feat(web): Remote Updates API — dashboard endpoints for pushing code to workers

Adds /api/v1/swarm-updates/{hosts,push,push-self,rollback} behind
require_admin. Reuses the existing UpdaterClient + tar_working_tree + the
per-host asyncio.gather pattern from api_deploy_swarm.py; tarball is
built exactly once per /push request and fanned out to every selected
worker. /hosts filters out decommissioned hosts and agent-only
enrollments (no updater bundle = not a target).

Connection drops during /update-self are treated as success — the
updater re-execs itself mid-response, so httpx always raises.

Pydantic models live in decnet/web/db/models.py (single source of
truth). 24 tests cover happy paths, rollback, transport failures,
include_self ordering (skip on rolled-back agents), validation, and
RBAC gating.

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,
+    )