hermes-agent/gateway/relay/adapter.py
Ben Barclay 6e4e5967f7
feat(relay): multi-platform-per-agent — list identity, provision-loop, N-hello, per-frame egress (Phase 1.5) (#52830)
Cut over the agent half of Shape A (D-Q1.5a/b.1/c) to front a SET of platforms on
one relay WS:

- relay_platform_identities() parses GATEWAY_RELAY_PLATFORMS (list) +
  GATEWAY_RELAY_BOT_IDS (JSON keyed map {platform:{botId,username?}}). Cut over
  from the scalar GATEWAY_RELAY_PLATFORM/_BOT_ID (no fallback, D-Q1.5c).
- self_provision_relay() loops one /relay/provision per platform under one
  gatewayId+secret, partial-failure-tolerant.
- WebSocketRelayTransport takes the identity SET, sends one hello per identity
  (connector accumulates the advertised set), and stamps the per-frame
  OutboundFrame.platform + its matching advertised botId on outbound.
- RelayAdapter remembers each chat's underlying source.platform (mirroring the
  existing guild/dm scope capture) and tags the reply's egress platform.
- send_relay_policy() declares one relevance policy per fronted platform (the
  connector keys policy by (tenant,platform,instanceId)).

Single-platform deploys are byte-identical on the wire (1-element list, no per-frame
tag -> connector session-default fallback). typecheck/ruff clean; relay unit 221 pass
(+10 new); all 15 cross-repo E2E drivers green vs connector origin/main.
2026-06-26 17:32:46 +10:00

546 lines
28 KiB
Python

"""RelayAdapter — one generic gateway adapter fronted by the connector. EXPERIMENTAL.
A single ``BasePlatformAdapter`` subclass that, at handshake, receives a
``CapabilityDescriptor`` from the connector telling it which platform it is
fronting and which capabilities to advertise to the ``GatewayStreamConsumer``.
It implements the four abstract methods (``connect`` / ``disconnect`` / ``send``
/ ``get_chat_info``) plus the capability surface (``MAX_MESSAGE_LENGTH``,
``message_len_fn``, ``supports_draft_streaming``) by delegating wire I/O to an
injected transport and reading capabilities off the descriptor.
There is NO per-platform gateway code: the connector is the only side that knows
"this chat_id maps to a Discord channel, send it via the Discord websocket."
The gateway sees an ordinary ``MessageEvent`` in and calls ``adapter.send`` out.
EXPERIMENTAL: the transport protocol and descriptor schema may change without a
deprecation cycle until >=2 Class-1 platforms validate them.
"""
from __future__ import annotations
import asyncio
import logging
from typing import Any, Callable, Dict, Optional
from gateway.config import Platform, PlatformConfig
from gateway.platforms.base import BasePlatformAdapter, MessageEvent, SendResult
from gateway.relay.descriptor import CapabilityDescriptor
from gateway.relay.transport import RelayTransport
from gateway.session import SessionSource
logger = logging.getLogger(__name__)
def _utf16_len(text: str) -> int:
"""Count UTF-16 code units (Telegram's length unit)."""
return len(text.encode("utf-16-le")) // 2
# Table-driven length-unit selection from the descriptor's ``len_unit``.
_LEN_FNS: Dict[str, Callable[[str], int]] = {
"chars": len,
"utf16": _utf16_len,
}
class RelayAdapter(BasePlatformAdapter):
"""Generic relay adapter advertising a connector-negotiated capability profile."""
def __init__(
self,
config: PlatformConfig,
descriptor: CapabilityDescriptor,
transport: Optional[RelayTransport] = None,
) -> None:
# The relay adapter fronts many platforms but presents as a single
# logical platform to the runner; Platform.RELAY identifies it.
super().__init__(config, Platform.RELAY)
self.descriptor = descriptor
self._transport = transport
# Capability surface read by stream_consumer (getattr(..., 4096)).
self.MAX_MESSAGE_LENGTH = descriptor.max_message_length
# chat_id -> guild_id (Discord) / workspace scope, learned from inbound
# events. The connector's egress guard resolves the owning tenant from
# the OUTBOUND action's metadata.guild_id; the gateway's generic delivery
# path (run.py _thread_metadata_for_source) only carries thread_id, so we
# re-attach the scope here from what we saw inbound. Keyed by chat_id
# (channel) since that's what send() receives. See routedEgressGuard.ts.
self._scope_by_chat: Dict[str, str] = {}
# chat_id -> author user_id for DM channels (no guild_id). A DM reply has
# no guild discriminator, so the connector resolves its tenant from the
# recipient's author binding; we re-attach this user_id as
# metadata.user_id on the outbound action so it can. See _capture_scope.
self._dm_user_by_chat: Dict[str, str] = {}
# chat_id -> the UNDERLYING platform (e.g. "discord", "telegram") this
# chat belongs to (Phase 1.5 multi-platform-per-agent). One relay adapter
# fronts N platforms on one WS; an outbound reply must egress through the
# platform the inbound came from. We remember it per chat_id from the
# inbound event's source.platform and stamp it on the OutboundFrame so the
# connector dispatches to the right sender. Empty for a single-platform
# gateway (the connector falls back to its session default). See
# _capture_scope / send.
self._platform_by_chat: Dict[str, str] = {}
self.supports_code_blocks = descriptor.markdown_dialect not in ("", "plain")
# Phase 7 Unit 7d-B: watches the transport for a terminal auth revocation
# (a 4401 close after a successful handshake = the operator opted this
# instance out of the relay). On revocation we surface a clean,
# non-retryable "relay disabled" fatal so the dashboard stops showing a
# red "retrying" spin against a dead credential.
self._revocation_monitor: Optional[asyncio.Task[None]] = None
# ── capability surface (from descriptor) ─────────────────────────────
@property
def authorization_is_upstream(self) -> bool:
"""Relay authorization is enforced by the connector, not locally.
The connector authenticates this gateway's WS (per-instance secret) and
performs owner-only author-binding resolution before delivering, so any
inbound relay event was already authorized as THIS instance's bound user
(``user_instance_binding``, keyed on the connector-observed author id).
The instance therefore must not default-deny relay users for lack of a
local ``RELAY_ALLOWED_USERS`` env allowlist. See
``BasePlatformAdapter.authorization_is_upstream``.
"""
return True
@property
def message_len_fn(self) -> Callable[[str], int]:
return _LEN_FNS.get(self.descriptor.len_unit, len)
def supports_draft_streaming(
self,
chat_type: Optional[str] = None,
metadata: Optional[Dict[str, Any]] = None,
) -> bool:
return self.descriptor.supports_draft_streaming
# ── abstract methods (delegated to the transport) ────────────────────
async def connect(self, *, is_reconnect: bool = False) -> bool:
# ``is_reconnect`` is part of the BasePlatformAdapter.connect contract:
# the gateway's reconnect watcher (gateway/run.py) re-establishes a
# platform after a fatal adapter error by building a fresh adapter and
# calling ``connect(is_reconnect=True)``. Relay MUST accept the kwarg or
# that recovery path raises TypeError and the relay platform can never
# come back through the watcher.
#
# Relay deliberately IGNORES the flag. The flag exists so adapters with a
# server-side update queue (e.g. Telegram's Bot API) preserve that queue
# across an outage instead of dropping it (#46621). Relay has no such
# gateway-side queue: messages buffered during a gap live in the
# CONNECTOR's durable buffer and are replayed when the transport
# re-handshakes. Routine WS drops are handled entirely by the transport's
# own reconnect supervisor (WebSocketRelayTransport, reconnect=True);
# a watcher-driven reconnect builds a fresh transport from scratch (the
# fatal-error handler disconnect()s the old adapter first, cancelling its
# supervisor), so there is nothing at the adapter layer to preserve.
if self._transport is None:
raise RuntimeError("RelayAdapter has no transport configured")
self._transport.set_inbound_handler(self._on_inbound)
# Inbound interrupts (connector -> owning gateway) arrive as
# interrupt_inbound frames over the SAME outbound WS; bridge them to the
# adapter's interrupt path. WS-only: there is no inbound HTTP receiver.
set_interrupt = getattr(self._transport, "set_interrupt_inbound_handler", None)
if callable(set_interrupt):
set_interrupt(self.on_interrupt)
# Passthrough-plane forwards (Discord interactions, Twilio, …) also ride
# the SAME outbound WS (Phase 5 §5.1) — the connector edge-ACKed and
# forwards the real request here, so a hosted gateway needs no public
# inbound port. Bridge them to the adapter's passthrough handler.
set_passthrough = getattr(self._transport, "set_passthrough_handler", None)
if callable(set_passthrough):
set_passthrough(self._on_passthrough)
ok = await self._transport.connect()
if not ok:
return False
# Negotiate the real capability descriptor from the connector and adopt
# it — the placeholder passed at construction is replaced by what the
# connector advertises for the platform this gateway actually fronts.
try:
descriptor = await self._transport.handshake()
except Exception as exc: # noqa: BLE001 - a failed handshake = a failed connect
logger.warning("relay handshake failed: %s", exc)
return False
self._apply_descriptor(descriptor)
# Inbound (messages + interrupts) is delivered over the outbound WS via
# the connector's relay bus — there is NO inbound HTTP endpoint (hosted
# gateways have no public IP). The transport's reader already dispatches
# `inbound` / `interrupt_inbound` frames to the handlers wired above.
# Phase 7 Unit 7d-B: start watching for a terminal auth revocation
# (opt-out). Only meaningful when the transport exposes `auth_revoked`
# (the production WebSocket transport); the test/stub transports don't.
if hasattr(self._transport, "auth_revoked"):
self._start_revocation_monitor()
return True
def _start_revocation_monitor(self) -> None:
"""Spawn (once) the task that turns a transport auth-revocation into a
clean non-retryable 'relay disabled' fatal. Idempotent."""
if self._revocation_monitor is not None and not self._revocation_monitor.done():
return
try:
self._revocation_monitor = asyncio.create_task(
self._watch_for_revocation(), name="relay-revocation-monitor"
)
except RuntimeError:
# No running loop (e.g. a unit test calling connect() synchronously
# via a stub) — nothing to monitor.
self._revocation_monitor = None
async def _watch_for_revocation(self, poll_interval_s: float = 1.0) -> None:
"""Poll the transport for a terminal 4401 revocation (opt-out). On
revocation, surface a non-retryable `relay_disabled` fatal so the
dashboard renders a clean 'Relay disabled' state instead of a red
'retrying' spin, and notify the gateway's fatal-error handler so the
adapter is cleanly removed (it is NOT queued for reconnection, because
the credential is dead until the instance is recreated)."""
transport = self._transport
try:
while True:
if transport is None or getattr(transport, "auth_revoked", False):
break
await asyncio.sleep(poll_interval_s)
except asyncio.CancelledError:
raise
if transport is None or not getattr(transport, "auth_revoked", False):
return
logger.warning(
"relay credential revoked (opt-out) — marking the relay adapter disabled"
)
# Non-retryable: a revoked secret never comes back without a recreate, so
# _handle_adapter_fatal_error must NOT queue it for reconnection.
self._set_fatal_error(
"relay_disabled",
"Relay disabled (opted out — recreate the instance to re-enable)",
retryable=False,
)
try:
await self._notify_fatal_error()
except Exception: # noqa: BLE001 - notification is best-effort
logger.debug("relay revocation fatal-error notify failed", exc_info=True)
def _apply_descriptor(self, descriptor: CapabilityDescriptor) -> None:
"""Adopt a (re)negotiated descriptor into the live capability surface."""
self.descriptor = descriptor
self.MAX_MESSAGE_LENGTH = descriptor.max_message_length
self.supports_code_blocks = descriptor.markdown_dialect not in ("", "plain")
async def _on_inbound(self, event) -> None:
"""Bridge a connector-delivered MessageEvent into the normal adapter path."""
self._capture_scope(event)
await self.handle_message(event)
def _capture_scope(self, event) -> None:
"""Remember a chat_id's egress discriminator from an inbound event so our
outbound (the agent's reply) can re-assert it for the connector's egress
tenant resolution. Never raises — scope tracking must not break inbound.
Two cases, matching the connector's two tenant-resolution paths:
- GUILD message: remember chat_id -> guild_id. The connector resolves
the tenant from metadata.guild_id (routing table).
- DM (no guild_id): remember chat_id -> the authentic author user_id.
A DM carries no guild discriminator, so the connector instead resolves
the tenant from the recipient's author binding (resolveByUser); it
needs the user_id on the OUTBOUND action to do that. Without this, a
DM reply has no resolvable discriminator and the connector's egress
guard declines it as "target not routed to an onboarded tenant".
See gateway-gateway routedEgressGuard.ts / discordTenantOf.
"""
try:
src = getattr(event, "source", None)
if not src:
return
chat = getattr(src, "chat_id", None)
if not chat:
return
# Phase 1.5: remember the underlying platform for this chat so the
# reply egresses through the right sender (one relay adapter fronts N
# platforms). source.platform is a Platform enum (e.g. Platform.DISCORD,
# mapped from the connector's "discord" by ws_transport _frame_to_event);
# record its string VALUE, skipping the generic RELAY fallback (a
# single-platform connector that didn't tag a concrete platform — the
# connector's session default handles egress then).
platform = getattr(src, "platform", None)
platform_value = getattr(platform, "value", platform)
if platform_value and platform_value != "relay":
self._platform_by_chat[str(chat)] = str(platform_value)
guild = getattr(src, "guild_id", None)
if guild:
self._scope_by_chat[str(chat)] = str(guild)
return
# DM: no guild_id. Remember the authentic author id for outbound
# author-binding resolution (the user we're replying to in this DM).
user_id = getattr(src, "user_id", None)
if user_id:
self._dm_user_by_chat[str(chat)] = str(user_id)
except Exception: # noqa: BLE001 - scope tracking must never break inbound
pass
def _with_scope(self, chat_id: str, metadata: Optional[Dict[str, Any]]) -> Dict[str, Any]:
"""Ensure the outbound metadata carries the discriminator the connector's
egress guard needs to resolve the owning tenant. Two cases:
- GUILD reply: re-attach metadata.guild_id (routing-table resolution).
- DM reply: there is no guild_id, so re-attach metadata.user_id — the
authentic author id we saw inbound — which the connector resolves to
the tenant via the recipient's author binding (resolveByUser). Without
one of these, egress is declined as 'target not routed to an onboarded
tenant'. See gateway-gateway routedEgressGuard.ts / discordTenantOf.
No-op when the relevant value is already present or unknown for this chat.
"""
meta: Dict[str, Any] = dict(metadata or {})
if not meta.get("guild_id"):
scope = self._scope_by_chat.get(str(chat_id))
if scope:
meta["guild_id"] = scope
# DM author-binding discriminator. Only meaningful when there's no guild
# (a guild reply resolves by guild_id); harmless to carry otherwise, but
# we only set it when this chat is a known DM and the field is absent.
if not meta.get("guild_id") and not meta.get("user_id"):
dm_user = self._dm_user_by_chat.get(str(chat_id))
if dm_user:
meta["user_id"] = dm_user
return meta
def _platform_is_fronted(self, platform: str) -> bool:
"""Whether ``platform`` is one of the platforms this gateway fronts over
the relay (Phase 1.5). Reads the transport's advertised identity set; used
to decide whether a follow-up's platform-prefixed `kind` names a real
fronted platform worth tagging on the frame (vs. leaving egress to the
session default). Safe when the transport is absent or single-identity."""
ids = getattr(self._transport, "_identities", None)
if not ids:
return False
return any(p == platform for p, _ in ids)
async def on_interrupt(self, session_key: str, chat_id: str) -> None:
"""Bridge a connector-delivered /stop into the adapter's interrupt path.
The connector forwards a mid-turn interrupt down the socket owned by
the gateway instance running ``session_key``; this routes it to the
existing per-session interrupt mechanism (sets the
``_active_sessions[session_key]`` Event and clears typing), cancelling
the right turn without touching sibling sessions.
"""
await self.interrupt_session_activity(session_key, chat_id)
async def _on_passthrough(self, forward, buffer_id: Optional[str] = None) -> None:
"""Handle a connector-forwarded passthrough request (Phase 5 §5.1).
The passthrough plane (Discord interactions, Twilio webhooks, …) answers
the provider's latency-critical ACK at the connector EDGE, then forwards
the real, ALREADY-SANITIZED request to this gateway over the outbound WS.
The connector is the trust boundary: it verified the provider signature
at the edge and stripped any shared-identity credential (e.g. a Discord
interaction follow-up token) into its vault — so this body carries no
token, and the agent later acts on it via the token-less ``follow_up``
path (``send_follow_up``), never holding the credential.
For a Discord interaction we decode the (JSON) body and convert it to a
normalized ``MessageEvent`` so it flows through the SAME agent path as a
chat message (``handle_message``); the agent's reply egresses over the
normal outbound/follow_up path. Non-JSON or non-interaction forwards are
logged and dropped for now (Twilio/SMS over the relay is a later unit).
NEVER raises: a malformed forward must not kill the read loop.
NOTE (open semantic sub-design, flagged for review): the interaction ->
MessageEvent mapping below is the v1 default. The exact agent UX for a
slash-command / button interaction (vs. a plain message) — command name
surfacing, option rendering, deferred-vs-immediate response — is the open
piece tracked in the spec; the TRANSPORT + receive mechanism (this whole
path) is settled.
"""
try:
platform = getattr(forward, "platform", "") or ""
if platform == "discord":
event = self._discord_interaction_to_event(forward)
if event is not None:
self._capture_scope(event)
await self.handle_message(event)
return
logger.info(
"relay passthrough_forward dropped (no handler): platform=%s method=%s path=%s",
platform,
getattr(forward, "method", "?"),
getattr(forward, "path", "?"),
)
except Exception: # noqa: BLE001 - a bad forward must never break the reader
logger.warning("relay passthrough_forward handling failed", exc_info=True)
def _discord_interaction_to_event(self, forward):
"""Convert a forwarded Discord interaction body to a MessageEvent, or None.
Builds the session source the same way the connector does for an
interaction (``interactionSessionSource`` on the connector side), so the
agent's session key matches the one the connector bound the follow-up
capability under. Returns None when the body isn't a usable interaction
(e.g. a PING, which the connector already answers at the edge and never
forwards).
"""
import json
from gateway.platforms.base import MessageType
try:
payload = json.loads(bytes(getattr(forward, "body", b"")).decode("utf-8"))
except Exception: # noqa: BLE001
return None
if not isinstance(payload, dict):
return None
# type 1 = PING (answered at the edge, never forwarded); 2 = APPLICATION_COMMAND;
# 3 = MESSAGE_COMPONENT; 5 = MODAL_SUBMIT. Surface a best-effort text.
itype = payload.get("type")
data = payload.get("data") or {}
if itype == 2:
text = str(data.get("name") or "")
elif itype == 3:
text = str(data.get("custom_id") or "")
else:
text = ""
member = payload.get("member") or {}
user = (member.get("user") if isinstance(member, dict) else None) or payload.get("user") or {}
channel_id = str(payload.get("channel_id") or "")
guild_id = payload.get("guild_id")
source = SessionSource(
platform=Platform.RELAY,
chat_id=channel_id,
chat_type="channel" if guild_id else "dm",
user_id=str(user.get("id")) if isinstance(user, dict) and user.get("id") else None,
user_name=str(user.get("username")) if isinstance(user, dict) and user.get("username") else None,
guild_id=str(guild_id) if guild_id else None,
message_id=str(payload.get("id")) if payload.get("id") else None,
)
return MessageEvent(text=text, message_type=MessageType.TEXT, source=source)
async def disconnect(self) -> None:
# Phase 7 Unit 7d-B: stop the revocation monitor first so it can't fire a
# spurious fatal during/after a deliberate teardown.
if self._revocation_monitor is not None:
self._revocation_monitor.cancel()
try:
await self._revocation_monitor
except (asyncio.CancelledError, Exception): # noqa: BLE001 - best-effort teardown
pass
self._revocation_monitor = None
if self._transport is not None:
# Phase 5 §5.3: emit going_idle as part of the gateway's EXISTING
# drain/shutdown transition (the runner calls adapter.disconnect()
# when the gateway enters `draining`). Asking the connector to flip
# this instance to buffered-only BEFORE we tear down the socket means
# inbound that arrives while we're asleep buffers durably and replays
# on reconnect, instead of being pushed at a closing socket. The
# connector is authoritative (it acks the flip); we stay serving until
# the ack (Q-5.3c). Best-effort + guarded: a transport without go_idle
# (the stub) or a failed/timed-out ack must not block shutdown — we
# proceed to disconnect exactly as before, no regression.
go_idle = getattr(self._transport, "go_idle", None)
if callable(go_idle):
try:
result: Any = go_idle()
if asyncio.iscoroutine(result):
await result
except Exception: # noqa: BLE001 - going-idle is an optimization, never blocks drain
logger.debug("relay going_idle failed during drain", exc_info=True)
await self._transport.disconnect()
async def go_dormant(self) -> bool:
"""Quiesce the relay for a scale-to-zero suspend (D12 / Phase 0).
Unlike ``disconnect()`` (terminal teardown for shutdown/restart), this
keeps the adapter's reconnect path armed so the gateway re-dials and
drains its buffered backlog when the machine wakes. Delegates to the
transport's ``go_dormant()`` when available; a transport without it (the
stub) is a no-op that returns False, so callers degrade safely.
NOTE: deliberately does NOT stop the revocation monitor — going dormant
is not a teardown; the monitor stays live so a real opt-out/revocation
during dormancy is still surfaced on wake.
"""
if self._transport is None:
return False
go_dormant = getattr(self._transport, "go_dormant", None)
if not callable(go_dormant):
return False
try:
result: Any = go_dormant()
if asyncio.iscoroutine(result):
return bool(await result)
return bool(result)
except Exception: # noqa: BLE001 - dormancy is best-effort, never blocks the idle path
logger.debug("relay go_dormant failed", exc_info=True)
return False
async def send(
self,
chat_id: str,
content: str,
reply_to: Optional[str] = None,
metadata: Optional[Dict[str, Any]] = None,
) -> SendResult:
if self._transport is None:
return SendResult(success=False, error="no transport")
result = await self._transport.send_outbound(
{
"op": "send",
"chat_id": chat_id,
"content": content,
"reply_to": reply_to,
"metadata": self._with_scope(chat_id, metadata),
},
platform=self._platform_by_chat.get(str(chat_id)),
)
return SendResult(
success=bool(result.get("success")),
message_id=result.get("message_id"),
error=result.get("error"),
)
async def get_chat_info(self, chat_id: str) -> Dict[str, Any]:
# Proxied to the connector (it owns the platform connection / cache).
if self._transport is None:
return {"name": chat_id, "type": "dm"}
return await self._transport.get_chat_info(chat_id)
async def send_follow_up(
self,
session_key: str,
kind: str,
content: str,
metadata: Optional[Dict[str, Any]] = None,
) -> SendResult:
"""Send via a shared-identity capability bound to a session (A2 outbound).
The gateway never holds the credential: it names the session it is
already in plus the capability ``kind``, and the connector resolves the
real value from its vault and egresses (enforcing the tenant match). Used
e.g. to post a Discord interaction follow-up as the shared bot without
the token ever reaching the gateway. See RelayTransport.send_follow_up.
"""
if self._transport is None:
return SendResult(success=False, error="no transport")
# Phase 1.5: the capability `kind` is platform-prefixed (e.g.
# "discord.interaction_token"), so derive the egress platform from it when
# it names one we front — that tags the OutboundFrame so a multi-platform
# gateway routes the follow-up through the right sender. Falls back to the
# session default (connector-side) when the prefix isn't a fronted platform.
follow_up_platform = None
if kind and "." in kind:
prefix = kind.split(".", 1)[0]
if self._platform_is_fronted(prefix):
follow_up_platform = prefix
result = await self._transport.send_follow_up(
{
"op": "follow_up",
"session_key": session_key,
"kind": kind,
"content": content,
"metadata": metadata or {},
},
platform=follow_up_platform,
)
return SendResult(
success=bool(result.get("success")),
message_id=result.get("message_id"),
error=result.get("error"),
)