decnet/web/db/models.py

@@ -373,3 +373,72 @@ class SwarmHostHealth(BaseModel):
 
 class SwarmCheckResponse(BaseModel):
     results: list[SwarmHostHealth]
+
+
+# --- Remote Updates (master → worker /updater) DTOs ---
+# Powers the dashboard's Remote Updates page. The master dashboard calls
+# these (auth-gated) endpoints; internally they fan out to each worker's
+# updater daemon over mTLS via UpdaterClient.
+
+class HostReleaseInfo(BaseModel):
+    host_uuid: str
+    host_name: str
+    address: str
+    reachable: bool
+    # These fields mirror the updater's /health payload when reachable; they
+    # are all Optional so an unreachable host still serializes cleanly.
+    agent_status: Optional[str] = None
+    current_sha: Optional[str] = None
+    previous_sha: Optional[str] = None
+    releases: list[dict[str, Any]] = PydanticField(default_factory=list)
+    detail: Optional[str] = None  # populated when unreachable
+
+
+class HostReleasesResponse(BaseModel):
+    hosts: list[HostReleaseInfo]
+
+
+class PushUpdateRequest(BaseModel):
+    host_uuids: Optional[list[str]] = PydanticField(
+        default=None,
+        description="Target specific hosts; mutually exclusive with 'all'.",
+    )
+    all: bool = PydanticField(default=False, description="Target every non-decommissioned host with an updater bundle.")
+    include_self: bool = PydanticField(
+        default=False,
+        description="After a successful /update, also push /update-self to upgrade the updater itself.",
+    )
+    exclude: list[str] = PydanticField(
+        default_factory=list,
+        description="Additional tarball exclude globs (on top of the built-in defaults).",
+    )
+
+
+class PushUpdateResult(BaseModel):
+    host_uuid: str
+    host_name: str
+    # updated = /update 200. rolled-back = /update 409 (auto-recovered).
+    # failed = transport error or non-200/409 response. self-updated = /update-self succeeded.
+    status: Literal["updated", "rolled-back", "failed", "self-updated", "self-failed"]
+    http_status: Optional[int] = None
+    sha: Optional[str] = None
+    detail: Optional[str] = None
+    stderr: Optional[str] = None
+
+
+class PushUpdateResponse(BaseModel):
+    sha: str
+    tarball_bytes: int
+    results: list[PushUpdateResult]
+
+
+class RollbackRequest(BaseModel):
+    host_uuid: str = PydanticField(..., description="Host to roll back to its previous release slot.")
+
+
+class RollbackResponse(BaseModel):
+    host_uuid: str
+    host_name: str
+    status: Literal["rolled-back", "failed"]
+    http_status: Optional[int] = None
+    detail: Optional[str] = None

decnet/web/router/__init__.py

@@ -21,6 +21,7 @@ from .config.api_manage_users import router as config_users_router
 from .config.api_reinit import router as config_reinit_router
 from .health.api_get_health import router as health_router
 from .artifacts.api_get_artifact import router as artifacts_router
+from .swarm_updates import swarm_updates_router
 
 api_router = APIRouter()
 
@@ -60,3 +61,6 @@ api_router.include_router(config_reinit_router)
 
 # Artifacts (captured attacker file drops)
 api_router.include_router(artifacts_router)
+
+# Remote Updates (dashboard → worker updater daemons)
+api_router.include_router(swarm_updates_router)

decnet/web/router/swarm_updates/__init__.py

@@ -0,0 +1,23 @@
+"""Remote Updates — master dashboard's surface for pushing code to workers.
+
+These are *not* the swarm-controller's /swarm routes (those run on a
+separate process, auth-free, internal-only). They live on the main web
+API, go through ``require_admin``, and are the interface the React
+dashboard calls to fan updates out to worker ``decnet updater`` daemons
+via ``UpdaterClient``.
+
+Mounted under ``/api/v1/swarm-updates`` by the main api router.
+"""
+from fastapi import APIRouter
+
+from .api_list_host_releases import router as list_host_releases_router
+from .api_push_update import router as push_update_router
+from .api_push_update_self import router as push_update_self_router
+from .api_rollback_host import router as rollback_host_router
+
+swarm_updates_router = APIRouter(prefix="/swarm-updates")
+
+swarm_updates_router.include_router(list_host_releases_router)
+swarm_updates_router.include_router(push_update_router)
+swarm_updates_router.include_router(push_update_self_router)
+swarm_updates_router.include_router(rollback_host_router)

decnet/web/router/swarm_updates/api_list_host_releases.py

@@ -0,0 +1,82 @@
+"""GET /swarm-updates/hosts — per-host updater health + release slots.
+
+Fans out an ``UpdaterClient.health()`` probe to every enrolled host that
+has an updater bundle. Each probe is isolated: a single unreachable host
+never fails the whole list (that's normal partial-failure behaviour for
+a fleet view).
+"""
+from __future__ import annotations
+
+import asyncio
+from typing import Any
+
+from fastapi import APIRouter, Depends
+
+from decnet.logging import get_logger
+from decnet.swarm.updater_client import UpdaterClient
+from decnet.web.db.models import HostReleaseInfo, HostReleasesResponse
+from decnet.web.db.repository import BaseRepository
+from decnet.web.dependencies import get_repo, require_admin
+
+log = get_logger("swarm_updates.list")
+
+router = APIRouter()
+
+
+def _extract_shas(releases: list[dict[str, Any]]) -> tuple[str | None, str | None]:
+    """Pick the (current, previous) SHA from the updater's releases list.
+
+    The updater reports releases as ``[{"slot": "active"|"prev", "sha": ...,
+    ...}]`` in no guaranteed order, so pull by slot name rather than index.
+    """
+    current = next((r.get("sha") for r in releases if r.get("slot") == "active"), None)
+    previous = next((r.get("sha") for r in releases if r.get("slot") == "prev"), None)
+    return current, previous
+
+
+async def _probe_host(host: dict[str, Any]) -> HostReleaseInfo:
+    try:
+        async with UpdaterClient(host=host) as u:
+            body = await u.health()
+    except Exception as exc:  # noqa: BLE001
+        return HostReleaseInfo(
+            host_uuid=host["uuid"],
+            host_name=host["name"],
+            address=host["address"],
+            reachable=False,
+            detail=f"{type(exc).__name__}: {exc}",
+        )
+    releases = body.get("releases") or []
+    current, previous = _extract_shas(releases)
+    return HostReleaseInfo(
+        host_uuid=host["uuid"],
+        host_name=host["name"],
+        address=host["address"],
+        reachable=True,
+        agent_status=body.get("agent_status") or body.get("status"),
+        current_sha=current,
+        previous_sha=previous,
+        releases=releases,
+    )
+
+
+@router.get(
+    "/hosts",
+    response_model=HostReleasesResponse,
+    tags=["Swarm Updates"],
+)
+async def api_list_host_releases(
+    admin: dict = Depends(require_admin),
+    repo: BaseRepository = Depends(get_repo),
+) -> HostReleasesResponse:
+    rows = await repo.list_swarm_hosts()
+    # Only hosts actually capable of receiving updates — decommissioned
+    # hosts and agent-only enrollments are filtered out.
+    targets = [
+        r for r in rows
+        if r.get("status") != "decommissioned" and r.get("updater_cert_fingerprint")
+    ]
+    if not targets:
+        return HostReleasesResponse(hosts=[])
+    results = await asyncio.gather(*(_probe_host(h) for h in targets))
+    return HostReleasesResponse(hosts=list(results))

decnet/web/router/swarm_updates/api_push_update.py

@@ -0,0 +1,152 @@
+"""POST /swarm-updates/push — fan a tarball of the master's tree to workers.
+
+Mirrors the ``decnet swarm update`` CLI flow: build the tarball once,
+dispatch concurrently, collect per-host statuses. Returns HTTP 200 even
+when individual hosts failed — the operator reads per-host ``status``.
+"""
+from __future__ import annotations
+
+import asyncio
+import pathlib
+from typing import Any
+
+from fastapi import APIRouter, Depends, HTTPException
+
+from decnet.logging import get_logger
+from decnet.swarm.tar_tree import detect_git_sha, tar_working_tree
+from decnet.swarm.updater_client import UpdaterClient
+from decnet.web.db.models import PushUpdateRequest, PushUpdateResponse, PushUpdateResult
+from decnet.web.db.repository import BaseRepository
+from decnet.web.dependencies import get_repo, require_admin
+
+log = get_logger("swarm_updates.push")
+
+router = APIRouter()
+
+
+def _master_tree_root() -> pathlib.Path:
+    """Resolve the master's install tree to tar.
+
+    Walks up from this file: ``decnet/web/router/swarm_updates/`` → 3 parents
+    lands on the repo root. Matches the layout shipped via ``pip install -e .``
+    and the dev checkout at ``~/Tools/DECNET``.
+    """
+    return pathlib.Path(__file__).resolve().parents[4]
+
+
+def _classify_update(status_code: int) -> str:
+    if status_code == 200:
+        return "updated"
+    if status_code == 409:
+        return "rolled-back"
+    return "failed"
+
+
+async def _resolve_targets(
+    repo: BaseRepository,
+    req: PushUpdateRequest,
+) -> list[dict[str, Any]]:
+    if req.all == bool(req.host_uuids):
+        raise HTTPException(
+            status_code=400,
+            detail="Specify exactly one of host_uuids or all=true.",
+        )
+    rows = await repo.list_swarm_hosts()
+    rows = [r for r in rows if r.get("updater_cert_fingerprint")]
+    if req.all:
+        targets = [r for r in rows if r.get("status") != "decommissioned"]
+    else:
+        wanted = set(req.host_uuids or [])
+        targets = [r for r in rows if r["uuid"] in wanted]
+        missing = wanted - {r["uuid"] for r in targets}
+        if missing:
+            raise HTTPException(
+                status_code=404,
+                detail=f"Unknown or updater-less host(s): {sorted(missing)}",
+            )
+    if not targets:
+        raise HTTPException(
+            status_code=404,
+            detail="No targets: no enrolled hosts have an updater bundle.",
+        )
+    return targets
+
+
+async def _push_one(
+    host: dict[str, Any],
+    tarball: bytes,
+    sha: str,
+    include_self: bool,
+) -> PushUpdateResult:
+    try:
+        async with UpdaterClient(host=host) as u:
+            r = await u.update(tarball, sha=sha)
+            body = r.json() if r.content else {}
+            status = _classify_update(r.status_code)
+            stderr = body.get("stderr") if isinstance(body, dict) else None
+
+            if include_self and r.status_code == 200:
+                # Agent first, updater second — a broken updater push must never
+                # strand the fleet on an old agent.
+                try:
+                    rs = await u.update_self(tarball, sha=sha)
+                    self_ok = rs.status_code in (200, 0)  # 0 = connection dropped (expected)
+                except Exception as exc:  # noqa: BLE001
+                    # Connection drop on update-self is expected and not an error.
+                    self_ok = _is_expected_connection_drop(exc)
+                    if not self_ok:
+                        return PushUpdateResult(
+                            host_uuid=host["uuid"], host_name=host["name"],
+                            status="self-failed", http_status=r.status_code, sha=sha,
+                            detail=f"agent updated OK but self-update failed: {exc}",
+                            stderr=stderr,
+                        )
+                status = "self-updated" if self_ok else "self-failed"
+
+            return PushUpdateResult(
+                host_uuid=host["uuid"], host_name=host["name"],
+                status=status, http_status=r.status_code, sha=sha,
+                detail=body.get("error") or body.get("probe") if isinstance(body, dict) else None,
+                stderr=stderr,
+            )
+    except Exception as exc:  # noqa: BLE001
+        log.exception("swarm_updates.push failed host=%s", host.get("name"))
+        return PushUpdateResult(
+            host_uuid=host["uuid"], host_name=host["name"],
+            status="failed",
+            detail=f"{type(exc).__name__}: {exc}",
+        )
+
+
+def _is_expected_connection_drop(exc: BaseException) -> bool:
+    """update-self re-execs the updater mid-response; httpx raises on the drop."""
+    import httpx
+    return isinstance(exc, (httpx.RemoteProtocolError, httpx.ReadError, httpx.ConnectError))
+
+
+@router.post(
+    "/push",
+    response_model=PushUpdateResponse,
+    tags=["Swarm Updates"],
+)
+async def api_push_update(
+    req: PushUpdateRequest,
+    admin: dict = Depends(require_admin),
+    repo: BaseRepository = Depends(get_repo),
+) -> PushUpdateResponse:
+    targets = await _resolve_targets(repo, req)
+    tree_root = _master_tree_root()
+    sha = detect_git_sha(tree_root)
+    tarball = tar_working_tree(tree_root, extra_excludes=req.exclude)
+    log.info(
+        "swarm_updates.push sha=%s tarball=%d hosts=%d include_self=%s",
+        sha or "(not a git repo)", len(tarball), len(targets), req.include_self,
+    )
+    results = await asyncio.gather(
+        *(_push_one(h, tarball, sha, req.include_self) for h in targets)
+    )
+    return PushUpdateResponse(
+        sha=sha,
+        tarball_bytes=len(tarball),
+        results=list(results),
+    )

decnet/web/router/swarm_updates/api_push_update_self.py

@@ -0,0 +1,92 @@
+"""POST /swarm-updates/push-self — push only to workers' /update-self.
+
+Use case: the agent is fine but the updater itself needs an upgrade (e.g.
+a fix to ``executor.py``). Uploading only ``/update-self`` avoids a
+redundant agent restart on healthy workers.
+
+No auto-rollback: the updater re-execs itself on success, so a broken
+push leaves the worker on the old code — verified by polling ``/health``
+after the request returns.
+"""
+from __future__ import annotations
+
+import asyncio
+from typing import Any
+
+from fastapi import APIRouter, Depends
+
+from decnet.logging import get_logger
+from decnet.swarm.tar_tree import detect_git_sha, tar_working_tree
+from decnet.swarm.updater_client import UpdaterClient
+from decnet.web.db.models import PushUpdateRequest, PushUpdateResponse, PushUpdateResult
+from decnet.web.db.repository import BaseRepository
+from decnet.web.dependencies import get_repo, require_admin
+
+from .api_push_update import _is_expected_connection_drop, _master_tree_root, _resolve_targets
+
+log = get_logger("swarm_updates.push_self")
+
+router = APIRouter()
+
+
+async def _push_self_one(host: dict[str, Any], tarball: bytes, sha: str) -> PushUpdateResult:
+    try:
+        async with UpdaterClient(host=host) as u:
+            try:
+                r = await u.update_self(tarball, sha=sha)
+                http_status = r.status_code
+                body = r.json() if r.content else {}
+                ok = http_status == 200
+                detail = (body.get("error") or body.get("probe")) if isinstance(body, dict) else None
+                stderr = body.get("stderr") if isinstance(body, dict) else None
+            except Exception as exc:  # noqa: BLE001
+                # Connection drops during self-update are expected — the updater
+                # re-execs itself mid-response.
+                if _is_expected_connection_drop(exc):
+                    return PushUpdateResult(
+                        host_uuid=host["uuid"], host_name=host["name"],
+                        status="self-updated", sha=sha,
+                        detail="updater re-exec dropped connection (expected)",
+                    )
+                raise
+        return PushUpdateResult(
+            host_uuid=host["uuid"], host_name=host["name"],
+            status="self-updated" if ok else "self-failed",
+            http_status=http_status, sha=sha,
+            detail=detail, stderr=stderr,
+        )
+    except Exception as exc:  # noqa: BLE001
+        log.exception("swarm_updates.push_self failed host=%s", host.get("name"))
+        return PushUpdateResult(
+            host_uuid=host["uuid"], host_name=host["name"],
+            status="self-failed",
+            detail=f"{type(exc).__name__}: {exc}",
+        )
+
+
+@router.post(
+    "/push-self",
+    response_model=PushUpdateResponse,
+    tags=["Swarm Updates"],
+)
+async def api_push_update_self(
+    req: PushUpdateRequest,
+    admin: dict = Depends(require_admin),
+    repo: BaseRepository = Depends(get_repo),
+) -> PushUpdateResponse:
+    targets = await _resolve_targets(repo, req)
+    tree_root = _master_tree_root()
+    sha = detect_git_sha(tree_root)
+    tarball = tar_working_tree(tree_root, extra_excludes=req.exclude)
+    log.info(
+        "swarm_updates.push_self sha=%s tarball=%d hosts=%d",
+        sha or "(not a git repo)", len(tarball), len(targets),
+    )
+    results = await asyncio.gather(
+        *(_push_self_one(h, tarball, sha) for h in targets)
+    )
+    return PushUpdateResponse(
+        sha=sha,
+        tarball_bytes=len(tarball),
+        results=list(results),
+    )

decnet/web/router/swarm_updates/api_rollback_host.py

@@ -0,0 +1,70 @@
+"""POST /swarm-updates/rollback — manual rollback on a single host.
+
+Calls the worker updater's ``/rollback`` which swaps the ``current``
+symlink back to ``releases/prev``. Fails with 404 if the target has no
+previous release slot.
+"""
+from __future__ import annotations
+
+from fastapi import APIRouter, Depends, HTTPException
+
+from decnet.logging import get_logger
+from decnet.swarm.updater_client import UpdaterClient
+from decnet.web.db.models import RollbackRequest, RollbackResponse
+from decnet.web.db.repository import BaseRepository
+from decnet.web.dependencies import get_repo, require_admin
+
+log = get_logger("swarm_updates.rollback")
+
+router = APIRouter()
+
+
+@router.post(
+    "/rollback",
+    response_model=RollbackResponse,
+    tags=["Swarm Updates"],
+)
+async def api_rollback_host(
+    req: RollbackRequest,
+    admin: dict = Depends(require_admin),
+    repo: BaseRepository = Depends(get_repo),
+) -> RollbackResponse:
+    host = await repo.get_swarm_host_by_uuid(req.host_uuid)
+    if host is None:
+        raise HTTPException(status_code=404, detail=f"Unknown host: {req.host_uuid}")
+    if not host.get("updater_cert_fingerprint"):
+        raise HTTPException(
+            status_code=400,
+            detail=f"Host '{host['name']}' has no updater bundle — nothing to roll back.",
+        )
+
+    try:
+        async with UpdaterClient(host=host) as u:
+            r = await u.rollback()
+    except Exception as exc:  # noqa: BLE001
+        log.exception("swarm_updates.rollback transport failure host=%s", host["name"])
+        return RollbackResponse(
+            host_uuid=host["uuid"], host_name=host["name"],
+            status="failed",
+            detail=f"{type(exc).__name__}: {exc}",
+        )
+
+    body = r.json() if r.content else {}
+    if r.status_code == 404:
+        # No previous release — surface as 404 so the UI can render the
+        # "nothing to roll back" state distinctly from a transport error.
+        raise HTTPException(
+            status_code=404,
+            detail=body.get("detail") if isinstance(body, dict) else "No previous release on worker.",
+        )
+    if r.status_code != 200:
+        return RollbackResponse(
+            host_uuid=host["uuid"], host_name=host["name"],
+            status="failed", http_status=r.status_code,
+            detail=(body.get("error") or body.get("detail")) if isinstance(body, dict) else None,
+        )
+    return RollbackResponse(
+        host_uuid=host["uuid"], host_name=host["name"],
+        status="rolled-back", http_status=r.status_code,
+        detail=body.get("status") if isinstance(body, dict) else None,
+    )

tests/api/swarm_updates/conftest.py

@@ -0,0 +1,151 @@
+"""Shared fixtures for /api/v1/swarm-updates tests.
+
+The tests never talk to a real worker — ``UpdaterClient`` is monkeypatched
+to a recording fake. That keeps the tests fast and lets us assert call
+shapes (tarball-once, per-host dispatch, include_self ordering) without
+standing up TLS endpoints.
+"""
+from __future__ import annotations
+
+import uuid as _uuid
+from datetime import datetime, timezone
+from typing import Any
+
+import httpx
+import pytest
+
+from decnet.web.dependencies import repo
+
+
+async def _add_host(
+    name: str,
+    address: str = "10.0.0.1",
+    *,
+    with_updater: bool = True,
+    status: str = "enrolled",
+) -> dict[str, Any]:
+    uuid = str(_uuid.uuid4())
+    await repo.add_swarm_host({
+        "uuid": uuid,
+        "name": name,
+        "address": address,
+        "agent_port": 8765,
+        "status": status,
+        "client_cert_fingerprint": "abc123",
+        "updater_cert_fingerprint": "def456" if with_updater else None,
+        "cert_bundle_path": f"/tmp/{name}",
+        "enrolled_at": datetime.now(timezone.utc),
+        "notes": None,
+    })
+    return {"uuid": uuid, "name": name, "address": address}
+
+
+@pytest.fixture
+def add_host():
+    return _add_host
+
+
+@pytest.fixture
+def fake_updater(monkeypatch):
+    """Install a fake ``UpdaterClient`` + tar builder into every route module.
+
+    The returned ``Fake`` exposes hooks so individual tests decide per-host
+    behaviour: response codes, exceptions, update-self outcomes, etc.
+    """
+
+    class FakeResponse:
+        def __init__(self, status_code: int, body: dict[str, Any] | None = None):
+            self.status_code = status_code
+            self._body = body or {}
+            self.content = b"payload"
+
+        def json(self) -> dict[str, Any]:
+            return self._body
+
+    class FakeUpdaterClient:
+        calls: list[tuple[str, str, dict]] = []  # (host_name, method, kwargs)
+        health_responses: dict[str, dict[str, Any]] = {}
+        update_responses: dict[str, FakeResponse | BaseException] = {}
+        update_self_responses: dict[str, FakeResponse | BaseException] = {}
+        rollback_responses: dict[str, FakeResponse | BaseException] = {}
+
+        def __init__(self, host=None, **_kw):
+            self._name = host["name"] if host else "?"
+
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, *exc):
+            return None
+
+        async def health(self):
+            FakeUpdaterClient.calls.append((self._name, "health", {}))
+            resp = FakeUpdaterClient.health_responses.get(self._name)
+            if isinstance(resp, BaseException):
+                raise resp
+            return resp or {"status": "ok", "releases": []}
+
+        async def update(self, tarball, sha=""):
+            FakeUpdaterClient.calls.append((self._name, "update", {"tarball": tarball, "sha": sha}))
+            resp = FakeUpdaterClient.update_responses.get(self._name, FakeResponse(200, {"probe": "ok"}))
+            if isinstance(resp, BaseException):
+                raise resp
+            return resp
+
+        async def update_self(self, tarball, sha=""):
+            FakeUpdaterClient.calls.append((self._name, "update_self", {"tarball": tarball, "sha": sha}))
+            resp = FakeUpdaterClient.update_self_responses.get(self._name, FakeResponse(200))
+            if isinstance(resp, BaseException):
+                raise resp
+            return resp
+
+        async def rollback(self):
+            FakeUpdaterClient.calls.append((self._name, "rollback", {}))
+            resp = FakeUpdaterClient.rollback_responses.get(self._name, FakeResponse(200, {"status": "rolled back"}))
+            if isinstance(resp, BaseException):
+                raise resp
+            return resp
+
+    # Reset class-level state each test — fixtures are function-scoped but
+    # the class dicts survive otherwise.
+    FakeUpdaterClient.calls = []
+    FakeUpdaterClient.health_responses = {}
+    FakeUpdaterClient.update_responses = {}
+    FakeUpdaterClient.update_self_responses = {}
+    FakeUpdaterClient.rollback_responses = {}
+
+    for mod in (
+        "decnet.web.router.swarm_updates.api_list_host_releases",
+        "decnet.web.router.swarm_updates.api_push_update",
+        "decnet.web.router.swarm_updates.api_push_update_self",
+        "decnet.web.router.swarm_updates.api_rollback_host",
+    ):
+        monkeypatch.setattr(f"{mod}.UpdaterClient", FakeUpdaterClient)
+
+    # Stub the tarball builders so tests don't spend seconds re-tarring the
+    # repo on every assertion. The byte contents don't matter for the route
+    # contract — the updater side is faked.
+    monkeypatch.setattr(
+        "decnet.web.router.swarm_updates.api_push_update.tar_working_tree",
+        lambda root, extra_excludes=None: b"tarball-bytes",
+    )
+    monkeypatch.setattr(
+        "decnet.web.router.swarm_updates.api_push_update.detect_git_sha",
+        lambda root: "deadbeef",
+    )
+    monkeypatch.setattr(
+        "decnet.web.router.swarm_updates.api_push_update_self.tar_working_tree",
+        lambda root, extra_excludes=None: b"tarball-bytes",
+    )
+    monkeypatch.setattr(
+        "decnet.web.router.swarm_updates.api_push_update_self.detect_git_sha",
+        lambda root: "deadbeef",
+    )
+
+    return {"client": FakeUpdaterClient, "Response": FakeResponse}
+
+
+@pytest.fixture
+def connection_drop_exc():
+    """A realistic 'updater re-exec mid-response' exception."""
+    return httpx.RemoteProtocolError("server disconnected")

tests/api/swarm_updates/test_list_host_releases.py

@@ -0,0 +1,69 @@
+"""GET /api/v1/swarm-updates/hosts — per-host updater health fan-out."""
+from __future__ import annotations
+
+import pytest
+
+
+@pytest.mark.anyio
+async def test_admin_lists_reachable_and_unreachable_hosts(
+    client, auth_token, add_host, fake_updater,
+):
+    await add_host("alpha", "10.0.0.1")
+    await add_host("beta", "10.0.0.2")
+
+    fake_updater["client"].health_responses = {
+        "alpha": {
+            "status": "ok",
+            "agent_status": "ok",
+            "releases": [
+                {"slot": "active", "sha": "aaaa111", "healthy": True},
+                {"slot": "prev", "sha": "0000000", "healthy": True},
+            ],
+        },
+        "beta": RuntimeError("TLS handshake failed"),
+    }
+
+    resp = await client.get(
+        "/api/v1/swarm-updates/hosts",
+        headers={"Authorization": f"Bearer {auth_token}"},
+    )
+    assert resp.status_code == 200
+    hosts = {h["host_name"]: h for h in resp.json()["hosts"]}
+    assert hosts["alpha"]["reachable"] is True
+    assert hosts["alpha"]["current_sha"] == "aaaa111"
+    assert hosts["alpha"]["previous_sha"] == "0000000"
+    assert hosts["beta"]["reachable"] is False
+    assert "TLS handshake" in hosts["beta"]["detail"]
+
+
+@pytest.mark.anyio
+async def test_decommissioned_and_agent_only_hosts_are_excluded(
+    client, auth_token, add_host, fake_updater,
+):
+    await add_host("good", "10.0.0.1", with_updater=True)
+    await add_host("gone", "10.0.0.2", with_updater=True, status="decommissioned")
+    await add_host("agentonly", "10.0.0.3", with_updater=False)
+
+    resp = await client.get(
+        "/api/v1/swarm-updates/hosts",
+        headers={"Authorization": f"Bearer {auth_token}"},
+    )
+    assert resp.status_code == 200
+    names = {h["host_name"] for h in resp.json()["hosts"]}
+    assert names == {"good"}
+
+
+@pytest.mark.anyio
+async def test_viewer_is_forbidden(client, viewer_token, add_host, fake_updater):
+    await add_host("alpha")
+    resp = await client.get(
+        "/api/v1/swarm-updates/hosts",
+        headers={"Authorization": f"Bearer {viewer_token}"},
+    )
+    assert resp.status_code == 403
+
+
+@pytest.mark.anyio
+async def test_unauth_returns_401(client):
+    resp = await client.get("/api/v1/swarm-updates/hosts")
+    assert resp.status_code == 401

tests/api/swarm_updates/test_push_update.py

@@ -0,0 +1,176 @@
+"""POST /api/v1/swarm-updates/push — happy paths, rollback, validation."""
+from __future__ import annotations
+
+import pytest
+
+
+@pytest.mark.anyio
+async def test_push_to_single_host_success(client, auth_token, add_host, fake_updater):
+    h = await add_host("alpha")
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuids": [h["uuid"]]},
+    )
+    assert resp.status_code == 200
+    body = resp.json()
+    assert body["sha"] == "deadbeef"
+    assert body["tarball_bytes"] == len(b"tarball-bytes")
+    assert body["results"][0]["status"] == "updated"
+    assert body["results"][0]["host_name"] == "alpha"
+
+
+@pytest.mark.anyio
+async def test_push_reports_rollback_on_409(client, auth_token, add_host, fake_updater):
+    h = await add_host("alpha")
+    Resp = fake_updater["Response"]
+    fake_updater["client"].update_responses = {
+        "alpha": Resp(409, {"error": "probe timed out", "stderr": "boom", "rolled_back": True}),
+    }
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuids": [h["uuid"]]},
+    )
+    assert resp.status_code == 200
+    result = resp.json()["results"][0]
+    assert result["status"] == "rolled-back"
+    assert result["http_status"] == 409
+    assert result["stderr"] == "boom"
+
+
+@pytest.mark.anyio
+async def test_push_all_aggregates_mixed_results(client, auth_token, add_host, fake_updater):
+    await add_host("alpha", "10.0.0.1")
+    await add_host("beta", "10.0.0.2")
+    Resp = fake_updater["Response"]
+    fake_updater["client"].update_responses = {
+        "alpha": Resp(200, {"probe": "ok"}),
+        "beta": RuntimeError("connect timeout"),
+    }
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"all": True},
+    )
+    assert resp.status_code == 200
+    statuses = {r["host_name"]: r["status"] for r in resp.json()["results"]}
+    assert statuses == {"alpha": "updated", "beta": "failed"}
+
+
+@pytest.mark.anyio
+async def test_tarball_built_once_across_multi_host_push(
+    client, auth_token, add_host, fake_updater, monkeypatch,
+):
+    await add_host("alpha", "10.0.0.1")
+    await add_host("beta", "10.0.0.2")
+    calls = {"count": 0}
+
+    def counted(root, extra_excludes=None):
+        calls["count"] += 1
+        return b"tarball-bytes"
+
+    monkeypatch.setattr(
+        "decnet.web.router.swarm_updates.api_push_update.tar_working_tree", counted,
+    )
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"all": True},
+    )
+    assert resp.status_code == 200
+    assert calls["count"] == 1
+
+
+@pytest.mark.anyio
+async def test_include_self_only_runs_update_self_on_success(
+    client, auth_token, add_host, fake_updater,
+):
+    await add_host("alpha", "10.0.0.1")
+    await add_host("beta", "10.0.0.2")
+    Resp = fake_updater["Response"]
+    fake_updater["client"].update_responses = {
+        "alpha": Resp(200, {"probe": "ok"}),
+        "beta": Resp(409, {"error": "bad", "rolled_back": True}),
+    }
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"all": True, "include_self": True},
+    )
+    assert resp.status_code == 200
+    results = {r["host_name"]: r for r in resp.json()["results"]}
+    assert results["alpha"]["status"] == "self-updated"
+    assert results["beta"]["status"] == "rolled-back"
+    # update_self must NOT have been called on beta (rolled-back agent).
+    methods_called = [(name, m) for name, m, _ in fake_updater["client"].calls]
+    assert ("beta", "update_self") not in methods_called
+    assert ("alpha", "update_self") in methods_called
+
+
+@pytest.mark.anyio
+async def test_include_self_tolerates_expected_connection_drop(
+    client, auth_token, add_host, fake_updater, connection_drop_exc,
+):
+    await add_host("alpha", "10.0.0.1")
+    fake_updater["client"].update_self_responses = {
+        "alpha": connection_drop_exc,
+    }
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"all": True, "include_self": True},
+    )
+    assert resp.status_code == 200
+    assert resp.json()["results"][0]["status"] == "self-updated"
+
+
+@pytest.mark.anyio
+async def test_host_and_all_are_mutually_exclusive(
+    client, auth_token, add_host, fake_updater,
+):
+    h = await add_host("alpha")
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuids": [h["uuid"]], "all": True},
+    )
+    assert resp.status_code == 400
+
+
+@pytest.mark.anyio
+async def test_neither_host_nor_all_rejected(client, auth_token, fake_updater):
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={},
+    )
+    assert resp.status_code == 400
+
+
+@pytest.mark.anyio
+async def test_unknown_host_uuid_returns_404(client, auth_token, fake_updater):
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuids": ["nonexistent"]},
+    )
+    assert resp.status_code == 404
+
+
+@pytest.mark.anyio
+async def test_viewer_is_forbidden(client, viewer_token, add_host, fake_updater):
+    h = await add_host("alpha")
+    resp = await client.post(
+        "/api/v1/swarm-updates/push",
+        headers={"Authorization": f"Bearer {viewer_token}"},
+        json={"host_uuids": [h["uuid"]]},
+    )
+    assert resp.status_code == 403

tests/api/swarm_updates/test_push_update_self.py

@@ -0,0 +1,67 @@
+"""POST /api/v1/swarm-updates/push-self — updater-only upgrade path."""
+from __future__ import annotations
+
+import pytest
+
+
+@pytest.mark.anyio
+async def test_push_self_only_calls_update_self(client, auth_token, add_host, fake_updater):
+    await add_host("alpha")
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push-self",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"all": True},
+    )
+    assert resp.status_code == 200
+    assert resp.json()["results"][0]["status"] == "self-updated"
+    methods = [m for _, m, _ in fake_updater["client"].calls]
+    assert "update" not in methods
+    assert "update_self" in methods
+
+
+@pytest.mark.anyio
+async def test_push_self_reports_failure(client, auth_token, add_host, fake_updater):
+    await add_host("alpha")
+    Resp = fake_updater["Response"]
+    fake_updater["client"].update_self_responses = {
+        "alpha": Resp(500, {"error": "pip failed", "stderr": "no module named typer"}),
+    }
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push-self",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"all": True},
+    )
+    assert resp.status_code == 200
+    result = resp.json()["results"][0]
+    assert result["status"] == "self-failed"
+    assert result["http_status"] == 500
+    assert "typer" in (result["stderr"] or "")
+
+
+@pytest.mark.anyio
+async def test_push_self_treats_connection_drop_as_success(
+    client, auth_token, add_host, fake_updater, connection_drop_exc,
+):
+    await add_host("alpha")
+    fake_updater["client"].update_self_responses = {"alpha": connection_drop_exc}
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/push-self",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"all": True},
+    )
+    assert resp.status_code == 200
+    assert resp.json()["results"][0]["status"] == "self-updated"
+
+
+@pytest.mark.anyio
+async def test_viewer_is_forbidden(client, viewer_token, add_host, fake_updater):
+    await add_host("alpha")
+    resp = await client.post(
+        "/api/v1/swarm-updates/push-self",
+        headers={"Authorization": f"Bearer {viewer_token}"},
+        json={"all": True},
+    )
+    assert resp.status_code == 403

tests/api/swarm_updates/test_rollback_host.py

@@ -0,0 +1,86 @@
+"""POST /api/v1/swarm-updates/rollback — single-host manual rollback."""
+from __future__ import annotations
+
+import pytest
+
+
+@pytest.mark.anyio
+async def test_rollback_happy_path(client, auth_token, add_host, fake_updater):
+    h = await add_host("alpha")
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/rollback",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuid": h["uuid"]},
+    )
+    assert resp.status_code == 200
+    body = resp.json()
+    assert body["status"] == "rolled-back"
+    assert body["host_name"] == "alpha"
+
+
+@pytest.mark.anyio
+async def test_rollback_404_when_no_previous(client, auth_token, add_host, fake_updater):
+    h = await add_host("alpha")
+    Resp = fake_updater["Response"]
+    fake_updater["client"].rollback_responses = {
+        "alpha": Resp(404, {"detail": "no previous release"}),
+    }
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/rollback",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuid": h["uuid"]},
+    )
+    assert resp.status_code == 404
+    assert "no previous" in resp.json()["detail"].lower()
+
+
+@pytest.mark.anyio
+async def test_rollback_transport_failure_reported(client, auth_token, add_host, fake_updater):
+    h = await add_host("alpha")
+    fake_updater["client"].rollback_responses = {"alpha": RuntimeError("TLS handshake failed")}
+
+    resp = await client.post(
+        "/api/v1/swarm-updates/rollback",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuid": h["uuid"]},
+    )
+    assert resp.status_code == 200
+    body = resp.json()
+    assert body["status"] == "failed"
+    assert "TLS handshake" in body["detail"]
+
+
+@pytest.mark.anyio
+async def test_rollback_unknown_host(client, auth_token, fake_updater):
+    resp = await client.post(
+        "/api/v1/swarm-updates/rollback",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuid": "nonexistent"},
+    )
+    assert resp.status_code == 404
+
+
+@pytest.mark.anyio
+async def test_rollback_on_agent_only_host_rejected(
+    client, auth_token, add_host, fake_updater,
+):
+    h = await add_host("alpha", with_updater=False)
+    resp = await client.post(
+        "/api/v1/swarm-updates/rollback",
+        headers={"Authorization": f"Bearer {auth_token}"},
+        json={"host_uuid": h["uuid"]},
+    )
+    assert resp.status_code == 400
+
+
+@pytest.mark.anyio
+async def test_viewer_is_forbidden(client, viewer_token, add_host, fake_updater):
+    h = await add_host("alpha")
+    resp = await client.post(
+        "/api/v1/swarm-updates/rollback",
+        headers={"Authorization": f"Bearer {viewer_token}"},
+        json={"host_uuid": h["uuid"]},
+    )
+    assert resp.status_code == 403