | `PUT /store/{ns}/{key}` | upsert a JSON value in a namespace (`201` create, `200` replace) |
57//! | `GET /store/{ns}/{key}` / `DELETE /store/{ns}/{key}` | fetch / delete one item |
58//! | `GET /store/{ns}` | list a namespace's items |
59//! | `POST /tasks` | R0.6 durable task queue: enqueue a task `{kind, payload, pool?, max_attempts?, idempotency_key?, worker_version?}` → `201 {task_id, deduplicated}` (`200` + `deduplicated: true` when the idempotency key already names a live task in this tenant; `429 quota_exceeded` when the tenant is over its wave-3 task quota — see `ServerConfig::with_task_quota`) |
60//! | `POST /tasks/outbox` | R0.6 wave 2b: enqueue through the transactional outbox (same payload as `POST /tasks`, same quota gate) → `202 {task_id, deduplicated}`; the relay publishes the task into the queue within one poll interval, at-least-once, deduped on the idempotency key |
61//! | `POST /tasks/claim` | R0.6: claim the oldest claimable task `{worker_id, pools?, worker_version?, lease_ms}` → `200 {task}` with a fresh lease, `204` when nothing is claimable. Wave 3: pools at their configured concurrency limit (`ServerConfig::with_pool_limit`) hand out nothing, and a task pinned with `worker_version` is leased only to a worker advertising that exact version |
62//! | `GET /tasks/metrics` | R0.6 wave 3: the autoscaling signals, tenant-scoped — per-pool queue depth, live leases, lease saturation against the configured limit, and oldest-visible-task age. Metrics, not a mechanism: the autoscaler is the operator's |
63//! | `POST /tasks/{id}/heartbeat` | R0.6: extend the held lease `{worker_id, lease_ms}` → `{lease_expires_at}`; `409` when the lease is lost |
64//! | `POST /tasks/{id}/complete` | R0.6: settle the held lease `{worker_id, result, receipt?}` → updated task record; `409` when the lease is lost. The optional `receipt` (wave 2b: `{provider, provider_id, idempotency_key, task_id?}`) is the worker's report of an idempotent effect's provider confirmation; the server journals it into the task's run as an `effect_receipt` event |
65//! | `POST /tasks/{id}/fail` | R0.6: record a failed attempt `{worker_id, error_class, message, retryable}` → `{requeued, next_attempt_at, dead, escalation}` (backoff + jitter requeue, or dead-letter); `409` when the lease is lost. R0.7 wave 2: a failed agent turn (non-`cancelled`) also feeds the agent's supervision policy — `escalation` reports where the escalation notice landed when the restart intensity was exceeded |
66//! | `GET /tasks/{id}` | R0.6: the task record (404 unknown/cross-tenant) |
67//! | `GET /tasks?status=…` | R0.6: the tenant's tasks, oldest first; `status=dead` is the DLQ listing |
68//! | `POST /agents` | R0.7 Agent Fabric (wave 1): register an agent `{agent_id, manifest, metadata?}` — the manifest is a core `CapabilityManifest` (`agent_kind`, `manifest_version`, `accepts`, …) → `201`; `409` when the id is taken |
69//! | `GET /agents` / `GET /agents/{id}` | R0.7: list the tenant's agents / fetch one registration (404 unknown/cross-tenant) |
70//! | `POST /agents/{id}/mailbox` | R0.7: send a message into the agent's mailbox `{kind, payload, idempotency_key?, …}` → `201 {task_id, deduplicated}` (`200` when the key already names a live message); `400` when the manifest's `accepts` does not declare the kind; same quota gate as `POST /tasks` |
71//! | `GET /agents/{id}/status` | R0.7: the agent's activation lease plus mailbox gauges (`queued` / `in_flight` / `dead` message counts) |
72//! | `POST /agents/{id}/activate` | R0.7: claim the agent's single activation lease `{worker_id, lease_ms}` → `200 {owner, fencing, lease_expires_at}`; `409` + the current lease when another host holds it live (expired leases are stolen, fencing bumped) |
73//! | `POST /agents/{id}/activate/heartbeat` | R0.7: renew the held activation `{worker_id, fencing, lease_ms}`; `409` on fencing loss |
74//! | `POST /agents/{id}/activate/release` | R0.7: drop the held activation `{worker_id, fencing}` so a draining host's replacement activates promptly; `409` on fencing loss |
75//! | `POST /agents/{id}/mailbox/next` | R0.7: claim the oldest queued mailbox message as one turn `{worker_id, fencing, lease_ms}` → `200 {task}` leased to the worker, `204` when empty or a turn is already in flight (one message at a time per agent is server-enforced), `409` when the activation lease is not held. The turn settles through the ordinary `/tasks/{id}/heartbeat|complete|fail` protocol |
76//! | `GET /agents/{id}/supervision` | R0.7 wave 2: the agent's supervision evidence — declared policy, latches, full attempt history, and the journaled `SupervisionEvent` / `AgentExit` events of its supervision journal (integrity re-verified on read) |
77//! | `POST /agents/{id}/cancel` | R0.7 wave 2: cancel the agent's outstanding mailbox traffic — queued/retry-scheduled messages go terminal-`cancelled`, the leased turn keeps its lease with `cancel_requested` set — and journal an `AgentExit`; idempotent (`200` with empty lists when nothing is outstanding) |
78//! | `POST /agents/{id}/restart` | R0.7 wave 2: the operator's manual restart — records a `manual_restart` supervision attempt, clears the escalation/deadline latches, and journals the event so the agent's next turn runs fresh |
79//! | `POST /teams/{team_id}/cancel` | R0.7 wave 2: cancel every agent registered with the `team_id` label, member by member with the agent-cancel semantics; `404` when no tenant agent declares the team |
80//! | `POST /memory` | R0.8 Rusty Learn (wave 1): write a governed memory record `{kind, scope, content, author, key?, tags?, priority?, confidence?, written_at?, valid_from?, valid_until?, expires_at?, supersedes?, evidence?, run_id?, parent?}` → `201 {memory_id, created, record}` (`200` + `created: false` when the content address is already stored — writes are idempotent by construction). Gates: `run` scope is runtime-only (`400`); `agent` scope requires the agent registered with `StateScope::Private` declared (`404` / `403`); `tenant` scope id must be the caller's tenant (`403`); confidence defaults to `1.0` for human authors and is required otherwise (`400`). With `run_id`, the write is journaled into that run as a `memory_write` event (best-effort) |
81//! | `GET /memory/{memory_id}` | R0.8: fetch one record by content address (`404` unknown/cross-tenant); artifact-spilled bodies are re-inlined, so the served record is self-contained |
82//! | `POST /memory/query` | R0.8: structured retrieval — the `MemoryQuery` filters plus optional `budget`, `run_id`, `parent`. `as_of` resolves at read time. With `budget`, answers the deterministic token-bounded `MemoryAssembly` (`422` when a hard budget overflows); without, the rank-ordered records. With `run_id` (budget required), the read is journaled into that run as a `memory_read` event (best-effort) |
83//! | `POST /memory/corrections` | R0.8 (wave 2): submit a human correction `{correction_id, author, target, corrected, scope, rationale?, run_id?, parent?}` → `201 {correction_id, attribution, candidate, memory_id, created, record, superseded, example_id}`. Author attribution is mandatory (validated at deserialization). Run scope adopts directly; agent scope or wider yields an attributed candidate (`candidacy: pending`, queryable via `candidates_only`). A `run_event` target additionally yields an `example`-kind record — the journaled input plus the corrected behavior — journaled into the corrected run through the memory-write seam; a `memory` target inherits the target's key, and same-key correction writes auto-supersede the prior record |
84//! | `POST /memory/consolidate` | R0.8 (wave 2): enqueue a consolidation as a durable `memory_consolidation` task `{scope, memory_ids, distiller, key?, tags?, priority?, pool?, run_id?, parent?}` → `201 {task_id, deduplicated, kind}` (deduped on scope + sorted source set). The claiming worker owns the distillation semantics and writes the `summary` record through the governed write path with the sources in `evidence.source_memory_ids` (which supersedes them) and the task payload's `written_at`; the task settles through the unchanged lease protocol |
85//! | `POST /memory/conflicts` | R0.8 (wave 2): the conflict review listing `{scope?}` → `{conflicts: [{scope, key, memory_ids, overlap}]}` — live same-key records with overlapping validity and contradictory content. Flags only; nothing is resolved |
86//! | `POST /memory/forget` | R0.8 (wave 2): erase one record `{memory_id, reason, run_id?, parent?}` → `200 {forgotten, invalidated, tombstone}`. Real deletion; dependent summaries are invalidated (deleted) transitively; a metadata-only `memory_forget` tombstone is journaled into the named run (best-effort) |
87//! | `POST /memory/forget_scope` | R0.8 (wave 2): erase every record at a scope address `{scope, reason, run_id?, parent?}` → `200 {forgotten, invalidated, tombstones}` — one tombstone per forgotten record; idempotent (empty scope → `200` with empty lists); tenant scope is self-only (`403`) |
88//! | `POST /learn/candidates` | R0.8 (wave 3): register a distilled candidate `{candidate, run_id, parent?}` → `201 {candidate_id, created, record}` (`200` + `created: false` when the content address is already stored — creation converges). The address is verified (`422`); the `candidate_created` event is journaled into `run_id`'s journal before the store write — hard-fail: an unresolvable run stops the transition (`404`) |
89//! | `GET /learn/candidates` / `GET /learn/candidates/{id}` | R0.8 (wave 3): the tenant's candidates (sorted by id) / fetch one record (`404` unknown/cross-tenant) |
90//! | `POST /learn/candidates/{id}/evaluate` | R0.8 (wave 3): drive the configured `CandidateEvaluator` `{request, run_id, parent?}` → `200 {candidate_id, status, evaluation}`; `409` when no evaluator is configured or the lifecycle forbids re-evaluation; `422` when the evaluation fails or violates the seam contract (it must name this candidate and the request's dataset version) |
91//! | `POST /learn/candidates/{id}/promote` | R0.8 (wave 3): run the promotion gate `{run_id, approval?, parent?}` → `200 {candidate_id, status, receipt, pointer}`. `403` on approval failures (out-of-envelope promotion needs an `ApprovalToken` scoped to the candidate's promotion effect id — non-transferable), `422` on evidence failures, `409` when the candidate is not `evaluated`. The status flip and the version-pointer move are one store transition |
92//! | `POST /learn/candidates/{id}/rollback` | R0.8 (wave 3): re-point the surface to the displaced version `{run_id, cause, parent?}` → `200 {candidate_id, status, receipt, pointer}`; `409` when the candidate is not `promoted` or the pointer no longer serves it. Byte-exact: the pointer's `to` is the promotion's recorded `previous`, and candidates are content-addressed — the restored version is the version that served |
93//! | `GET /learn/versions` | R0.8 (wave 3): the tenant's version pointers, sorted by surface |
94//! | `POST /policy/versions` | R0.8 (wave 4): register an immutable executor policy body `{version?, policy}` → `201 {version, record}` (`200` converged when the version already names exactly this body; `409` when it names a different one — registry immutability; `400` invalid version). Without `version`, the content-derived `policy-{hash12}` name is minted. The reserved `static-v0` floor is never registerable |
95//! | `GET /policy/versions` / `GET /policy/versions/{version}` | R0.8 (wave 4): the tenant's registered policy bodies (sorted by version) / fetch one (`404` unknown/cross-tenant; the floor resolves as a synthetic record) |
96//! | `POST /policy/activations` | R0.8 (wave 4): move the active-version pointer `{version}` → `200 {version, active}`; `422` when the version is not registered (the floor is always activatable — reverting to pre-learning behavior needs no candidate) |
97//! | `GET /policy/active` | R0.8 (wave 4): the tenant's active policy — the last activation's body, or the floor when the registry never moved |
98//! | `GET /policy/epochs` | R0.8 (wave 4): the epoch history — each activation's reign window plus the admission bindings recorded inside it (and the implicit floor epoch covering pre-activation bindings) |
99//! | `POST /capsules` | R0.9 Rusty Capsules (wave 1): register an immutable capsule manifest `{manifest}` → `201 {capsule_id, record}` (`200` converged when the content address already names exactly this manifest; `409` when the `(name, version)` pin is claimed by a different address — registry immutability; `422` when the manifest fails validation). The manifest's content address is derived; callers never mint ids |
100//! | `GET /capsules` / `GET /capsules/{id}` | R0.9 (wave 1): the tenant's registered capsule manifests (sorted by content address) / fetch one (`404` unknown/cross-tenant) |
101//! | `POST /capsules/resolve` | R0.9 (wave 1): resolve a run's capsule pins `{pins: {name: version}, run_id, parent?}` → `200 {resolutions}` — each pin's stored manifest re-derives its content address before answering (a tampered record fails closed, `422`; an unknown pin is `404`), and one `capsule_resolved` event per pin is journaled into `run_id`'s journal (hard-fail: an unresolvable run is `404`). R0.9 wave 2 adds the optional `budget` field and, with the `capsules` feature, the admission composition: Cedar decides the admission and each declared grant (`403` on refusal, one journaled `capsule_denied` per forbidden grant), overlays intersect the effective grants, and budgets compose (`422` on a token/cost overspend) |
102//! | `POST /capsule_policies/versions` / `GET /capsule_policies/versions` / `GET /capsule_policies/versions/{version}` | R0.9 (wave 2, feature `capsules`): register an immutable Cedar policy body `{policy_text, version?}` → `201 {version, record}` (`200` converged; `409` when the version names a different body; `422` unparseable) / list the tenant's bodies (sorted by version) / fetch one (`404` unknown/cross-tenant). Without the feature: `503 capsule_policy_unavailable` |
103//! | `GET /capsule_policies/active` / `POST /capsule_policies/active` | R0.9 (wave 2, feature `capsules`): the tenant's active policy body (`404` in the unconfigured posture) / move the active-version pointer `{version}` → `200 {active}` (`422` unregistered version); the move eagerly refreshes the revocation cache. Without the feature: `503 capsule_policy_unavailable` |
104//! | `POST /capsules/overlays` / `GET /capsules/overlays` / `GET /capsules/overlays/{name}` | R0.9 (wave 2, feature `capsules`): attach (or replace) a tenant overlay `{overlay}` → `201` new / `200` replaced (`403` when the active policy refuses a widening attach; `422` invalid) / list the tenant's overlays (sorted by name) / fetch one (`404` unknown/cross-tenant). Without the feature: `503 capsule_policy_unavailable` |
105//! | `GET /runs/{id}/receipt` | R0.9 (wave 3): the run's signed `RunReceipt` — minted on first request over the run's reverified journal (manifest and executor policy read back from its last checkpoint header), then stored and served while the journal's head stands; a run whose journal advanced gets a fresh receipt. `409` when nothing is persisted yet (queued or pre-checkpoint); `404` unknown/cross-tenant |
106//! | `POST /receipts/verify` | R0.9 (wave 3): verify caller-supplied evidence `{snapshot, receipt, key_id?}` → `200` with the typed `VerifiedRun` summary, or `422 receipt_verification_failed` naming the mismatched component (`journal_head` / `manifest_digest` / `effect_ledger` / `denials_ledger` / `capsule_resolutions` / `capsule_policies` / `signer_key_id` / `signature`). The public key resolves from the deployment's key history by `key_id` (default: the receipt's `signer`); an unknown key id is `404` |
107//! | `GET /receipt_keys` | R0.9 (wave 3): the deployment's signing-key history (public keys, registration and retirement instants) plus the active key id — what an auditor needs to verify receipts offline |
108//! | `POST /receipt_keys/rotate` | R0.9 (wave 3): rotate the signing key → `201 {previous_key_id, key_id, public_key, event_id}`; the new key id is journaled as a `signing_key_rotated` event in the deployment's receipts journal, and receipts signed by the retired key keep verifying against the history |
109//! | `GET /receipt_keys/journal` | R0.9 (wave 3): the deployment's key-lineage journal — the chained `signing_key_rotated` events, integrity re-verified on read |
110//! | `POST /mcp` | R0.9 (wave 4): the MCP bridge (pinned revision `2025-03-26`) — every registered graph as one tool. `initialize` / `ping` / `notifications/*` / `tools/list` (tool schemas derived from each graph's state spec: append channels are arrays, deep-merge channels objects) / `tools/call` (runs the graph on a fresh thread; plain-JSON answer, or SSE `notifications/progress` + the final response when the request accepts `text/event-stream`; a mid-stream disconnect cancels the run). JSON-RPC errors in the envelope with HTTP 200; notifications answer `202` |
111//! | `GET /.well-known/agent-card.json` | R0.9 (wave 4): the A2A agent card (pinned spec `0.3.0`), derived from the registry on every read — one skill per registered graph; deterministic, no timestamps |
112//! | `POST /a2a` | R0.9 (wave 4): the A2A JSON-RPC task surface — `message/send` (enqueue a durable task, `kind = "a2a"`, idempotent on `messageId`; a capsule data part `{"capsule": {name, version}, "input"}` routes to the in-process executor over the `a2a-capsule` pool, plain messages queue on `a2a` for external workers) / `message/stream` (the same plus SSE status events) / `tasks/get` / `tasks/cancel`. The context id maps to one Flight Recorder journal (`a2a-{tenant}-{contextId}`), so capsule executions leave their evidence on the native `/runs/{id}/events` and `/fixture` endpoints |
113//! | `PUT /capsules/{id}/blob` | R0.9 (wave 4): upload the component bytes a registered manifest's `build_digest` commits to (raw body) → `201 {capsule_id, sha256, bytes}`; `404` unknown capsule, `422` digest mismatch, `409` different bytes under a taken address (registry immutability over bytes) |
114//!
115//! Runs support `command.resume` (HITL), `config.recursion_limit`, the
116//! `reject` / `enqueue` multitask strategies (one active run per thread),
117//! `assistant_id` (resolved to its bound graph, with the assistant's
118//! `config.recursion_limit` as a default), and `checkpoint.checkpoint_id`
119//! (time-travel replay from that checkpoint instead of the latest).
120
121mod a2a;
122mod agents;
123mod assistants;
124mod auth;
125pub mod capsule_policy;
126mod capsules;
127mod coordination;
128mod crons;
129mod error;
130mod journals;
131mod learn;
132mod mcp_bridge;
133mod memory;
134mod outbox;
135mod policy;
136mod receipts;
137mod replay;
138mod routes;
139mod runs;
140mod server_store;
141mod sse;
142mod store;
143mod supervision;
144mod tasks;
145mod threads;
146mod triggers;
147
148use std::collections::HashMap;
149use std::net::SocketAddr;
150use std::path::PathBuf;
151use std::sync::Arc;
152
153use axum::Router;
154use rusty_agent_runtime::capsule::ResourceBudget;
155use rusty_agent_runtime::graph::Graph;
156use rusty_agent_runtime::learn::{CandidateEvaluator, PromotionEnvelope};
157use rusty_agent_runtime::state::StateSpec;
158use tokio_util::sync::CancellationToken;
159
160pub use error::ApiError;
161pub use learn::{DatasetSource, DirectoryDatasetSource, EvalCandidateEvaluator, EvaluationAgent};
162pub use runs::RunStatus;
163
164/// Default bound on graceful shutdown (25 s): how long
165/// [`serve_with_shutdown`] lets in-flight requests, runs, and the outbox
166/// relay finish after the shutdown signal before the server stops anyway.
167///
168/// Chosen to fit under two outer backstops: Kubernetes' default 30 s
169/// pod-termination grace (so the drain finishes before SIGKILL would
170/// arrive) and the task queue's default 30 s lease (so a drained server has
171/// released its work well before lease expiry would reassign it). The
172/// durable machinery — checkpoint resume, lease expiry — remains the
173/// correctness net; the grace bound only decides how *fast* the common
174/// case is. Override with [`ServerConfig::with_shutdown_grace`].
175pub const DEFAULT_SHUTDOWN_GRACE: std::time::Duration = std::time::Duration::from_secs(25);
176
177/// Names the JSON-file layout already owns at the store root
178/// (`agent_leases/`, `agents/`, `assistants/`, `capsules/`,
179/// `capsule_policies/`, `coordinations/`, `crons/`, `journals/`, `learn/`,
180/// `memory/`, `memory_artifacts/`, `outbox/`, `policy/`, `store/`,
181/// `tasks/`, `threads/`, `trigger_events/`, `triggers/`, plus the
182/// `latest` pointer file inside each thread's checkpoint dir).
183/// Client-chosen ids and tenant ids claiming one of these would write
184/// checkpoints into platform directories (or platform records into
185/// checkpoint dirs), so both `validate_client_id` and
186/// [`ServerConfig::with_tenant_key`] reject them.
187pub(crate) const RESERVED_NAMES: &[&str] = &[
188 "agent_leases",
189 "agents",
190 "assistants",
191 "capsules",
192 "capsule_policies",
193 "coordinations",
194 "crons",
195 "journals",
196 "keys",
197 "learn",
198 "memory",
199 "memory_artifacts",
200 "outbox",
201 "policy",
202 "receipts",
203 "store",
204 "tasks",
205 "threads",
206 "trigger_events",
207 "triggers",
208 "latest",
209];
210
211/// One registered graph: the compiled topology plus the state schema the
212/// executor needs to drive it.
213#[derive(Debug, Clone)]
214struct GraphEntry {
215 graph: Graph,
216 spec: StateSpec,
217}
218
219/// The set of graphs this server hosts — the Rust analog of the `graphs` map
220/// in LangGraph's `langgraph.json`. Registration is compile-checked in user
221/// code; a `GraphRegistry` is heterogeneous (each entry carries its own
222/// [`StateSpec`]), which is safe because `State` is a JSON map at runtime.
223#[derive(Debug, Default, Clone)]
224pub struct GraphRegistry {
225 entries: HashMap<String, GraphEntry>,
226}
227
228impl GraphRegistry {
229 pub fn new() -> Self {
230 Self::default()
231 }
232
233 /// Register a compiled graph under `name`, together with the state spec
234 /// the executor should merge its node updates through. Re-registering a
235 /// name replaces the previous entry.
236 pub fn register(
237 &mut self,
238 name: impl Into<String>,
239 graph: Graph,
240 spec: StateSpec,
241 ) -> &mut Self {
242 self.entries.insert(name.into(), GraphEntry { graph, spec });
243 self
244 }
245
246 /// `true` if a graph is registered under `name`.
247 pub fn contains(&self, name: &str) -> bool {
248 self.entries.contains_key(name)
249 }
250
251 /// All registered graph names, sorted for stable output.
252 pub fn names(&self) -> Vec<String> {
253 let mut names: Vec<String> = self.entries.keys().cloned().collect();
254 names.sort_unstable();
255 names
256 }
257
258 /// The declared channel names of a registered graph's spec, sorted.
259 pub fn channel_names(&self, name: &str) -> Vec<String> {
260 let mut channels: Vec<String> = self
261 .entries
262 .get(name)
263 .map(|entry| entry.spec.channel_names().map(str::to_owned).collect())
264 .unwrap_or_default();
265 channels.sort_unstable();
266 channels
267 }
268
269 pub(crate) fn get(&self, name: &str) -> Option<(Graph, StateSpec)> {
270 self.entries
271 .get(name)
272 .map(|entry| (entry.graph.clone(), entry.spec.clone()))
273 }
274}
275
276/// Tenant quota for the durable task queue (R0.6 wave 3): three gauges,
277/// each capped independently — tasks queued, tasks in flight, dead-letter
278/// depth — enforced at **submission** (`POST /tasks`, `POST /tasks/outbox`,
279/// and `update_state`'s atomic `enqueue` list). Over quota answers `429
280/// quota_exceeded`, the honest shape for "retry this submission later".
281///
282/// The gauges are the store's `TaskUsage` definitions: the backlog
283/// counts scheduled retries *and* outbox rows pending publication (a flood
284/// through the outbox must not bypass the quota), in flight counts every
285/// `leased` record, and the DLQ counts because an unbounded dead-letter
286/// queue is a quiet disk-full outage. `None` (the default for every field)
287/// means unlimited, preserving pre-wave-3 behavior.
288#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
289pub struct TaskQuota {
290 /// Cap on the tenant's backlog (queued + retry-scheduled + pending
291 /// outbox rows). A submission that would push the backlog past the cap
292 /// is rejected.
293 pub max_queued: Option<usize>,
294
295 /// Cap on tasks in flight (status `leased`). Pure backpressure: a
296 /// tenant already at the cap has its new submissions rejected until
297 /// workers settle.
298 pub max_in_flight: Option<usize>,
299
300 /// Cap on dead-letter depth. A tenant at the cap must inspect and
301 /// re-drive its DLQ before more work is accepted.
302 pub max_dlq: Option<usize>,
303}
304
305impl TaskQuota {
306 /// Every gauge capped: `TaskQuota { max_queued: Some(q),
307 /// max_in_flight: Some(f), max_dlq: Some(d) }`. Field-by-field
308 /// construction works too — the struct's fields are public and each
309 /// `None` means unlimited.
310 pub fn capped(max_queued: usize, max_in_flight: usize, max_dlq: usize) -> Self {
311 Self {
312 max_queued: Some(max_queued),
313 max_in_flight: Some(max_in_flight),
314 max_dlq: Some(max_dlq),
315 }
316 }
317
318 /// `true` when no gauge is capped — the quota gate short-circuits
319 /// without a store read (the default configuration's fast path).
320 pub(crate) fn is_unlimited(&self) -> bool {
321 self.max_queued.is_none() && self.max_in_flight.is_none() && self.max_dlq.is_none()
322 }
323}
324
325/// Server configuration.
326///
327/// Checkpointing is rooted at `store_path` via
328/// [`rusty_agent_runtime::checkpoint::JsonFileCheckpointer`]. Auth maps static API
329/// keys (checked against the `X-Api-Key` header) to tenants: the legacy
330/// single [`ServerConfig::with_api_key`] maps its key to the `default`
331/// tenant, while [`ServerConfig::with_tenant_key`] adds `(tenant, key)`
332/// pairs for multi-tenant deployments. With no keys configured the server
333/// runs in open (dev) mode — no header required, everything lives in the
334/// `default` tenant. Every tenant-scoped resource (threads + checkpoints,
335/// assistants, crons, KV namespaces) is isolated per tenant; cross-tenant
336/// access answers `404`. With the `postgres` feature,
337/// `ServerConfig::with_postgres` switches **both** the run checkpointer
338/// (core's `PostgresCheckpointer`) and the server store (assistants /
339/// crons / threads / KV) to Postgres.
340#[derive(Debug, Clone)]
341pub struct ServerConfig {
342 /// Address to bind when using [`serve`] (default `0.0.0.0:8080`).
343 pub bind_addr: SocketAddr,
344
345 /// Root directory for checkpoint files
346 /// (`{store_path}/{thread_id}/{checkpoint_id}.json`). Also roots the
347 /// JSON-file assistants/crons/threads/KV persistence. Unused for
348 /// checkpointing when `database_url` is set (still used as the
349 /// `store_path` reported by `GET /info`).
350 pub store_path: PathBuf,
351
352 /// Postgres connection URL. When set (requires the `postgres` feature —
353 /// see `ServerConfig::with_postgres`), checkpoints live in core's
354 /// `rusty_checkpoints` table and the platform surface in the
355 /// `server_assistants` / `server_crons` / `server_threads` /
356 /// `server_kv` / `server_journals` tables, all auto-migrated on
357 /// connect. Connections are established lazily on first use.
358 pub database_url: Option<String>,
359
360 /// Per-thread in-flight run cap used as the **enqueue queue depth**
361 /// (default 1). There is always at most one *active* run per thread.
362 pub max_concurrent_runs_per_thread: usize,
363
364 /// Static API key required via the `X-Api-Key` header, mapped to the
365 /// `default` tenant (legacy single-key mode). `None` (the default) with
366 /// an empty [`ServerConfig::api_keys`] is dev mode: no authentication.
367 pub api_key: Option<String>,
368
369 /// Additional `(api_key, tenant)` pairs for multi-tenant deployments
370 /// (see [`ServerConfig::with_tenant_key`]). Each key maps to exactly one
371 /// tenant; every tenant's threads, assistants, crons, and KV namespaces
372 /// are isolated from all others (cross-tenant access answers `404`).
373 pub api_keys: Vec<(String, String)>,
374
375 /// Per-run SSE event-log capacity (frames retained for replay, default
376 /// 1000).
377 pub event_log_capacity: usize,
378
379 /// How often the transactional-outbox relay publishes pending rows into
380 /// the task queue (R0.6 wave 2b, default 250 ms). The relay is
381 /// crash-safe at any interval — pending rows survive restarts and
382 /// publishing dedupes on the task's idempotency key — so this only
383 /// tunes outbox-to-queue latency, not correctness.
384 pub outbox_relay_interval: std::time::Duration,
385
386 /// Bound on graceful shutdown (R0.6 wave 2c, default
387 /// [`DEFAULT_SHUTDOWN_GRACE`]): after the shutdown signal, in-flight
388 /// requests and runs get this long to finish before the server stops
389 /// regardless. Runs stopped by the drain are resumable from their last
390 /// checkpoint, so a short grace costs a resume, never work.
391 pub shutdown_grace: std::time::Duration,
392
393 /// Per-pool in-flight caps for the durable task queue (R0.6 wave 3):
394 /// pool name → maximum live leases. The claim path counts a pool's
395 /// unexpired leases and stops handing out new ones at the cap, so
396 /// pools coexist without starving each other (a saturated GPU pool
397 /// never blocks IO-pool claims). Pools without an entry are
398 /// uncapped — the pre-wave-3 behavior. See
399 /// [`ServerConfig::with_pool_limit`].
400 pub task_pool_limits: HashMap<String, usize>,
401
402 /// The default tenant quota (R0.6 wave 3), applied to every tenant
403 /// without an entry in [`ServerConfig::tenant_task_quotas`]. All
404 /// gauges uncapped by default. See [`TaskQuota`].
405 pub task_quota: TaskQuota,
406
407 /// Per-tenant quota overrides (R0.6 wave 3), replacing
408 /// [`ServerConfig::task_quota`] wholesale for the named tenant (an
409 /// override *is* the tenant's quota, not a patch on the default).
410 /// See [`ServerConfig::with_tenant_quota`].
411 pub tenant_task_quotas: HashMap<String, TaskQuota>,
412
413 /// The promotion envelope (R0.8 Rusty Learn, wave 3): the declared,
414 /// per-deployment standing approval the promotion gate evaluates
415 /// every promotion against. Defaults to
416 /// [`PromotionEnvelope::r08_default`] — memory-set candidates at
417 /// run/agent scope auto-promote on cleared evidence; everything else
418 /// requires an approval token. See
419 /// [`ServerConfig::with_promotion_envelope`].
420 pub promotion_envelope: PromotionEnvelope,
421
422 /// The candidate evaluator (R0.8 wave 3): the evaluation
423 /// composition `POST /learn/candidates/{id}/evaluate` drives. `None`
424 /// (the default) answers `409` — a deployment without an evaluator
425 /// can hold and inspect candidates, but promotion is gated on
426 /// evidence, and evidence requires an evaluator. See
427 /// [`ServerConfig::with_candidate_evaluator`].
428 pub candidate_evaluator: Option<Arc<dyn CandidateEvaluator>>,
429
430 /// Operator-authored Cedar policy files (R0.9 Rusty Capsules, wave
431 /// 2), loaded for the `default` tenant at startup — the deployment's
432 /// standing authorization, the way static API keys are its standing
433 /// authentication. Requires the `capsules` feature; each file is
434 /// registered under its content-derived version
435 /// (`cedar-{sha256[..12]}`), never activated — activation is an
436 /// explicit operator move through `POST /capsule_policies/active`. See
437 /// [`ServerConfig::with_capsule_policy_file`].
438 pub capsule_policy_files: Vec<PathBuf>,
439
440 /// The tenant-wide budget ceiling (R0.9 wave 2): the outermost bound
441 /// every capsule budget composes under. Fuel, memory, wall time, and
442 /// output bytes clamp down to it; a run declaring more `max_tokens`
443 /// or `max_cost_usd` than the ceiling (or than the run's own budget)
444 /// is refused `422` — token and cost bounds cannot be retrofitted
445 /// mid-run the way fuel can, so admission refuses the overspend. See
446 /// [`ServerConfig::with_capsule_budget_ceiling`].
447 pub capsule_budget_ceiling: Option<ResourceBudget>,
448
449 /// The deployment's network-egress seam for capsules the A2A bridge
450 /// executes in-process (R0.9 wave 4): the [`NetworkConnector`] every
451 /// bridge-built capsule host is constructed with. Without one, A2A
452 /// tasks carrying a capsule payload fail closed with
453 /// `capsule_execution_unavailable` — the bridge never executes guest
454 /// code against an egress path the operator did not explicitly wire.
455 /// Requires the `capsules` feature. See
456 /// [`ServerConfig::with_capsule_connector`].
457 #[cfg(feature = "capsules")]
458 pub capsule_connector: Option<Arc<dyn rusty_agent_runtime::capsule_host::NetworkConnector>>,
459}
460
461impl Default for ServerConfig {
462 fn default() -> Self {
463 Self {
464 bind_addr: SocketAddr::from(([0, 0, 0, 0], 8080)),
465 store_path: PathBuf::from("./data/checkpoints"),
466 database_url: None,
467 max_concurrent_runs_per_thread: 1,
468 api_key: None,
469 api_keys: Vec::new(),
470 event_log_capacity: 1000,
471 outbox_relay_interval: crate::outbox::DEFAULT_RELAY_INTERVAL,
472 shutdown_grace: DEFAULT_SHUTDOWN_GRACE,
473 task_pool_limits: HashMap::new(),
474 task_quota: TaskQuota::default(),
475 tenant_task_quotas: HashMap::new(),
476 promotion_envelope: PromotionEnvelope::r08_default(),
477 candidate_evaluator: None,
478 capsule_policy_files: Vec::new(),
479 capsule_budget_ceiling: None,
480 #[cfg(feature = "capsules")]
481 capsule_connector: None,
482 }
483 }
484}
485
486impl ServerConfig {
487 /// A config with the given bind address and checkpoint store root;
488 /// everything else at its default.
489 pub fn new(bind_addr: SocketAddr, store_path: impl Into<PathBuf>) -> Self {
490 Self {
491 bind_addr,
492 store_path: store_path.into(),
493 ..Self::default()
494 }
495 }
496
497 /// Builder-style: require an API key on every request. The key maps to
498 /// the `default` tenant (legacy single-tenant mode).
499 pub fn with_api_key(mut self, key: impl Into<String>) -> Self {
500 self.api_key = Some(key.into());
501 self
502 }
503
504 /// Builder-style: map an API key to a tenant (multi-tenant mode). Every
505 /// request presenting `key` via `X-Api-Key` runs as `tenant`, fully
506 /// isolated from all other tenants. Tenant ids must match
507 /// `[A-Za-z0-9._-]` (1–64 chars) and must not be a reserved layout name
508 /// (`assistants`, `crons`, `store`, `threads`, `latest`) — they become a
509 /// path segment in the JSON-file layout and a `{tenant}/` id prefix
510 /// everywhere else.
511 ///
512 /// # Panics
513 ///
514 /// Panics on an empty key or an invalid tenant id (configuration is a
515 /// programmer error, caught at startup).
516 pub fn with_tenant_key(mut self, tenant: impl Into<String>, key: impl Into<String>) -> Self {
517 let tenant = tenant.into();
518 let key = key.into();
519 let valid = !tenant.is_empty()
520 && tenant.len() <= 64
521 && tenant
522 .chars()
523 .all(|c| c.is_ascii_alphanumeric() || matches!(c, '_' | '-' | '.'))
524 && !RESERVED_NAMES.contains(&tenant.as_str());
525 assert!(
526 valid,
527 "invalid tenant id `{tenant}` (allowed: [A-Za-z0-9._-], 1..=64 chars, not a reserved name)"
528 );
529 assert!(
530 !key.is_empty(),
531 "API key for tenant `{tenant}` must not be empty"
532 );
533 self.api_keys.push((key, tenant));
534 self
535 }
536
537 /// `true` when at least one API key is configured (legacy `api_key` or
538 /// any tenant key), i.e. requests must authenticate.
539 pub fn auth_enabled(&self) -> bool {
540 self.api_key.is_some() || !self.api_keys.is_empty()
541 }
542
543 /// The tenant a presented API key maps to, or `None` for unknown keys.
544 /// Tenant keys are checked first (last registration wins on duplicate
545 /// keys); the legacy `api_key` maps to the `default` tenant.
546 pub fn tenant_for_key(&self, key: &str) -> Option<&str> {
547 if let Some((_, tenant)) = self.api_keys.iter().rev().find(|(k, _)| k == key) {
548 return Some(tenant.as_str());
549 }
550 if self.api_key.as_deref() == Some(key) {
551 return Some("default");
552 }
553 None
554 }
555
556 /// Builder-style: persist everything in Postgres at `url` (e.g.
557 /// `postgres://user:pass@localhost/rusty`). Switches the run
558 /// checkpointer to [`rusty_agent_runtime::checkpoint_postgres::PostgresCheckpointer`]
559 /// **and** the assistants/crons/threads/KV/journals server store to the
560 /// `server_*` tables. Schemas auto-migrate on (lazy) connect.
561 #[cfg(feature = "postgres")]
562 pub fn with_postgres(mut self, url: impl Into<String>) -> Self {
563 self.database_url = Some(url.into());
564 self
565 }
566
567 /// Builder-style: set the per-thread enqueue queue depth cap. Values
568 /// below 1 are clamped to 1 (a zero-deep queue would reject every
569 /// `enqueue` run).
570 pub fn with_max_concurrent_runs_per_thread(mut self, cap: usize) -> Self {
571 self.max_concurrent_runs_per_thread = cap;
572 self
573 }
574
575 /// Builder-style: set the per-run SSE event-log capacity. Values below
576 /// 16 are clamped to 16 (replay needs room for at least the
577 /// metadata/updates/end frames of a minimal run).
578 pub fn with_event_log_capacity(mut self, capacity: usize) -> Self {
579 self.event_log_capacity = capacity;
580 self
581 }
582
583 /// Builder-style: set the transactional-outbox relay's poll interval
584 /// (R0.6 wave 2b). Tests can set a long interval to drive publishing
585 /// deterministically; correctness never depends on the interval.
586 pub fn with_outbox_relay_interval(mut self, interval: std::time::Duration) -> Self {
587 self.outbox_relay_interval = interval;
588 self
589 }
590
591 /// Builder-style: set the graceful-shutdown bound (R0.6 wave 2c; see
592 /// [`DEFAULT_SHUTDOWN_GRACE`] for the default's rationale). Past this
593 /// window the server stops even if requests or runs are still in flight
594 /// — durability (checkpoint resume, lease expiry) is the backstop.
595 pub fn with_shutdown_grace(mut self, grace: std::time::Duration) -> Self {
596 self.shutdown_grace = grace;
597 self
598 }
599
600 /// Builder-style: cap the named pool's live leases (R0.6 wave 3). The
601 /// claim path counts a pool's unexpired leases and stops handing out
602 /// new ones at `max_in_flight`, so a saturated pool (GPU-bound work)
603 /// never starves the others (IO-bound work). A cap of `0` pauses the
604 /// pool: nothing is leased from it until the cap is raised. Pools
605 /// without an entry stay uncapped.
606 ///
607 /// # Panics
608 ///
609 /// Panics on an invalid pool name (configuration is a programmer error,
610 /// caught at startup); names must match `[A-Za-z0-9._-]` (1–128 chars),
611 /// the same rule the enqueue and claim paths validate.
612 pub fn with_pool_limit(mut self, pool: impl Into<String>, max_in_flight: usize) -> Self {
613 let pool = pool.into();
614 assert!(
615 crate::tasks::validate_pool(&pool).is_ok(),
616 "invalid pool name `{pool}` (allowed: [A-Za-z0-9._-], 1..=128 chars)"
617 );
618 self.task_pool_limits.insert(pool, max_in_flight);
619 self
620 }
621
622 /// Builder-style: set the default tenant quota for the durable task
623 /// queue (R0.6 wave 3), applied to every tenant without a
624 /// [`ServerConfig::with_tenant_quota`] override. Over quota, task
625 /// submissions answer `429 quota_exceeded`. The default is uncapped on
626 /// every gauge. See [`TaskQuota`] for the gauge definitions.
627 pub fn with_task_quota(mut self, quota: TaskQuota) -> Self {
628 self.task_quota = quota;
629 self
630 }
631
632 /// Builder-style: override the task quota for one tenant (R0.6 wave 3).
633 /// The override replaces [`ServerConfig::task_quota`] wholesale for
634 /// that tenant — an uncapped gauge in the override is unlimited even if
635 /// the default caps it.
636 ///
637 /// # Panics
638 ///
639 /// Panics on an invalid tenant id (same rule as
640 /// [`ServerConfig::with_tenant_key`]).
641 pub fn with_tenant_quota(mut self, tenant: impl Into<String>, quota: TaskQuota) -> Self {
642 let tenant = tenant.into();
643 let valid = !tenant.is_empty()
644 && tenant.len() <= 64
645 && tenant
646 .chars()
647 .all(|c| c.is_ascii_alphanumeric() || matches!(c, '_' | '-' | '.'))
648 && !RESERVED_NAMES.contains(&tenant.as_str());
649 assert!(
650 valid,
651 "invalid tenant id `{tenant}` (allowed: [A-Za-z0-9._-], 1..=64 chars, not a reserved name)"
652 );
653 self.tenant_task_quotas.insert(tenant, quota);
654 self
655 }
656
657 /// The quota governing `tenant`: its override when one is configured,
658 /// else the server-wide default ([`ServerConfig::task_quota`]).
659 pub(crate) fn quota_for(&self, tenant: &str) -> &TaskQuota {
660 self.tenant_task_quotas
661 .get(tenant)
662 .unwrap_or(&self.task_quota)
663 }
664
665 /// Builder-style: set the promotion envelope (R0.8 Rusty Learn, wave
666 /// 3) — the declared, per-deployment standing approval the promotion
667 /// gate evaluates every promotion against. Replaces
668 /// [`PromotionEnvelope::r08_default`].
669 ///
670 /// # Panics
671 ///
672 /// Panics when the envelope fails [`PromotionEnvelope::validate`]
673 /// (a canary fraction outside `(0, 1]`, a negative improvement bar —
674 /// configuration is a programmer error, caught at startup).
675 pub fn with_promotion_envelope(mut self, envelope: PromotionEnvelope) -> Self {
676 if let Err(error) = envelope.validate() {
677 panic!("invalid promotion envelope: {error}");
678 }
679 self.promotion_envelope = envelope;
680 self
681 }
682
683 /// Builder-style: register the candidate evaluator (R0.8 wave 3) —
684 /// the evaluation composition `POST /learn/candidates/{id}/evaluate`
685 /// drives. Without one the route answers `409`: promotion is gated
686 /// on evidence, and evidence requires an evaluator.
687 pub fn with_candidate_evaluator(mut self, evaluator: Arc<dyn CandidateEvaluator>) -> Self {
688 self.candidate_evaluator = Some(evaluator);
689 self
690 }
691
692 /// Builder-style: add one operator-authored Cedar policy file (R0.9
693 /// Rusty Capsules, wave 2), loaded for the `default` tenant at
694 /// startup. Files are *registered*, never activated — moving a
695 /// tenant's active pointer is an explicit operator move through
696 /// `POST /capsule_policies/active`, so a restart can never silently
697 /// change what serves. Without the `capsules` feature the files are
698 /// ignored (the plane's routes answer `503` either way).
699 pub fn with_capsule_policy_file(mut self, path: impl Into<PathBuf>) -> Self {
700 self.capsule_policy_files.push(path.into());
701 self
702 }
703
704 /// Builder-style: set the tenant-wide capsule budget ceiling (R0.9
705 /// wave 2) — the outermost bound every capsule budget composes
706 /// under. `None` (the default) leaves budgets bounded only by the
707 /// run's own declaration and each capsule's manifest.
708 pub fn with_capsule_budget_ceiling(mut self, ceiling: ResourceBudget) -> Self {
709 self.capsule_budget_ceiling = Some(ceiling);
710 self
711 }
712
713 /// Builder-style: set the network connector for capsules the A2A
714 /// bridge executes in-process (R0.9 wave 4) — the deployment's
715 /// explicit egress seam. `None` (the default) refuses capsule
716 /// execution over A2A with `capsule_execution_unavailable`: guest
717 /// code never runs against an egress path the operator did not wire.
718 #[cfg(feature = "capsules")]
719 pub fn with_capsule_connector(
720 mut self,
721 connector: Arc<dyn rusty_agent_runtime::capsule_host::NetworkConnector>,
722 ) -> Self {
723 self.capsule_connector = Some(connector);
724 self
725 }
726}
727
728/// Build the axum [`Router`] for a registry and config. Use this to embed the
729/// rusty-agent-server routes into a larger application, or to drive the API in tests
730/// via `tower::ServiceExt::oneshot`.
731///
732/// The router's drain control is internal and never fires (embedders who
733/// want cooperative drain should use [`router_with_shutdown`]).
734pub fn router(registry: GraphRegistry, config: ServerConfig) -> Router {
735 router_with_shutdown(registry, config, CancellationToken::new())
736}
737
738/// Build the axum [`Router`] with an explicit drain control (R0.6 wave 2c).
739///
740/// Cancelling `shutdown` starts the cooperative drain **inside** the
741/// application — in-flight runs stop at their next checkpoint boundary, the
742/// outbox relay finishes its current pass and stops, the cron scheduler
743/// stops firing, and new run submissions are rejected with `503` — while
744/// the HTTP surface keeps serving so in-flight requests can complete.
745/// Stopping HTTP itself is the embedder's job (axum's
746/// `with_graceful_shutdown`); [`serve_with_shutdown`] wires both halves
747/// together.
748pub fn router_with_shutdown(
749 registry: GraphRegistry,
750 config: ServerConfig,
751 shutdown: CancellationToken,
752) -> Router {
753 routes::router_with_shutdown(registry, config, shutdown)
754}
755
756/// Build the router and bind it to `config.bind_addr`. Blocks until the
757/// server shuts down: SIGINT or SIGTERM starts the graceful drain described
758/// on [`serve_with_shutdown`].
759pub async fn serve(registry: GraphRegistry, config: ServerConfig) -> std::io::Result<()> {
760 serve_with_shutdown(registry, config, shutdown_signal()).await
761}
762
763/// Build the router, bind it to `config.bind_addr`, and serve until
764/// `shutdown` resolves, then drain gracefully (R0.6 wave 2c). The drain is
765/// ordered so the common case of a rolling deploy strands nothing:
766///
767/// 1. **Stop taking new work.** axum stops accepting connections; the
768/// server's shared drain token fires, so new run submissions answer
769/// `503` and the cron scheduler stops firing.
770/// 2. **Let in-flight work land.** In-flight requests complete (axum waits
771/// for them); in-flight runs are cooperatively cancelled at their next
772/// super-step boundary — where a checkpoint was just persisted — and
773/// end terminal-`cancelled`, resumable by re-running the thread; the
774/// outbox relay finishes its current publish pass and stops (pending
775/// rows are durable and publish on the next process's first pass).
776/// 3. **Stop, bounded.** The whole drain is capped at
777/// [`ServerConfig::shutdown_grace`]. Past it the server returns anyway:
778/// anything still in flight is abandoned mid-step, which is exactly the
779/// crash case the durable machinery already covers — runs resume from
780/// their last checkpoint and leased tasks return to visibility within
781/// one lease period. Grace makes the common case fast; it is never the
782/// correctness mechanism.
783pub async fn serve_with_shutdown(
784 registry: GraphRegistry,
785 config: ServerConfig,
786 shutdown: impl std::future::Future<Output = ()> + Send + 'static,
787) -> std::io::Result<()> {
788 let addr = config.bind_addr;
789 let grace = config.shutdown_grace;
790 // Open (dev) mode on a non-loopback address exposes the full API — run
791 // creation, KV writes, checkpoint deletion — to the network. That's a
792 // legitimate dev choice, but it must never be a quiet one.
793 if !config.auth_enabled() && !addr.ip().is_loopback() {
794 tracing::warn!(
795 %addr,
796 "serving WITHOUT authentication on a non-loopback address; \
797 configure `with_api_key`/`with_tenant_key` or bind 127.0.0.1"
798 );
799 }
800 let draining = CancellationToken::new();
801 let app = router_with_shutdown(registry, config, draining.clone());
802 let listener = tokio::net::TcpListener::bind(addr).await?;
803 tracing::info!(%addr, "rusty-server listening");
804 let server = axum::serve(listener, app).with_graceful_shutdown({
805 let draining = draining.clone();
806 async move {
807 shutdown.await;
808 tracing::info!(
809 grace_ms = grace.as_millis() as u64,
810 "shutdown requested; draining"
811 );
812 // One token drives the whole cooperative drain: runs stop at
813 // their next checkpoint boundary, the relay finishes its pass,
814 // the cron scheduler stops, new runs are rejected.
815 draining.cancel();
816 }
817 });
818 tokio::select! {
819 // `WithGracefulShutdown` is `IntoFuture`, not `Future`, so it needs
820 // this async wrapper to sit in a `select!` arm.
821 result = async { server.await } => result,
822 () = async {
823 // The grace clock starts when the drain does. axum's graceful
824 // shutdown waits for in-flight connections without a bound of
825 // its own, so this arm is the bound.
826 draining.cancelled().await;
827 tokio::time::sleep(grace).await;
828 } => {
829 tracing::warn!(
830 grace_ms = grace.as_millis() as u64,
831 "drain grace expired; forcing shutdown — in-flight runs resume \
832 from their last checkpoint and leased tasks return at lease expiry"
833 );
834 Ok(())
835 }
836 }
837}
838
839/// The default shutdown signal for [`serve`]: resolves on SIGINT or SIGTERM
840/// (the two signals orchestrators and terminals actually send). Exposed so
841/// binaries that compose their own shutdown logic can reuse it.
842pub async fn shutdown_signal() {
843 let ctrl_c = tokio::signal::ctrl_c();
844 #[cfg(unix)]
845 {
846 // Signal-handler installation only fails when the fd limit is
847 // exhausted; a server that cannot hear SIGTERM must not start
848 // pretending it drains gracefully.
849 let mut sigterm = tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate())
850 .expect("installing the SIGTERM handler must succeed");
851 tokio::select! {
852 _ = ctrl_c => {},
853 _ = sigterm.recv() => {},
854 }
855 }
856 #[cfg(not(unix))]
857 {
858 let _ = ctrl_c.await;
859 }
860}