feat: refine Obsidian plugin routes and schemas for improved device management and API stability

This commit is contained in:
Anish Sarkar 2026-04-20 18:19:30 +05:30
parent 60d9e7ed8c
commit b5c9388c8a
9 changed files with 182 additions and 385 deletions

View file

@ -1,31 +1,8 @@
"""
Obsidian plugin ingestion routes.
"""Obsidian plugin ingestion routes (``/api/v1/obsidian/*``).
This is the public surface that the SurfSense Obsidian plugin
(``surfsense_obsidian/``) speaks to. It is a separate router from the
legacy server-path Obsidian connector the legacy code stays in place
until the ``obsidian-legacy-cleanup`` plan ships.
Endpoints
---------
- ``GET /api/v1/obsidian/health`` version handshake
- ``POST /api/v1/obsidian/connect`` register or get a vault row
- ``POST /api/v1/obsidian/sync`` batch upsert
- ``POST /api/v1/obsidian/rename`` batch rename
- ``DELETE /api/v1/obsidian/notes`` batch soft-delete
- ``GET /api/v1/obsidian/manifest`` reconcile manifest
Auth contract
-------------
Every endpoint requires ``Depends(current_active_user)`` the same JWT
bearer the rest of the API uses; future PAT migration is transparent.
API stability is provided by the ``/api/v1/...`` URL prefix and the
``capabilities`` array advertised on ``/health`` (additive only). There
is no plugin-version gate; "your plugin is out of date" notices are
delegated to Obsidian's built-in community-store updater.
Wire surface for the ``surfsense_obsidian/`` plugin. API stability is the
``/api/v1/`` prefix plus the additive ``capabilities`` array on /health;
no plugin-version gate.
"""
from __future__ import annotations
@ -67,14 +44,10 @@ logger = logging.getLogger(__name__)
router = APIRouter(prefix="/obsidian", tags=["obsidian-plugin"])
# Bumped manually whenever the wire contract gains a non-additive change.
# Additive (extra='ignore'-safe) changes do NOT bump this.
# Bumped only on non-additive wire changes; additive ones ride extra='ignore'.
OBSIDIAN_API_VERSION = "1"
# Capabilities advertised on /health and /connect. Plugins use this list
# for feature gating ("does this server understand attachments_v2?"). Add
# new strings, never rename/remove existing ones — older plugins ignore
# unknown entries safely.
# Plugins feature-gate on these. Add entries, never rename or remove.
OBSIDIAN_CAPABILITIES: list[str] = ["sync", "rename", "delete", "manifest"]
@ -90,18 +63,41 @@ def _build_handshake() -> dict[str, object]:
}
def _upsert_device(
existing_devices: object,
device_id: str,
now_iso: str,
) -> dict[str, dict[str, str]]:
"""Upsert ``device_id`` into ``{device_id: {first_seen_at, last_seen_at}}``.
Keyed by device_id for O(1) dedup; ``len(devices)`` is the count.
Timestamps are kept for a future stale-device pruner.
"""
devices: dict[str, dict[str, str]] = {}
if isinstance(existing_devices, dict):
for key, val in existing_devices.items():
if not isinstance(key, str) or not key or not isinstance(val, dict):
continue
devices[key] = {
"first_seen_at": str(val.get("first_seen_at") or now_iso),
"last_seen_at": str(val.get("last_seen_at") or now_iso),
}
prev = devices.get(device_id)
devices[device_id] = {
"first_seen_at": prev["first_seen_at"] if prev else now_iso,
"last_seen_at": now_iso,
}
return devices
async def _resolve_vault_connector(
session: AsyncSession,
*,
user: User,
vault_id: str,
) -> SearchSourceConnector:
"""Find the OBSIDIAN_CONNECTOR row that owns ``vault_id`` for this user.
Looked up by the (user_id, connector_type, config['vault_id']) tuple
so users can have multiple vaults each backed by its own connector
row (one per search space).
"""
"""Find the OBSIDIAN_CONNECTOR row that owns ``vault_id`` for this user."""
result = await session.execute(
select(SearchSourceConnector).where(
and_(
@ -136,12 +132,7 @@ async def _ensure_search_space_access(
user: User,
search_space_id: int,
) -> SearchSpace:
"""Confirm the user owns the requested search space.
Plugin currently does not support shared search spaces (RBAC roles)
that's a follow-up. Restricting to owner-only here keeps the
surface narrow and avoids leaking other members' connectors.
"""
"""Owner-only access to the search space (shared spaces are a follow-up)."""
result = await session.execute(
select(SearchSpace).where(
and_(SearchSpace.id == search_space_id, SearchSpace.user_id == user.id)
@ -168,11 +159,7 @@ async def _ensure_search_space_access(
async def obsidian_health(
user: User = Depends(current_active_user),
) -> HealthResponse:
"""Return the API contract handshake.
The plugin calls this once per ``onload`` and caches the result for
capability-gating decisions.
"""
"""Return the API contract handshake; plugin caches it per onload."""
return HealthResponse(
**_build_handshake(),
server_time_utc=datetime.now(UTC),
@ -187,9 +174,9 @@ async def obsidian_connect(
) -> ConnectResponse:
"""Register a vault, or return the existing connector row.
Idempotent on the (user_id, OBSIDIAN_CONNECTOR, vault_id) tuple so
re-installing the plugin or reconnecting from a new device picks up
the same connector and therefore the same documents.
Idempotent on (user_id, OBSIDIAN_CONNECTOR, vault_id). Called on every
plugin onload as a heartbeat upserts ``device_id`` into
``config['devices']`` so the web UI can show a "Devices: N" tile.
"""
await _ensure_search_space_access(
session, user=user, search_space_id=payload.search_space_id
@ -215,27 +202,31 @@ async def obsidian_connect(
if existing is not None:
cfg = dict(existing.config or {})
devices = _upsert_device(cfg.get("devices"), payload.device_id, now_iso)
cfg.update(
{
"vault_id": payload.vault_id,
"vault_name": payload.vault_name,
"source": "plugin",
"plugin_version": payload.plugin_version,
"device_id": payload.device_id,
"devices": devices,
"device_count": len(devices),
"last_connect_at": now_iso,
}
)
if payload.device_label:
cfg["device_label"] = payload.device_label
cfg.pop("legacy", None)
cfg.pop("vault_path", None)
existing.config = cfg
# Re-stamp on every connect so vault renames in Obsidian propagate;
# the web UI hides the Name input for Obsidian connectors.
existing.name = f"Obsidian — {payload.vault_name}"
existing.is_indexable = False
existing.search_space_id = payload.search_space_id
await session.commit()
await session.refresh(existing)
connector = existing
else:
devices = _upsert_device(None, payload.device_id, now_iso)
connector = SearchSourceConnector(
name=f"Obsidian — {payload.vault_name}",
connector_type=SearchSourceConnectorType.OBSIDIAN_CONNECTOR,
@ -245,8 +236,8 @@ async def obsidian_connect(
"vault_name": payload.vault_name,
"source": "plugin",
"plugin_version": payload.plugin_version,
"device_id": payload.device_id,
"device_label": payload.device_label,
"devices": devices,
"device_count": len(devices),
"files_synced": 0,
"last_connect_at": now_iso,
},
@ -271,11 +262,7 @@ async def obsidian_sync(
user: User = Depends(current_active_user),
session: AsyncSession = Depends(get_async_session),
) -> dict[str, object]:
"""Batch-upsert notes pushed by the plugin.
Returns per-note ack so the plugin can dequeue successes and retry
failures.
"""
"""Batch-upsert notes; returns per-note ack so the plugin can dequeue/retry."""
connector = await _resolve_vault_connector(
session, user=user, vault_id=payload.vault_id
)
@ -439,11 +426,7 @@ async def obsidian_manifest(
user: User = Depends(current_active_user),
session: AsyncSession = Depends(get_async_session),
) -> ManifestResponse:
"""Return the server-side ``{path: {hash, mtime}}`` manifest.
Used by the plugin's ``onload`` reconcile to find files that were
edited or deleted while the plugin was offline.
"""
"""Return ``{path: {hash, mtime}}`` for the plugin's onload reconcile diff."""
connector = await _resolve_vault_connector(
session, user=user, vault_id=vault_id
)

View file

@ -1,23 +1,8 @@
"""
Obsidian Plugin connector schemas.
"""Wire schemas spoken between the SurfSense Obsidian plugin and the backend.
Wire format spoken between the SurfSense Obsidian plugin
(``surfsense_obsidian/``) and the FastAPI backend.
Stability contract
------------------
Every request and response schema sets ``model_config = ConfigDict(extra='ignore')``.
This is the API stability contract not just hygiene:
- Old plugins talking to a newer backend silently drop any new response fields
they don't understand instead of failing validation.
- New plugins talking to an older backend can include forward-looking request
fields (e.g. attachments metadata) without the older backend rejecting them.
Hard breaking changes are reserved for the URL prefix (``/api/v2/...``).
Additive evolution is signaled via the ``capabilities`` array on
``HealthResponse`` / ``ConnectResponse`` older plugins ignore unknown
capability strings safely.
All schemas inherit ``extra='ignore'`` from :class:`_PluginBase` so additive
field changes never break either side; hard breaks live behind a new URL
prefix (``/api/v2/...``).
"""
from __future__ import annotations
@ -31,22 +16,13 @@ _PLUGIN_MODEL_CONFIG = ConfigDict(extra="ignore")
class _PluginBase(BaseModel):
"""Base class for all plugin payload schemas.
Carries the forward-compatibility config so subclasses don't have to
repeat it.
"""
"""Base schema carrying the shared forward-compatibility config."""
model_config = _PLUGIN_MODEL_CONFIG
class NotePayload(_PluginBase):
"""One Obsidian note as pushed by the plugin.
The plugin is the source of truth: ``content`` is the post-frontmatter
body, ``frontmatter``/``tags``/``headings``/etc. are precomputed by the
plugin via ``app.metadataCache`` so the backend doesn't have to re-parse.
"""
"""One Obsidian note as pushed by the plugin (the source of truth)."""
vault_id: str = Field(..., description="Stable plugin-generated UUID for this vault")
path: str = Field(..., description="Vault-relative path, e.g. 'notes/foo.md'")
@ -68,7 +44,7 @@ class NotePayload(_PluginBase):
class SyncBatchRequest(_PluginBase):
"""Batch upsert. Plugin sends 10-20 notes per request to amortize HTTP overhead."""
"""Batch upsert; plugin sends 10-20 notes per request."""
vault_id: str
notes: list[NotePayload] = Field(default_factory=list, max_length=100)
@ -90,8 +66,6 @@ class DeleteBatchRequest(_PluginBase):
class ManifestEntry(_PluginBase):
"""One row of the server-side manifest used by the plugin to reconcile."""
hash: str
mtime: datetime
@ -104,26 +78,18 @@ class ManifestResponse(_PluginBase):
class ConnectRequest(_PluginBase):
"""First-call handshake to register or look up a vault connector row."""
"""Vault registration / heartbeat. Replayed on every plugin onload."""
vault_id: str
vault_name: str
search_space_id: int
plugin_version: str
device_id: str
device_label: str | None = Field(
default=None,
description="User-friendly device name shown in the web UI (e.g. 'iPad Pro').",
)
class ConnectResponse(_PluginBase):
"""Returned from POST /connect.
Carries the same handshake fields as ``HealthResponse`` so the plugin
learns the contract on its very first call without an extra round-trip
to ``GET /health``.
"""
"""Carries the same handshake fields as ``HealthResponse`` so the plugin
learns the contract without a separate ``GET /health`` round-trip."""
connector_id: int
vault_id: str
@ -133,14 +99,7 @@ class ConnectResponse(_PluginBase):
class HealthResponse(_PluginBase):
"""API contract handshake.
The plugin calls ``GET /health`` once per ``onload`` and caches the
result. ``capabilities`` is a forward-extensible string list: future
additions (``'pat_auth'``, ``'scoped_pat'``, ``'attachments_v2'``,
``'shared_search_spaces'``...) ship without breaking older plugins
because they only enable extra behavior, never gate existing endpoints.
"""
"""API contract handshake. ``capabilities`` is additive-only string list."""
api_version: str
capabilities: list[str]