Enforce AccessControl on overlay ops in OverlayOperationEnv::invoke_with_policy
(alknet-call) so the hub's calls to browser-registered ops are gated by the
browser's AccessControl — matching OperationRegistry::invoke semantics for
internal composition (caller identity = parent handler_identity.as_identity()).
Add src/websocket/overlay.rs with 19 integration tests covering the connection-
local overlay acceptance criteria: browser ops land in the per-CallConnection
overlay (not PeerCompositeEnv), no PeerId for the browser, register_imported()/
register_imported_all() populate the overlay, hub outgoing calls route through
overlay_env() (not PeerRef::Specific), PeerRef::Specific('browser-X') routes to
NOT_FOUND, AccessControl gates hub calls (allowed/forbidden/default), overlay is
per-connection isolated and dropped on WS close, WS close aborts in-flight calls
with ADR-016 cascade, bidirectionality, and browser-with-no-ops use-case scoping.
Implements src/server/gateway_routes.rs: POST /call, GET /search, GET /schema,
POST /batch, POST /subscribe (SSE). All delegate to GatewayDispatch::invoke; auth
via ResolvedIdentity extractor; errors mapped via call_error_to_http_response
(identity-aware 401/403 split). Internal ops → 404. /schema adds ACL pre-check.
/subscribe projects ResponseEnvelope as SSE. /batch loops over invoke. Wired real
handlers into adapter.rs replacing placeholder 501s. 157 tests pass.
Note: /subscribe SSE completes after single event (registry invoke returns single
ResponseEnvelope, no streaming handler yet — research §6 OQ#5).
# Conflicts:
# crates/alknet-http/src/server/adapter.rs
Implement the sole HTTP invoke path (ADR-042/047) in
src/server/gateway_routes.rs: POST /call reads { operation, input },
resolves identity via the shared ResolvedIdentity extractor, dispatches
via GatewayDispatch::invoke, and returns ResponseEnvelope as JSON with
errors mapped via call_error_to_http_response. GET /search dispatches
services/list (AccessControl-filtered); GET /schema dispatches
services/schema with an ACL pre-check (unauthorized -> 403). POST /batch
loops over invoke returning an array; POST /subscribe projects the
response as SSE (text/event-stream) with data frames for call.responded
and an error event for call.aborted. Internal ops return 404. Wire the
real handlers into adapter.rs router, replacing the placeholder 501s.
Implements src/websocket/upgrade.rs: WS upgrade at /alknet/call using axum
WebSocketUpgrade, bearer auth (no token → 401), CallConnection::new_overlay_only,
native EventEnvelope over binary WS messages (no length prefix, text → close 1002),
Dispatcher::dispatch_requested for call.requested (AccessControl gates), handle_abort
for call.aborted, PendingRequestMap correlation, fail_all_pending on disconnect
(ADR-016 cascade), bidirectionality via connection-local overlay. Wired /alknet/call
route into adapter.rs. 168 tests pass.
Add src/server/auth.rs with bearer_auth_middleware axum layer that
extracts the Authorization: Bearer header, resolves via
IdentityProvider::resolve_from_token, and stashes Option<Identity> in
request extensions. Shared by HTTP gateway routes and the to_mcp rmcp
service (research §4.4). No token, malformed header, or failed
resolution all yield None (unauthenticated, not an error) — Bearer-only
auth mechanism (ADR-004).
Includes ResolvedIdentity axum extractor reading from extensions, and
wires the middleware into the HttpAdapter router around the
gateway/openapi/mcp routes (excluding the raw /healthz route).
GET /healthz: raw route, no auth, no OperationContext, returns 200 OK
with plain-text 'ok' (ADR-036). Decoy fallback for unknown paths via
DecoyConfig: fake nginx 404 (default), static site serving, or redirect.
Decoy does not leak alknet presence (no alknet headers/format). Custom
routes take precedence over decoy (decoy is fallback only). Wire real
handlers into HttpAdapter router replacing placeholder 501s.
FromMCP (OperationAdapter, feature-gated behind mcp) discovers remote MCP
tools over streamable HTTP via rmcp's StreamableHttpClientTransport, calls
tools/list, and registers each as a HandlerRegistration bundle with a
forwarding handler that calls the remote tool via tools/call. Output
handling follows the structuredContent-preferred-over-content-blocks rule:
declared outputSchema + structuredContent is the composable result; absent
outputSchema falls back to the MCP ContentBlock union. isError:true maps to
a CallError with the error content. No-env-vars invariant: the handler reads
context.capabilities (injected at registration), never std::env::var (ADR-014).
Streamable HTTP only — stdio is not built (ADR-037). Provenance is FromMCP
(leaf: composition_authority None, scoped_env None, Internal by default,
ADR-015/022). Includes unit tests for schema/mapping logic and an integration
test that spins up a real rmcp streamable HTTP server and exercises the
forwarding handler end-to-end.
Parses OpenAPI 3.x documents into HandlerRegistration bundles with
reqwest-backed forwarding handlers that inject credentials from
OperationContext.capabilities (no-env-vars invariant, ADR-014).
Error codes are prefixed HTTP_<status> (ADR-023); ops are Internal
leaves with FromOpenAPI provenance (ADR-015/022). SSE subscriptions
are consumed via parseSSEFrames; JSON/text/binary response branching
mirrors the TS prior art.
Wires the axum Router (gateway endpoints + /healthz + /openapi.json + MCP +
custom routes via extra_routes merge ADR-046) and drives hyper's HTTP/1.1 or
HTTP/2 connection driver over a single QUIC bidirectional stream. The
QUIC-to-hyper bridge wraps the (SendStream, RecvStream) pair as a
TokioIo-compatible duplex and feeds it to hyper-util's auto Builder (which
auto-detects HTTP/1.1 vs HTTP/2). h3 ALPN is not registered (ADR-044).
Route handlers, healthz/decoy logic, openapi.json, the MCP route, and the WS
upgrade handler are wired as 501 Not Implemented placeholders for their
respective tasks. The router state holds Arc<OperationRegistry> +
Arc<dyn IdentityProvider>; the router is built once at construction and
cloned per connection (cheap Arc clone). DecoyConfig defaults to NotFound.
Adds hyper-util dependency (server, service, tokio features).
Implements SharedHttpClient (ArcSwap<ClientWithMiddleware>) with HttpClientConfig
(pool/timeout/retry/optional CA bundle+client cert), RetryTransientMiddleware from
reqwest-retry, and inlined RetryAfterMiddleware (~90 lines, bounded HashMap with LRU
eviction, parses Retry-After seconds + HTTP-date, sleeps on 429/503). reload() via
ArcSwap. No env-var reads; per-request credential injection only. 24 unit tests.
Adds SharedHttpClient wrapping ArcSwap<ClientWithMiddleware> with a
RetryTransientMiddleware + inlined RetryAfterMiddleware stack.
HttpClientConfig covers pool, timeout, retry policy, and optional CA
bundle/client cert. reload() rebuilds and swaps via ArcSwap. No env-var
reads; credential injection is per-request, not at construction.
Make Dispatcher::dispatch_requested pub and extract abort-cascade handling
into a pub handle_abort method so the WebSocket handler can feed deserialized
EventEnvelopes directly to the shared Dispatcher without a QUIC Connection.
CallConnection gains a new_overlay_only(identity) constructor (Option A) that
holds the Layer 2 overlay, PendingRequestMap, and resolved bearer Identity
without a QUIC Connection; identity() reads the stored field for the non-QUIC
case. compose_root_env uses the new identity() accessor for both paths.
The existing QUIC path (CallAdapter, CallClient, run_loop, handle_stream) is
unchanged — outgoing client methods guard on connection().is_none().
Implements call_error_to_http_status, call_error_to_http_status_with_identity,
and call_error_to_http_response in src/gateway/error.rs. Five protocol codes
map to fixed statuses (404/422/504/500 + 401/403 split for FORBIDDEN).
HTTP_<status>-prefixed operation-level codes parse status from prefix. Unknown
operation-level codes default to 500. Retry-After header for retryable 503/429.
21 unit tests.
# Conflicts:
# crates/alknet-http/src/gateway/mod.rs
Add gateway/error.rs with call_error_to_http_status,
call_error_to_http_status_with_identity, and call_error_to_http_response.
Maps the five protocol codes (NOT_FOUND/FORBIDDEN/INVALID_INPUT/TIMEOUT/
INTERNAL) to fixed HTTP statuses, splits FORBIDDEN into 401 (no identity) /
403 (identity present), maps HTTP_<status>-prefixed operation-level codes
to the status number (from_openapi fidelity), and defaults unknown
operation-level codes to 500. Retryable 503/429 errors carry a Retry-After
header when details.retry_after is present.
Thin concrete struct (not a trait) holding Arc<OperationRegistry> +
Arc<dyn IdentityProvider>. Exposes resolve_bearer() (delegates to
identity_provider.resolve_from_token) and invoke() which builds a root
OperationContext for wire-ingress (internal: false, forwarded_for: None,
fresh UUID v4 request_id, deadline now+30s) carrying the registration
bundle's composition_authority/capabilities/scoped_env, then calls
OperationRegistry::invoke. Dispatches services/list and services/schema
unchanged (registered ops); AccessControl filtering in services/list
sees the caller's resolved identity. Re-exported from lib.rs.
Duplicates Dispatcher::build_root_context construction (the alknet-call
version is pub(crate) and tangled with CallConnection peer/session
overlays); the invariants (internal: false, forwarded_for: None) are
the load-bearing part and identical to the wire-ingress path.
Add crates/alknet-http with Cargo.toml, src/lib.rs, and the five
subsystem modules (server, gateway, client, adapters, websocket) per
ADR-039 (server + client host colocated). The mcp feature gate pulls in
rmcp with streamable HTTP transport features only (ADR-037 — no stdio);
h3/WebTransport is absent (deferred per ADR-044). alknet-core and
alknet-call use workspace path deps. The crate is added to the workspace
members list.
Break the alknet-http architecture spec into atomic, dependency-ordered
tasks in tasks/http/, following the taskgraph frontmatter conventions
used by the call/core/vault crates.
Tasks span 7 phases across 5 module subdirectories (server/, gateway/,
client/, adapters/, websocket/):
- Phase 0: crate-init (foundation)
- Phase 1: gateway-dispatch-spine, error-mapping, shared-http-client
(shared infrastructure)
- Phase 2: http-adapter, bearer-auth-middleware, gateway-endpoints,
healthz-decoy (HTTP server surface)
- Phase 3: to-openapi (OpenAPI gateway projection)
- Phase 4: from-openapi (OpenAPI adapter, reqwest forwarding)
- Phase 5: dispatcher-transport-abstraction, upgrade-handler,
connection-overlay (WebSocket browser bidirectional path)
- Phase 6: from-mcp, to-mcp (MCP adapters, feature-gated)
- Phase 7: review-http, review-websocket, review-mcp, review-http-final
(quality checkpoints)
The gateway-dispatch-spine task implements the thin shared core
recommended by the gateway-factoring research (concrete struct, not a
trait). The dispatcher-transport-abstraction task is a cross-crate
change to alknet-call (exposes EventEnvelope-level dispatch API for
non-QUIC transports) — the highest-risk task. WebTransport/h3 is
deferred per ADR-044 and has no tasks; from_wss is out of scope.
Validated: 19 tasks, no cycles, 8 parallel generations, critical path
length 8 (through the WebSocket strand).
Three issues found in the http crate spec sanity check that would have
caused problems during task decomposition, now fixed:
C1 — /subscribe GET→POST: the gateway's /subscribe is an invoke endpoint
carrying { operation, input } in the body, but was listed as GET (which
has no body). Flipped to POST with Accept: text/event-stream negotiating
the SSE response, consistent with /call's flat-JSON-body invariant.
Browsers using EventSource can't POST but use WebSocket for the
bidirectional path; the HTTP gateway's /subscribe is for non-browser
HTTP clients (fetch + ReadableStream). Touches ADR-042, ADR-047,
ADR-048, http-adapters.md, http-server.md.
C2 — stale direct-call references: three spots contradicted ADR-047
(which removed the POST /{service}/{op} direct-call surface) and
ADR-046 §3 (which states /{service}/{op} is no longer reserved).
Cleaned up in http-server.md (custom-routes intro + collision list) and
ADR-046 §6 (default-surface list).
W2 — from_mcp output handling: the spec's fallback for tools without
outputSchema was Type.Unknown(), but the correct fallback is the MCP
ContentBlock union (text|image|audio|resource|resource_link) — a
well-defined MCP type, not Unknown. Fixed http-mcp.md with the full
structuredContent-preferred-over-content-blocks logic (matching the TS
adapter and rmcp SDK), enriched references with specific rmcp source
files. Also added shared-dispatch-spine notes to http-mcp.md and
http-adapters.md cross-referencing the new research findings.
Research (docs/research/alknet-http-gateway-factoring/findings.md):
to_mcp and to_openapi share a dispatch spine (resolve → invoke → map).
Recommendation: extract a thin shared struct now, not a GatewayDispatch
trait — the server-integration layers (axum routes vs rmcp
StreamableHttpService) and wire-framing stay per-gateway. A third
gateway is not on the horizon; if one appears its server-integration
needs its own shape anyway.
Minor: WS route precedence note (websocket.md), OpenAPISpec
shared-type-not-shape clarification (http-adapters.md), date bumps.
Document the two codebases that inform the ShaderGenerator's op table
and the wgpu+handlebars+remote-GPU patterns:
- wonnx (MIT/Apache-2.0, archived): comprehensive ONNX op set in
Tera-templated WGSL at wonnx/templates/ — arithmetic, activation,
gemm, conv, batchnorm, softmax, etc. Port the shader implementations,
swap Tera for handlebars. compiler.rs's add_raw_template +
include_str! pattern maps 1:1 to handlebars-rs register_template_string.
- Handlebars + wgpu + remote-GPU patterns (private reference, patterns
reusable): validates the handlebars-rs side and the vast.ai deployment
shape. Patterns carried over: {{> partial}} includes for shared
fragments, inline-able constant tables via switch statements (SHA-256
k-values, universal across wgpu versions), default-valued template
parameters, wgpu-on-remote-GPU sync. sha256 as a base shader
demonstrating non-ML compute on the same dispatch surface.
Updated the WGSL codegen probe POC to reference wonnx's op set as the
porting source.
Extract the shared JS+wgpu substrate (verified by the alknet-desktop POCs)
as alknet-runtime — the generalized QuickJS-NG + wgpu runtime that both
alknet-desktop (render) and alknet-compute (tensor compute) build on. Key
property driving the split: wgpu on llvmpipe is genuinely useful compute
with no physical GPU (WGSL → optimized SIMD beats JS for non-trivial
workloads), so wgpu is unconditional in the runtime rather than a feature
flag.
Reframes the original alknet-tensor architecture-summary as alknet-compute
(builds on alknet-runtime + alknet-tensor) with ShaderGenerator as a trait
(WGSL first impl, SPIR-V/GLSL/naga-IR later per wgpu multi-input-language
support). alknet-tensor/metatensor-format.md is now clearly the pure binary
format crate (no JS or wgpu dep), usable standalone by a pure-Rust model
server.
Layering: alknet-runtime depends on alknet-call (registry authority stays
per ADR-013); alknet-compute and alknet-desktop depend on alknet-runtime;
alknet-tensor is a pure-format sibling.
Promote the WebSocket browser path from a section in http-server.md to a
first-class spec (websocket.md) and commit the contract-pattern decision
(ADR-048): a WS connection carries the native EventEnvelope call-protocol
session, not the HTTP gateway shape. The gateway endpoints are HTTP-only;
discovery on WS is via services/list/services/schema as ordinary call-protocol
ops; subscriptions project as native call.responded events (no SSE).
ADR-044 already decided WS as the v1 browser bidirectional path; ADR-048
clarifies the shape of what ADR-044 committed (§1 implies native session;
the ADR makes it an explicit implementer-visible rule). The from_wss adapter
(importing a remote node's ops over WS) is recorded as out-of-scope with a
concrete reversal trigger so it is not re-derived later.
Spec cleanup: http-server.md WS section collapsed to a stub pointer;
websocket.md Why section references ADRs rather than re-arguing them;
length-prefix decision made canonical (no prefix on WS — message boundary
is the delimiter); default upgrade path pinned (/alknet/call) with HTTP/2
extended CONNECT noted; indexes (README, http/README, overview) updated.
OQ-39 (to_openapi published-spec versioning) resolved by ADR-045:
info.version semver tracks the gateway endpoint contract, not the
operation set — per-caller operations discovered via /search do not
bump the version. The gateway pattern (ADR-042) dissolved most of the
original churn concern.
ADR-046: assembly-layer custom HTTP routes on HttpAdapter. The HTTP
router had no documented extension point for deployment-specific
endpoints (e.g., an OAI-compatible proxy at /v1/chat/completions). Adds
extra_routes: Option<Router> at construction; raw HTTP, not operations;
default surface takes precedence on collision. The mechanism is the
one-way door; specific routes are two-way.
ADR-047: remove the direct-call POST /{service}/{op} HTTP surface. The
gateway /call is the sole invoke path — the simplified contract is a
few fixed endpoints, not a per-operation REST tree. The direct-call
surface re-introduced the 'dump the full API regardless of privs'
failure mode at the HTTP level that the gateway /search was built to
escape. ADR-036's routing decision is superseded; its non-routing
clauses (SSE, Bearer auth, /healthz, stealth, error mapping) survive.
A deployment wanting a REST-like per-operation surface builds it as a
custom route projection (ADR-046).
ADR-044 updated with the tradeoff framing (WSS is the right tool for
the call-protocol-from-browser case; WebTransport is the right tool for
the generalized ALPN-stream-proxy case we don't have yet — coexist, not
migrate) and the @alkdev/pubsub concrete prior art (the EventEnvelope
{type,id,payload} the call protocol was derived from already has a
working WebSocket client/server; the sync is a small adjustment, not a
from-scratch build).
call-protocol.md references the pubsub lineage for the
transport-agnosticism claim.
OQ-40 resolved: alknet-http owns a shared reqwest_middleware::ClientWithMiddleware
(not a bare reqwest::Client) with a two-layer middleware stack —
RetryTransientMiddleware (reqwest-retry, exponential backoff on transient
failures) + inlined RetryAfterMiddleware (from melotic/reqwest-retry-after, MIT,
~50 lines, inlined to bound the upstream's unbounded HashMap storage). The two
are complementary: reqwest-retry's default strategy does not honor Retry-After.
Hot-reload is rebuild-and-swap via ArcSwap (same pattern as
ConfigIdentityProvider, ADR-035); a rebuild drops the connection pool, which
is acceptable since a config change wanting a fresh pool is the trigger. The
three one-way constraints stand unchanged: alknet-http owns its client (no
env-var config, no shared global), credentials inject per-request from
OperationContext.capabilities, outbound TLS uses the system trust store.
Records the downstream layering boundary: the agent crate's provider SSE
normalization (the solid part of aisdk's pattern — Vercel-UI-message
normalization) sits on top of this client, consuming the reqwest::Response
stream; it does not replace the client. The aisdk core/client.rs reference for
client construction is dropped (env-var config + hand-rolled retry are the
anti-patterns discarded); the from_openapi.ts SSE normalization reference in
the forwarding-handler section is kept (separate, solid pattern).
No ADR — the decision is internal to alknet-http: the client type does not
cross crate boundaries (alknet-call never sees reqwest), the library choice is
reversible, and it does not touch the system's structure, constraints, or
cross-crate API surface.
Updates: http-adapters.md (HTTP client section rewritten, references updated,
constraints/OQ bullets updated), http-mcp.md (OQ-40 status flip), open-
questions.md (OQ-40 resolved with full config-shape table), README.md (OQ-40
folded into the existing two-way-doors bucket), and three secondary docs
(crates/http/README.md, overview.md, http-server.md) that carried stale 'open'
OQ-40 references.
