@rebasepro/server-postgres 0.17.3 → 0.18.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (149) hide show
  1. package/LICENSE +0 -1
  2. package/README.md +31 -0
  3. package/dist/backup-argv.d.ts +19 -0
  4. package/dist/branch-argv.d.ts +28 -0
  5. package/dist/branch-prune.d.ts +77 -0
  6. package/dist/cli-collections-path.d.ts +31 -0
  7. package/dist/cli-errors.d.ts +123 -0
  8. package/dist/cli-flags.d.ts +101 -0
  9. package/dist/cli-helpers.d.ts +94 -10
  10. package/dist/collection-index-C7LopMyX.js +247 -0
  11. package/dist/collection-index-C7LopMyX.js.map +1 -0
  12. package/dist/databasePoolManager.d.ts +27 -0
  13. package/dist/{ensure-collection-policies-UCqgv_8c.js → ensure-collection-policies-CagOzEPS.js} +6 -6
  14. package/dist/{ensure-collection-policies-UCqgv_8c.js.map → ensure-collection-policies-CagOzEPS.js.map} +1 -1
  15. package/dist/{ensure-collection-tables-DgVixhX3.js → ensure-collection-tables-DbAhrjnG.js} +614 -52
  16. package/dist/ensure-collection-tables-DbAhrjnG.js.map +1 -0
  17. package/dist/index.d.ts +12 -0
  18. package/dist/index.es.js +19307 -3059
  19. package/dist/index.es.js.map +1 -1
  20. package/dist/{rls-enforcement-DvAbL9YJ.js → rls-enforcement-ClJoTboT.js} +53 -189
  21. package/dist/rls-enforcement-ClJoTboT.js.map +1 -0
  22. package/dist/schema/column-type-drift.d.ts +79 -0
  23. package/dist/schema/doctor.d.ts +42 -0
  24. package/dist/schema/ensure-collection-tables.d.ts +90 -0
  25. package/dist/schema/generate-next-step.d.ts +30 -0
  26. package/dist/schema/generate-postgres-ddl-logic.d.ts +28 -0
  27. package/dist/schema/generated-schema-staleness.d.ts +37 -0
  28. package/dist/schema/introspect-db-queries.d.ts +34 -0
  29. package/dist/security/rls-enforcement.d.ts +63 -2
  30. package/dist/services/BranchService.d.ts +86 -1
  31. package/dist/services/collection-helpers.d.ts +1 -1
  32. package/dist/services/realtimeService.d.ts +30 -2
  33. package/dist/services/row-pipeline.d.ts +12 -1
  34. package/dist/utils/pg-error-utils.d.ts +3 -1
  35. package/dist/utils/pg-numeric-number-patch.d.ts +31 -0
  36. package/package.json +32 -27
  37. package/dist/backup-service-DCk7KhhL.js +0 -9044
  38. package/dist/backup-service-DCk7KhhL.js.map +0 -1
  39. package/dist/collection-index-BRUg10H5.js +0 -5690
  40. package/dist/collection-index-BRUg10H5.js.map +0 -1
  41. package/dist/connection-GOKU3Hu5.js +0 -277
  42. package/dist/connection-GOKU3Hu5.js.map +0 -1
  43. package/dist/ensure-collection-tables-DgVixhX3.js.map +0 -1
  44. package/dist/rls-bootstrap-sql-B5C9LoJ6.js +0 -244
  45. package/dist/rls-bootstrap-sql-B5C9LoJ6.js.map +0 -1
  46. package/dist/rls-enforcement-DvAbL9YJ.js.map +0 -1
  47. package/dist/src-DiB5RP2Z.js +0 -541
  48. package/dist/src-DiB5RP2Z.js.map +0 -1
  49. package/dist/websocket-BZ4H5wUz.js +0 -8198
  50. package/dist/websocket-BZ4H5wUz.js.map +0 -1
  51. package/src/PostgresAdapter.ts +0 -91
  52. package/src/PostgresBackendDriver.ts +0 -1857
  53. package/src/PostgresBootstrapper.ts +0 -1157
  54. package/src/auth/ensure-tables.ts +0 -896
  55. package/src/auth/schema-version.ts +0 -260
  56. package/src/auth/services.ts +0 -1610
  57. package/src/backup/backup-cli.ts +0 -520
  58. package/src/backup/backup-cron.ts +0 -212
  59. package/src/backup/backup-logic.ts +0 -93
  60. package/src/backup/backup-service.ts +0 -485
  61. package/src/backup/index.ts +0 -12
  62. package/src/backup/pg-tools.ts +0 -403
  63. package/src/backup/retention.ts +0 -86
  64. package/src/cli-errors.ts +0 -265
  65. package/src/cli-helpers.ts +0 -569
  66. package/src/cli-output.ts +0 -43
  67. package/src/cli.ts +0 -1339
  68. package/src/collections/PostgresCollectionRegistry.ts +0 -108
  69. package/src/collections/buildRegistry.ts +0 -68
  70. package/src/collections/validate-relations.ts +0 -402
  71. package/src/connection.ts +0 -333
  72. package/src/data-transformer.ts +0 -845
  73. package/src/databasePoolManager.ts +0 -95
  74. package/src/history/HistoryService.ts +0 -254
  75. package/src/history/ensure-history-table.ts +0 -53
  76. package/src/index.ts +0 -19
  77. package/src/interfaces.ts +0 -60
  78. package/src/module-dir.ts +0 -7
  79. package/src/schema/atlas-argv.ts +0 -94
  80. package/src/schema/auth-schema.ts +0 -298
  81. package/src/schema/auth-users-columns.ts +0 -131
  82. package/src/schema/carved-out-migration.ts +0 -404
  83. package/src/schema/classify-change.ts +0 -436
  84. package/src/schema/collection-index.ts +0 -427
  85. package/src/schema/collections-schema-version.ts +0 -103
  86. package/src/schema/destructive-sql.ts +0 -94
  87. package/src/schema/doctor-cli.ts +0 -66
  88. package/src/schema/doctor-policy-checks.ts +0 -105
  89. package/src/schema/doctor.ts +0 -837
  90. package/src/schema/drizzle-ddl.ts +0 -33
  91. package/src/schema/dynamic-tables.test.ts +0 -302
  92. package/src/schema/dynamic-tables.ts +0 -293
  93. package/src/schema/ensure-collection-policies.ts +0 -198
  94. package/src/schema/ensure-collection-tables.test.ts +0 -518
  95. package/src/schema/ensure-collection-tables.ts +0 -1267
  96. package/src/schema/generate-drizzle-schema-logic.ts +0 -957
  97. package/src/schema/generate-drizzle-schema.ts +0 -143
  98. package/src/schema/generate-postgres-ddl-logic.ts +0 -1339
  99. package/src/schema/generate-postgres-ddl.ts +0 -180
  100. package/src/schema/generate-schema-commit.ts +0 -267
  101. package/src/schema/generated-schema-staleness.ts +0 -284
  102. package/src/schema/introspect-db-constraints.ts +0 -385
  103. package/src/schema/introspect-db-inference.ts +0 -255
  104. package/src/schema/introspect-db-logic.ts +0 -1387
  105. package/src/schema/introspect-db-naming.ts +0 -15
  106. package/src/schema/introspect-db-project.ts +0 -78
  107. package/src/schema/introspect-db-queries.ts +0 -326
  108. package/src/schema/introspect-db-structure.ts +0 -670
  109. package/src/schema/introspect-db-types.ts +0 -56
  110. package/src/schema/introspect-db.ts +0 -257
  111. package/src/schema/introspect-runtime.test.ts +0 -260
  112. package/src/schema/introspect-runtime.ts +0 -327
  113. package/src/schema/non-sql-collections.test.ts +0 -131
  114. package/src/schema/rls-bootstrap-sql.ts +0 -288
  115. package/src/schema/search-column.ts +0 -643
  116. package/src/schema/test-schema.ts +0 -11
  117. package/src/schema/vector-index.ts +0 -491
  118. package/src/security/anonymous-grants.test.ts +0 -73
  119. package/src/security/policy-drift.test.ts +0 -435
  120. package/src/security/policy-drift.ts +0 -473
  121. package/src/security/rls-enforcement.ts +0 -505
  122. package/src/services/BranchService.ts +0 -326
  123. package/src/services/FetchService.ts +0 -2115
  124. package/src/services/PersistService.ts +0 -557
  125. package/src/services/RelationService.ts +0 -978
  126. package/src/services/RelationWriteService.ts +0 -653
  127. package/src/services/cdc/CdcListener.ts +0 -103
  128. package/src/services/cdc/junction-tables.ts +0 -91
  129. package/src/services/cdc/trigger-cdc.ts +0 -173
  130. package/src/services/channel-bus/ChannelBus.ts +0 -44
  131. package/src/services/channel-bus/PostgresChannelBus.ts +0 -299
  132. package/src/services/channel-bus/index.ts +0 -114
  133. package/src/services/channel-history.ts +0 -424
  134. package/src/services/channel-presence.ts +0 -185
  135. package/src/services/collection-helpers.ts +0 -420
  136. package/src/services/dataService.ts +0 -256
  137. package/src/services/index.ts +0 -14
  138. package/src/services/junction-writes.ts +0 -295
  139. package/src/services/nested-path.ts +0 -145
  140. package/src/services/pg-notify-listener.ts +0 -151
  141. package/src/services/realtimeService.ts +0 -2545
  142. package/src/services/row-pipeline.ts +0 -240
  143. package/src/services/write-denial.ts +0 -55
  144. package/src/types.ts +0 -4
  145. package/src/utils/connection-string.ts +0 -58
  146. package/src/utils/drizzle-conditions.ts +0 -2743
  147. package/src/utils/pg-array-null-patch.ts +0 -42
  148. package/src/utils/pg-error-utils.ts +0 -391
  149. package/src/websocket.ts +0 -855
@@ -1,2545 +0,0 @@
1
- import { WebSocket } from "ws";
2
- import { EventEmitter } from "events";
3
- import { Client as PgClient } from "pg";
4
- import { randomUUID } from "crypto";
5
- import { DataService } from "./dataService";
6
-
7
- import { ANONYMOUS_USER_ID, FetchCollectionProps, ListenCollectionProps, ListenOneProps, DataDriver, CollectionUpdateMessage, SingleUpdateMessage, CollectionPatchMessage, WebSocketMessage, FilterValues, LogicalCondition, OrderByTuple, CollectionConfig, RebaseCallContext, resolveClientListLimit, ListLimitError } from "@rebasepro/types";
8
- import { NodePgDatabase } from "drizzle-orm/node-postgres";
9
- import { sql as drizzleSql } from "drizzle-orm";
10
- import { RealtimeProvider, CollectionSubscriptionConfig, SingleSubscriptionConfig } from "../interfaces";
11
- import { PostgresCollectionRegistry } from "../collections/PostgresCollectionRegistry";
12
- import { buildPropertyCallbacks, getTableName, OrderBySpecError, parseOrderBySpecStrict } from "@rebasepro/common";
13
- import { applyAuthContext } from "../security/rls-enforcement";
14
- import { buildJunctionLinkMap, type JunctionLink } from "./cdc/junction-tables";
15
- import { logger } from "@rebasepro/server";
16
- import { sanitizeErrorForClient } from "../utils/pg-error-utils";
17
- import { CdcListener, type CdcChangeEvent } from "./cdc/CdcListener";
18
- import { deriveRowAddress, getPrimaryKeys, type PrimaryKeyInfo } from "./collection-helpers";
19
- import { ChannelHistoryStore, type ResolvedRetention } from "./channel-history";
20
- import { ChannelPresenceStore } from "./channel-presence";
21
- import { ChannelBus, ChannelBusFrame, MemoryChannelBus, frameByteLength } from "./channel-bus";
22
- import type { ChannelHistoryEntry, ChannelRetentionRule } from "@rebasepro/types";
23
-
24
- /** Channel name used for Postgres LISTEN/NOTIFY cross-instance realtime. */
25
- const PG_NOTIFY_CHANNEL = "rebase_entity_changes";
26
-
27
- /**
28
- * Auth context stored per-subscription so real-time refetches respect RLS.
29
- * Mirrors the session variables set by PostgresBackendDriver.withAuth().
30
- */
31
- export interface SubscriptionAuthContext {
32
- uid: string;
33
- roles: string[];
34
- }
35
-
36
- /** What a channel frame is asking to do. */
37
- export type ChannelAction = "join" | "broadcast" | "presence" | "history";
38
-
39
- /** Everything an authorizer is told about the frame it is asked to allow. */
40
- export interface ChannelAuthorizationRequest {
41
- /** The channel the frame names, exactly as the client wrote it. */
42
- channel: string;
43
- action: ChannelAction;
44
- /** The socket, not the principal — one user may hold several. */
45
- clientId: string;
46
- /** The socket's authenticated principal, or the anonymous one. */
47
- user?: SubscriptionAuthContext;
48
- }
49
-
50
- /**
51
- * The extension point for channel access rules.
52
- *
53
- * **This is deliberately not a product API yet.** The rule *language* — a
54
- * config key, a per-pattern DSL, how it composes with `securityRules` — is an
55
- * open design question (see `docs/channel-authorization.md`), and
56
- * inventing one here would be inventing the answer. What exists is the single
57
- * place every channel frame passes through, so that whatever shape the rules
58
- * eventually take has exactly one seam to plug into and no arm of the switch
59
- * can be forgotten.
60
- *
61
- * Returning `false` — or throwing — refuses the frame. It is consulted *after*
62
- * the membership floor below, so an authorizer can only ever narrow access,
63
- * never widen it.
64
- */
65
- export type ChannelAuthorizer = (request: ChannelAuthorizationRequest) => boolean | Promise<boolean>;
66
-
67
- interface DataDriverWithData extends DataDriver {
68
- data: unknown;
69
- }
70
-
71
- type RealTimeListenCollectionProps = ListenCollectionProps & {
72
- subscriptionId: string
73
- };
74
-
75
- /**
76
- * The narrowing a collection subscription was created with, kept so that every
77
- * refetch answers the same query the initial fetch did.
78
- *
79
- * Named once because it used to be written out inline in five places, and a
80
- * field missing from one of them is accepted over the wire and then silently
81
- * ignored: `offset` was declared on the incoming props and never stored, so a
82
- * live list on page three served page one, and `logical` was never stored
83
- * either, so an `or(...)` subscription was pushed every row in the table.
84
- */
85
- type StoredCollectionRequest = {
86
- filter?: Record<string, unknown>;
87
- logical?: LogicalCondition;
88
- orderBy?: string | OrderByTuple[];
89
- order?: "desc" | "asc";
90
- limit?: number;
91
- offset?: number;
92
- startAfter?: Record<string, unknown>;
93
- databaseId?: string;
94
- searchString?: string;
95
- /** Ask each row which declared search field matched — populates `_matches`. */
96
- searchExplain?: boolean;
97
- };
98
-
99
- type RealTimeListenEntityProps = ListenOneProps & { subscriptionId: string };
100
-
101
- /**
102
- * A registered subscription, plus the two counters that order its deliveries.
103
- *
104
- * Every update a subscription delivers is a full re-fetch, and more than one
105
- * thing starts one for the same subscription without coordinating: the initial
106
- * fetch at subscribe time, and a debounced refetch per notification (app
107
- * mutation, cross-instance NOTIFY, or CDC). A fetch that started earlier can
108
- * finish later, and the delivery replaces everything the subscriber has — so
109
- * the subscriber goes back to the state before the change and stays there,
110
- * silently, until the next write to that collection.
111
- *
112
- * The debounce is not a fix for this. It collapses a burst into one refetch and
113
- * does nothing about two refetches that overlap: notification A fires its timer
114
- * and starts fetch A, notification B arrives while A is still in flight, and B's
115
- * timer fires and starts fetch B regardless. See class 44 in
116
- * `docs/bug-classes.md`.
117
- *
118
- * `started` is taken before the work, `delivered` after it — which makes the
119
- * last delivery *started* the last one *delivered*.
120
- */
121
- type Subscription = {
122
- clientId: string;
123
- type: "collection" | "single";
124
- path: string;
125
- id?: string | number;
126
- // Store full collection request parameters for proper refetching
127
- collectionRequest?: StoredCollectionRequest;
128
- // Auth context for RLS — when set, refetches run in a transaction
129
- // with set_config('app.uid', ...) / set_config('app.user_roles', ...)
130
- authContext?: SubscriptionAuthContext;
131
- /** How many deliveries have been started for this subscription. */
132
- started: number;
133
- /** The highest started-sequence that has already reached the subscriber. */
134
- delivered: number;
135
- };
136
-
137
- /**
138
- * PostgreSQL-specific realtime service.
139
- * Handles WebSocket connections and subscriptions for real-time row updates.
140
- *
141
- * Implements the RealtimeProvider interface for database abstraction.
142
- */
143
- export class RealtimeService extends EventEmitter implements RealtimeProvider {
144
- /**
145
- * Declares to the multi-engine router that channel frames can be handled
146
- * here. Read by `createRoutedRealtimeService`, which otherwise would have to
147
- * guess — and guessed "the default provider", whichever engine that is.
148
- */
149
- public readonly supportsChannels = true;
150
-
151
- private clients = new Map<string, WebSocket>();
152
-
153
- // Broadcast channels: channel name → set of client IDs
154
- private channels = new Map<string, Set<string>>();
155
-
156
- // Presence: channel → Map<clientId, { state, lastSeen }>
157
- private presence = new Map<string, Map<string, { state: Record<string, unknown>; lastSeen: number }>>();
158
-
159
- /**
160
- * Ordered, replayable history for channels that opt into it.
161
- *
162
- * Undefined until {@link configureChannelHistory} is called, and inert even
163
- * then unless retention rules were supplied — so presence and ephemeral
164
- * notification channels never touch it. See `channel-history.ts`.
165
- */
166
- private channelHistory?: ChannelHistoryStore;
167
-
168
- /**
169
- * One promise chain per retained channel, so that assigning a sequence
170
- * number and fanning the message out happen in the same order for every
171
- * message on that channel.
172
- *
173
- * Without it, two concurrent broadcasts can be numbered 4 and 5 by the
174
- * database and still reach subscribers as 5 then 4 — live order and replay
175
- * order would disagree, which is exactly the divergence sequence numbers
176
- * are supposed to rule out. Keyed by channel, so unrelated channels never
177
- * wait on each other.
178
- */
179
- private channelSendQueues = new Map<string, Promise<void>>();
180
-
181
- /**
182
- * Cross-instance transport for channel frames and presence.
183
- *
184
- * Defaults to the memory bus, which publishes nowhere — so a single-instance
185
- * deployment runs the same fan-out it always did, with one resolved promise
186
- * per broadcast for company. See `channel-bus/ChannelBus.ts`.
187
- */
188
- private bus: ChannelBus = new MemoryChannelBus();
189
-
190
- /**
191
- * The shared presence roster, present only when a real bus is active.
192
- *
193
- * Fan-out alone is not enough for presence: `presence_state` has to answer
194
- * with everyone in the channel, and per-process maps can only answer for
195
- * this replica's clients. See `channel-presence.ts`.
196
- */
197
- private presenceStore?: ChannelPresenceStore;
198
-
199
- /** Sweeps roster rows left behind by instances that stopped heartbeating. */
200
- private presenceSweepInterval?: ReturnType<typeof setInterval>;
201
-
202
- /**
203
- * Channels whose oversized ephemeral broadcasts have already been reported,
204
- * so a hot channel logs the problem once rather than once per message.
205
- */
206
- private oversizedBroadcastWarned = new Set<string>();
207
-
208
- /**
209
- * Optional narrowing on top of the membership floor — see
210
- * {@link ChannelAuthorizer}. Unset by default, which leaves membership as
211
- * the whole of the rule.
212
- */
213
- private channelAuthorizer?: ChannelAuthorizer;
214
-
215
- /**
216
- * Whether a notification from another instance has ever arrived.
217
- *
218
- * The entity LISTEN handler sees a foreign `sid` on every cross-instance
219
- * change, which is proof that this deployment runs more than one pod — the
220
- * one fact needed to tell "the memory bus is fine here" from "broadcast and
221
- * presence silently reach a fraction of your users".
222
- */
223
- private foreignInstanceSeen = false;
224
-
225
- /** So the multi-pod memory-bus warning is emitted once, not once per join. */
226
- private memoryBusWarned = false;
227
-
228
- private presenceInterval?: ReturnType<typeof setInterval>;
229
- private static readonly PRESENCE_TIMEOUT_MS = 30000; // 30s
230
- /** How often stale roster rows from other instances are reaped. */
231
- private static readonly PRESENCE_SWEEP_INTERVAL_MS = 10000; // 10s
232
- private dataService: DataService;
233
- // Enhanced subscriptions storage with full request parameters
234
- private _subscriptions = new Map<string, Subscription>();
235
-
236
- // Add callback storage for DataDriver subscriptions
237
- private subscriptionCallbacks = new Map<string, (data: Record<string, unknown>[] | Record<string, unknown> | null) => void>();
238
-
239
- private driver?: DataDriver;
240
-
241
- // ── Cross-instance LISTEN/NOTIFY ──
242
- /** Unique identifier for this process instance, used to skip own notifications. */
243
- private readonly instanceId = `inst_${randomUUID().slice(0, 8)}`;
244
- /** Dedicated pg.Client for LISTEN (outside the Drizzle pool). */
245
- private listenClient?: PgClient;
246
- /** Connection string used for reconnecting the LISTEN client. */
247
- private listenConnectionString?: string;
248
- /** Whether cross-instance broadcasting is active. */
249
- private broadcasting = false;
250
- /** Reconnection timer handle. */
251
- private reconnectTimer?: ReturnType<typeof setTimeout>;
252
- /** Debounce timers for collection refetches to prevent refetch storms. */
253
- private refetchTimers = new Map<string, ReturnType<typeof setTimeout>>();
254
- /** Debounce window (ms) for coalescing rapid row updates into a single correctness refetch. */
255
- private static readonly REFETCH_DEBOUNCE_MS = 300;
256
-
257
- // ── Database-level Change Data Capture (CDC) ──
258
- /** Dedicated LISTEN client for DB-level change events (undefined unless CDC is enabled). */
259
- private cdcListener?: CdcListener;
260
- /** Whether database-level CDC is the active cross-instance change source. */
261
- private cdcActive = false;
262
- /** Junction table → the child lists its rows belong to, built when CDC starts. */
263
- private junctionLinkMap?: Map<string, JunctionLink[]>;
264
-
265
- /** Reverse lookup: `schema.table` (and bare `table`) → collection, built when CDC starts. */
266
- private cdcTableMap?: Map<string, CollectionConfig>;
267
- /**
268
- * Short-lived record of `path/id` keys this instance just fanned out via the
269
- * app path (a Rebase-API mutation). When CDC echoes the same committed change
270
- * back to *this* instance, we suppress the duplicate — the change was already
271
- * delivered locally. Other instances have no such record, so they still
272
- * deliver the CDC event. External writes (psql, cron, SQL editor) never match
273
- * and always flow through. Keyed → expiry timestamp (ms).
274
- */
275
- private recentAppEmits = new Map<string, number>();
276
- /** How long an app-emit key suppresses its own CDC echo. Covers NOTIFY round-trip latency. */
277
- private static readonly CDC_DEDUP_WINDOW_MS = 5000;
278
-
279
- constructor(private db: NodePgDatabase<any>, private registry: PostgresCollectionRegistry) {
280
- super();
281
- this.dataService = new DataService(db, registry);
282
- }
283
-
284
- /**
285
- * Restricted role that auth-scoped refetches run as (via `SET LOCAL ROLE`)
286
- * so RLS `select` policies bind. Set by the bootstrapper alongside
287
- * `PostgresBackendDriver.rlsUserRole`; undefined when the connection
288
- * is already subject to RLS natively. Without this, realtime refetches
289
- * would leak rows the initial (isolated) fetch correctly hid.
290
- */
291
- public rlsUserRole?: string;
292
-
293
- /** Whether to emit verbose debug logs (disabled in production). */
294
- private static readonly DEBUG = process.env.NODE_ENV !== "production";
295
- private debugLog(...args: unknown[]) {
296
- if (RealtimeService.DEBUG) console.debug(...args);
297
- }
298
-
299
- setDataDriver(driver: DataDriver) {
300
- this.driver = driver;
301
- }
302
-
303
- // Make subscriptions accessible for DataDriver
304
- get subscriptions() {
305
- return this._subscriptions;
306
- }
307
-
308
- /**
309
- * Claim a delivery slot for a subscription, before doing the work.
310
- *
311
- * Returns the check to run immediately before delivering. It refuses in
312
- * three cases, all of which used to deliver:
313
- *
314
- * - **Out of order.** A newer refetch has already delivered, so this one is
315
- * stale — the subscriber would go back to the state before the change.
316
- * - **Unsubscribed.** The subscription was cancelled while the fetch was in
317
- * flight. The `has(subscriptionId)` check the debounced refetches ran
318
- * *before* the await cannot answer this; only a check after it can.
319
- * - **Replaced.** The same id can name a *different* subscription by the
320
- * time a fetch lands — a re-subscribe overwrites the map entry, and the
321
- * old filter's rows would be delivered to the new subscriber.
322
- *
323
- * The last two are identity, not presence: the map has to still hold *this
324
- * exact object*, not merely something under this id.
325
- */
326
- private beginDelivery(subscriptionId: string, subscription: Subscription): () => boolean {
327
- const seq = ++subscription.started;
328
- return () => {
329
- if (this._subscriptions.get(subscriptionId) !== subscription) return false;
330
- if (seq <= subscription.delivered) return false;
331
- subscription.delivered = seq;
332
- return true;
333
- };
334
- }
335
-
336
- // Add public method to register DataDriver subscriptions
337
- registerDataDriverSubscription(subscriptionId: string, subscription: {
338
- clientId: string;
339
- type: "collection" | "single";
340
- path: string;
341
- id?: string | number;
342
- collectionRequest?: StoredCollectionRequest;
343
- authContext?: SubscriptionAuthContext;
344
- }) {
345
- this.debugLog("📋 [RealtimeService] Registering DataDriver subscription:", subscriptionId, subscription.authContext ? "(with auth)" : "(no auth)");
346
- this._subscriptions.set(subscriptionId, { ...subscription, started: 0, delivered: 0 });
347
- }
348
-
349
- // Add callback management methods
350
- addSubscriptionCallback(subscriptionId: string, callback: (data: Record<string, unknown>[] | Record<string, unknown> | null) => void) {
351
- this.debugLog("📋 [RealtimeService] Adding callback for subscription:", subscriptionId);
352
- this.subscriptionCallbacks.set(subscriptionId, callback);
353
- }
354
-
355
- removeSubscriptionCallback(subscriptionId: string) {
356
- this.debugLog("📋 [RealtimeService] Removing callback for subscription:", subscriptionId);
357
- this.subscriptionCallbacks.delete(subscriptionId);
358
- }
359
-
360
- // =============================================================================
361
- // RealtimeProvider Interface Methods
362
- // =============================================================================
363
-
364
- /**
365
- * Subscribe to collection changes (RealtimeProvider interface)
366
- */
367
- subscribeToCollection(
368
- subscriptionId: string,
369
- config: CollectionSubscriptionConfig,
370
- callback?: (rows: Record<string, unknown>[]) => void
371
- ): void {
372
- this._subscriptions.set(subscriptionId, {
373
- clientId: config.clientId,
374
- type: "collection",
375
- path: config.path,
376
- collectionRequest: {
377
- filter: config.filter as Record<string, unknown> | undefined,
378
- orderBy: config.orderBy,
379
- order: config.order,
380
- limit: config.limit,
381
- startAfter: config.startAfter as Record<string, unknown> | undefined,
382
- databaseId: config.databaseId,
383
- searchString: config.searchString,
384
- searchExplain: config.searchExplain
385
- },
386
- started: 0,
387
- delivered: 0
388
- });
389
-
390
- if (callback) {
391
- this.subscriptionCallbacks.set(subscriptionId, callback as (data: Record<string, unknown>[] | Record<string, unknown> | null) => void);
392
- }
393
- }
394
-
395
- /**
396
- * Subscribe to single row changes (RealtimeProvider interface)
397
- */
398
- subscribeToOne(
399
- subscriptionId: string,
400
- config: SingleSubscriptionConfig,
401
- callback?: (row: Record<string, unknown> | null) => void
402
- ): void {
403
- this._subscriptions.set(subscriptionId, {
404
- clientId: config.clientId,
405
- type: "single",
406
- path: config.path,
407
- id: config.id,
408
- started: 0,
409
- delivered: 0
410
- });
411
-
412
- if (callback) {
413
- this.subscriptionCallbacks.set(subscriptionId, callback as (data: Record<string, unknown>[] | Record<string, unknown> | null) => void);
414
- }
415
- }
416
-
417
- /**
418
- * Unsubscribe from a subscription (RealtimeProvider interface)
419
- */
420
- unsubscribe(subscriptionId: string): void {
421
- this._subscriptions.delete(subscriptionId);
422
- this.subscriptionCallbacks.delete(subscriptionId);
423
- }
424
-
425
- // =============================================================================
426
- // WebSocket Client Management
427
- // =============================================================================
428
-
429
- addClient(clientId: string, ws: WebSocket) {
430
- this.clients.set(clientId, ws);
431
-
432
- ws.on("close", () => {
433
- this.removeClient(clientId);
434
- });
435
-
436
- ws.on("error", (error) => {
437
- logger.error("WebSocket error for client", { detail: clientId, error });
438
- this.removeClient(clientId);
439
- });
440
- }
441
-
442
- // Public method to handle messages from external sources (like main WebSocket handler)
443
- async handleClientMessage(clientId: string, message: WebSocketMessage, authContext?: SubscriptionAuthContext) {
444
- await this.handleMessage(clientId, message, authContext);
445
- }
446
-
447
- async removeClient(clientId: string) {
448
- this.clients.delete(clientId);
449
-
450
- // Remove all subscriptions, callbacks, and pending refetch timers for this client
451
- for (const [subscriptionId, subscription] of this._subscriptions.entries()) {
452
- if (subscription.clientId === clientId) {
453
- this._subscriptions.delete(subscriptionId);
454
- this.subscriptionCallbacks.delete(subscriptionId);
455
-
456
- // Cancel any pending debounced refetch timers
457
- for (const prefix of ["ws_", "drv_", "wse_", "drve_"]) {
458
- const key = `${prefix}${subscriptionId}`;
459
- const timer = this.refetchTimers.get(key);
460
- if (timer) { clearTimeout(timer); this.refetchTimers.delete(key); }
461
- }
462
- }
463
- }
464
-
465
- // The shared rows go before the announcement, not after it. Every
466
- // `removePresence` below publishes a departure — to local members and
467
- // over the bus — while `sendPresenceState` answers a *newly arriving*
468
- // subscriber from this table. Announcing first leaves a window where the
469
- // table still lists someone who has left: a client hydrating inside it is
470
- // handed the ghost, and the diff that would have corrected it was
471
- // broadcast before that client existed, so it holds the ghost until the
472
- // TTL sweep rather than for the length of one statement.
473
- //
474
- // One statement for every channel the client was in, rather than one per
475
- // channel below — a disconnect is the common case, not a rare one. The
476
- // subscription and timer cleanup above stays synchronous on purpose: it
477
- // is what leaks if the database is slow, and it owes nothing to the
478
- // shared table. With no bus configured this is a no-op that never awaits
479
- // a query.
480
- await this.presenceStoreOp(() => this.presenceStore!.removeClient(clientId), "client removal");
481
-
482
- // Remove from all broadcast channels
483
- for (const [channel, members] of this.channels.entries()) {
484
- if (members.has(clientId)) {
485
- members.delete(clientId);
486
- this.removePresence(clientId, channel, { skipStore: true });
487
- if (members.size === 0) this.channels.delete(channel);
488
- }
489
- }
490
-
491
- // Remove from all presence channels
492
- for (const [channel] of this.presence) {
493
- this.removePresence(clientId, channel, { skipStore: true });
494
- }
495
- }
496
-
497
- private async handleMessage(clientId: string, message: WebSocketMessage, authContext?: SubscriptionAuthContext) {
498
- const payload = message.payload as Record<string, unknown> | undefined;
499
- switch (message.type) {
500
- case "subscribe_collection":
501
- await this.handleCollectionSubscription(clientId, message.payload as RealTimeListenCollectionProps, authContext);
502
- break;
503
- case "subscribe_one":
504
- await this.handleEntitySubscription(clientId, message.payload as RealTimeListenEntityProps, authContext);
505
- break;
506
- case "unsubscribe":
507
- await this.handleUnsubscribe(clientId, message.subscriptionId!);
508
- break;
509
-
510
- // ── Broadcast Channels & Presence ──
511
- //
512
- // One arm for all of them, because every one has to pass the same
513
- // gate and a switch with seven arms is a place to forget it once.
514
- // See `handleChannelMessage`.
515
- case "join_channel":
516
- case "leave_channel":
517
- case "broadcast":
518
- case "channel_history":
519
- case "presence_track":
520
- case "presence_untrack":
521
- case "presence_state":
522
- await this.handleChannelMessage(clientId, message.type, payload, authContext);
523
- break;
524
-
525
- default:
526
- this.sendError(clientId, "Unknown message type " + message.type, message.subscriptionId);
527
- }
528
- }
529
-
530
- private async handleCollectionSubscription(clientId: string, request: RealTimeListenCollectionProps, authContext?: SubscriptionAuthContext) {
531
- const subscriptionId = request.subscriptionId;
532
-
533
- try {
534
- // Early validation: ensure the requested collection exists in the registry
535
- const collection = this.registry.getCollectionByPath(request.path);
536
- if (!collection) {
537
- const registered = this.registry.getCollections().map(c => c.slug).join(", ");
538
- const msg = `Collection not found: '${request.path}'. Registered: [${registered}]`;
539
- logger.error(`[RealtimeService] ${msg}`);
540
- this.sendError(clientId, msg, subscriptionId);
541
- return;
542
- }
543
-
544
- // A vector search cannot be served here, and the parameter used to
545
- // be read for one thing only — the limit default below — and then
546
- // dropped: the stored request carries no `vectorSearch` and the
547
- // refetch has no branch for one. So `.vectorSearch(…).listen()`
548
- // delivered an ordinary `id DESC` listing, with no `_distance` and
549
- // no error, forever. Refusing says what the silence did not.
550
- if (request.vectorSearch) {
551
- const msg =
552
- "Realtime subscriptions do not support vector search: a subscription is re-run on every " +
553
- "matching write, and nothing here computes distances. Use `.vectorSearch(...).find()` for " +
554
- "the query, and subscribe without it if you need live updates.";
555
- logger.warn(`[RealtimeService] ${msg}`);
556
- this.sendError(clientId, msg, subscriptionId, "VECTOR_SEARCH_NOT_LIVE");
557
- return;
558
- }
559
-
560
- // Bound the client-supplied limit with the SAME guarantee the REST
561
- // ingress applies (`resolveClientListLimit`): default an absent
562
- // limit by mode, refuse one above the ceiling. A subscription is
563
- // re-fetched on every matching write, so an unbounded one is a DoS
564
- // amplified per write — resolve it once and reuse for the stored
565
- // request and the initial fetch.
566
- //
567
- // Refusing matters more here than on the REST route: a
568
- // `collection_update` frame carries rows and nothing else — no
569
- // `total`, no `hasMore` — so a subscriber handed a quietly smaller
570
- // page has no way at all to learn it is not seeing the collection.
571
- let boundedLimit: number;
572
- try {
573
- boundedLimit = resolveClientListLimit(request.limit);
574
- } catch (e) {
575
- if (!(e instanceof ListLimitError)) throw e;
576
- logger.warn(`[RealtimeService] Refused subscription to '${request.path}': ${e.message}`);
577
- this.sendError(clientId, e.message, subscriptionId, "INVALID_LIMIT");
578
- return;
579
- }
580
-
581
- // The sort arrives as whatever JSON the client put in the frame, so
582
- // its *shape* is checked here the way the REST ingress checks the
583
- // query parameter. Unchecked, a malformed entry reads as a field
584
- // name that resolves to no column, and under lenient unknown-field
585
- // handling the subscription then streams rows in no order at all
586
- // while reporting nothing wrong.
587
- let orderBy: OrderByTuple[] | undefined;
588
- try {
589
- orderBy = parseOrderBySpecStrict(request.orderBy, request.order);
590
- } catch (e) {
591
- if (!(e instanceof OrderBySpecError)) throw e;
592
- logger.warn(`[RealtimeService] Refused subscription to '${request.path}': ${e.message}`);
593
- this.sendError(clientId, e.message, subscriptionId, e.code);
594
- return;
595
- }
596
-
597
- // Store subscription with full request parameters and auth context for RLS
598
- const subscription: Subscription = {
599
- clientId,
600
- type: "collection",
601
- path: request.path,
602
- collectionRequest: {
603
- filter: request.filter,
604
- logical: request.logical,
605
- orderBy,
606
- order: request.order,
607
- limit: boundedLimit,
608
- offset: request.offset,
609
- startAfter: request.startAfter as Record<string, unknown> | undefined,
610
- databaseId: request.collection?.databaseId,
611
- searchString: request.searchString,
612
- searchExplain: request.searchExplain
613
- },
614
- authContext,
615
- started: 0,
616
- delivered: 0
617
- };
618
- this._subscriptions.set(subscriptionId, subscription);
619
-
620
- // The subscription is registered before this fetch runs, so a write
621
- // arriving in that window starts a refetch of its own — with nothing
622
- // ordering the two. Claim a slot first: this fetch is the oldest, so
623
- // if the refetch answers first, this one no longer delivers.
624
- const canDeliver = this.beginDelivery(subscriptionId, subscription);
625
-
626
- // Send initial data. Built from the request the subscription just
627
- // stored, so the first answer and every refetch after it cannot
628
- // describe different queries.
629
- const rows = await this.fetchCollectionWithAuth(
630
- request.path,
631
- subscription.collectionRequest!,
632
- authContext
633
- );
634
-
635
- if (canDeliver()) {
636
- this.sendCollectionUpdate(clientId, subscriptionId, rows, request.path);
637
- }
638
-
639
- } catch (error) {
640
- const sanitized = sanitizeErrorForClient(error, request.path);
641
- this.sendError(clientId, sanitized.message, subscriptionId, sanitized.code);
642
- }
643
- }
644
-
645
- private async handleEntitySubscription(clientId: string, request: RealTimeListenEntityProps, authContext?: SubscriptionAuthContext) {
646
- const subscriptionId = request.subscriptionId;
647
-
648
- try {
649
- // Early validation: ensure the requested collection exists in the registry
650
- const collection = this.registry.getCollectionByPath(request.path);
651
- if (!collection) {
652
- const registered = this.registry.getCollections().map(c => c.slug).join(", ");
653
- const msg = `Collection not found: '${request.path}'. Registered: [${registered}]`;
654
- logger.error(`[RealtimeService] ${msg}`);
655
- this.sendError(clientId, msg, subscriptionId);
656
- return;
657
- }
658
-
659
- // Store subscription in memory with auth context for RLS
660
- const subscription: Subscription = {
661
- clientId,
662
- type: "single",
663
- path: request.path,
664
- id: request.id,
665
- authContext,
666
- started: 0,
667
- delivered: 0
668
- };
669
- this._subscriptions.set(subscriptionId, subscription);
670
-
671
- // Same race as the collection case: a write landing between the
672
- // registration above and this fetch starts a refetch that can answer
673
- // first, and this one must not overwrite it afterwards.
674
- const canDeliver = this.beginDelivery(subscriptionId, subscription);
675
-
676
- // Send initial data
677
- const row = await this.fetchEntityWithAuth(
678
- request.path,
679
- String(request.id),
680
- authContext
681
- );
682
-
683
- if (canDeliver()) {
684
- this.sendSingleUpdate(clientId, subscriptionId, row || null);
685
- }
686
-
687
- } catch (error) {
688
- const sanitized = sanitizeErrorForClient(error, request.path);
689
- this.sendError(clientId, sanitized.message, subscriptionId, sanitized.code);
690
- }
691
- }
692
-
693
- private async handleUnsubscribe(_clientId: string, subscriptionId: string) {
694
- this._subscriptions.delete(subscriptionId);
695
- this.subscriptionCallbacks.delete(subscriptionId);
696
- // Cancel any pending debounced refetch
697
- for (const prefix of ["ws_", "drv_", "wse_", "drve_"]) {
698
- const key = `${prefix}${subscriptionId}`;
699
- const timer = this.refetchTimers.get(key);
700
- if (timer) { clearTimeout(timer); this.refetchTimers.delete(key); }
701
- }
702
- }
703
-
704
- /**
705
- * Enhanced notification method that handles nested relation updates.
706
- * @param broadcast When true (default), also sends a pg_notify so other instances
707
- * pick up the change. Set to false when handling an incoming
708
- * cross-instance notification to avoid infinite loops.
709
- * @param origin `"app"` (default) — a Rebase-API mutation on this instance;
710
- * `"cdc"` — a database-level change observed via CDC (any writer,
711
- * any instance). The origin drives de-duplication: an app emit
712
- * records the change so this instance can suppress the matching
713
- * CDC echo, while an unmatched CDC event is delivered normally.
714
- */
715
- async notifyUpdate(path: string, id: string, row: Record<string, unknown> | null, databaseId?: string, broadcast = true, origin: "app" | "cdc" = "app") {
716
- this.debugLog("🔔 [RealtimeService] notifyUpdate called for path:", path, "id:", id, "isDelete:", row === null, "origin:", origin);
717
-
718
- // De-duplicate against database-level CDC. The app path (a mutation made
719
- // through the Rebase API) fans out locally AND, once CDC is active, the
720
- // same committed change is echoed back to this instance via the WAL /
721
- // trigger stream. Record app emits so we can drop that echo here; deliver
722
- // any CDC event we did not originate (external writes, other instances).
723
- if (this.cdcActive) {
724
- const key = this.dedupKey(path, id, databaseId);
725
- if (origin === "cdc") {
726
- if (this.consumeAppEmit(key)) {
727
- this.debugLog("🔁 [RealtimeService] Suppressing CDC echo of local app mutation:", key);
728
- return;
729
- }
730
- } else {
731
- this.markAppEmit(key);
732
- }
733
- }
734
-
735
- // Get all paths that need to be notified - the direct path plus any parent paths
736
- const pathsToNotify = [path];
737
-
738
- // If this is a nested relation path (like "posts/70/tags"), also notify parent paths
739
- if (path.includes("/") && path.split("/").length > 1) {
740
- const parentPaths = this.getParentPaths(path);
741
- pathsToNotify.push(...parentPaths);
742
- this.debugLog(`🔗 [RealtimeService] Nested path detected. Will notify paths: ${pathsToNotify.join(", ")}`);
743
- }
744
-
745
- // Process each path that needs notification
746
- for (const notifyPath of pathsToNotify) {
747
- await this.notifyPathUpdate(notifyPath, path, id, row, databaseId);
748
- }
749
-
750
- // Broadcast to other instances via pg_notify (only for local mutations).
751
- // When CDC is active it IS the cross-instance channel — every instance
752
- // observes every commit through the change stream — so the legacy
753
- // per-mutation broadcast is redundant (and would double-deliver). Skip it.
754
- if (broadcast && this.broadcasting && !this.cdcActive) {
755
- try {
756
- await this.broadcastChange(path, id, databaseId);
757
- } catch (err) {
758
- logger.error("❌ [RealtimeService] Failed to broadcast change via pg_notify", { error: err });
759
- }
760
- }
761
-
762
- this.debugLog("🔔 [RealtimeService] notifyUpdate completed for path:", path);
763
- }
764
-
765
- /**
766
- * Notify subscriptions for a specific path.
767
- *
768
- * **A subscriber only ever receives rows re-read under its own scope.**
769
- * `row` is used to decide *that* something changed, never to say *what* —
770
- * every delivery below goes through a refetch that binds the subscription's
771
- * own auth context.
772
- *
773
- * It used to be conditional. The CDC path already did the right thing: it
774
- * discards the captured tuple and emits `{_rebase_invalidated: true}`, and
775
- * that marker selected the refetch branch. But the marker is produced in
776
- * exactly two places, and the *other* side of each branch here shipped the
777
- * row it was handed straight to the socket. Two of the three entry paths
778
- * took that side — every API mutation (`PostgresBackendDriver.save` passes
779
- * the row it just wrote, read under the **writer's** scope) and the legacy
780
- * cross-instance LISTEN handler (which re-reads on the owner connection,
781
- * bypassing RLS altogether). Path matching was the only filter applied: the
782
- * subscription's own `filter`/`logical` was never evaluated, and any
783
- * `afterRead` redaction was the writer's rather than the reader's.
784
- *
785
- * A single-row subscription was the sharpest case. `subscribe_one` on a row
786
- * RLS denies is accepted and answered `null`; the next update then pushed
787
- * the full row with no later correction. The collection variant was merely
788
- * papered over ~300 ms later by the debounced refetch — after the bytes had
789
- * already reached the browser.
790
- *
791
- * The same defect was found and fixed on the Mongo driver in `065e2b615`
792
- * (see `packages/server-mongo/test/realtime-authorization.test.ts`); this is
793
- * the Postgres half, stated as one rule rather than three patched branches.
794
- *
795
- * The cost is the instant row-level patch that used to precede the refetch:
796
- * cross-tab feedback now waits for the debounce. That is the price of not
797
- * being able to know, without asking the database as this subscriber,
798
- * whether this subscriber may see the row at all.
799
- */
800
- private async notifyPathUpdate(notifyPath: string, originalPath: string, id: string, row: Record<string, unknown> | null, _databaseId?: string) {
801
- this.debugLog(`📡 [RealtimeService] Notifying path: ${notifyPath} (original: ${originalPath})`);
802
-
803
- // Find all relevant subscriptions for this specific path
804
- const allSubscriptions = Array.from(this._subscriptions.entries()).filter(([, sub]) => {
805
- const isPathMatch = sub.path === notifyPath;
806
-
807
- // For row subscriptions, check if the id matches (only for exact path matches)
808
- if (sub.type === "single") {
809
- return isPathMatch && (notifyPath === originalPath ? sub.id === id : true);
810
- }
811
- // For collection subscriptions, it's always relevant if the path matches
812
- if (sub.type === "collection") {
813
- return isPathMatch;
814
- }
815
- return false;
816
- });
817
-
818
- this.debugLog(`📡 [RealtimeService] Found ${allSubscriptions.length} subscriptions for path: ${notifyPath}`);
819
-
820
- // Separate WebSocket subscriptions from DataDriver callback subscriptions
821
- const webSocketSubscriptions = allSubscriptions.filter(([, sub]) =>
822
- sub.clientId !== "driver" && this.clients.has(sub.clientId)
823
- );
824
-
825
- const driverSubscriptions = allSubscriptions.filter(([subscriptionId, sub]) =>
826
- sub.clientId === "driver" && this.subscriptionCallbacks.has(subscriptionId)
827
- );
828
-
829
- // Handle WebSocket subscriptions
830
- for (const [subscriptionId, subscription] of webSocketSubscriptions) {
831
- try {
832
- if (subscription.type === "single" && notifyPath === originalPath) {
833
- this.debouncedSingleRefetch(subscriptionId, notifyPath, id, subscription);
834
- } else if (subscription.type === "collection" && subscription.collectionRequest) {
835
- this.debouncedCollectionRefetch(subscriptionId, notifyPath, subscription);
836
- }
837
- } catch (error) {
838
- const sanitized = sanitizeErrorForClient(error, notifyPath);
839
- this.sendError(subscription.clientId, sanitized.message, subscriptionId, sanitized.code);
840
- }
841
- }
842
-
843
- // Handle DataDriver callback subscriptions
844
- for (const [subscriptionId, subscription] of driverSubscriptions) {
845
- try {
846
- const callback = this.subscriptionCallbacks.get(subscriptionId);
847
- if (!callback) continue;
848
-
849
- if (subscription.type === "single" && notifyPath === originalPath) {
850
- this.debouncedSingleDriverRefetch(subscriptionId, notifyPath, id, subscription, callback);
851
- } else if (subscription.type === "collection" && subscription.collectionRequest) {
852
- // Debounce collection refetches for DataDriver subscriptions too
853
- this.debouncedDriverRefetch(subscriptionId, notifyPath, subscription, callback);
854
- }
855
- } catch (error) {
856
- logger.error(`❌ [RealtimeService] Error processing DataDriver subscription ${subscriptionId}`, { error: error });
857
- }
858
- }
859
- }
860
-
861
- /**
862
- * Debounce a collection refetch for a WebSocket subscription.
863
- * Coalesces rapid row mutations into a single database query.
864
- */
865
- private debouncedCollectionRefetch(
866
- subscriptionId: string,
867
- notifyPath: string,
868
- subscription: Subscription
869
- ) {
870
- const timerKey = `ws_${subscriptionId}`;
871
- const existing = this.refetchTimers.get(timerKey);
872
- if (existing) clearTimeout(existing);
873
-
874
- this.refetchTimers.set(timerKey, setTimeout(async () => {
875
- this.refetchTimers.delete(timerKey);
876
- // Cheap bail before spending a query: the client may have
877
- // disconnected, or re-subscribed under the same id. It is only an
878
- // optimisation — `canDeliver()` after the await is what makes the
879
- // delivery safe, because the same things can happen *during* it.
880
- if (this._subscriptions.get(subscriptionId) !== subscription) return;
881
- // Claimed here rather than when the timer was scheduled: the
882
- // debounce coalesces, and no work exists to order until it fires.
883
- const canDeliver = this.beginDelivery(subscriptionId, subscription);
884
- try {
885
- const rows = await this.fetchCollectionWithAuth(notifyPath, subscription.collectionRequest!, subscription.authContext);
886
- if (canDeliver()) {
887
- this.sendCollectionUpdate(subscription.clientId, subscriptionId, rows, notifyPath);
888
- }
889
- } catch (error) {
890
- const sanitized = sanitizeErrorForClient(error, notifyPath);
891
- this.sendError(subscription.clientId, sanitized.message, subscriptionId, sanitized.code);
892
- }
893
- }, RealtimeService.REFETCH_DEBOUNCE_MS));
894
- }
895
-
896
- /**
897
- * Debounce a collection refetch for a DataDriver callback subscription.
898
- */
899
- private debouncedDriverRefetch(
900
- subscriptionId: string,
901
- notifyPath: string,
902
- subscription: Subscription,
903
- callback: (data: Record<string, unknown>[] | Record<string, unknown> | null) => void
904
- ) {
905
- const timerKey = `drv_${subscriptionId}`;
906
- const existing = this.refetchTimers.get(timerKey);
907
- if (existing) clearTimeout(existing);
908
-
909
- this.refetchTimers.set(timerKey, setTimeout(async () => {
910
- this.refetchTimers.delete(timerKey);
911
- if (this._subscriptions.get(subscriptionId) !== subscription) return;
912
- const canDeliver = this.beginDelivery(subscriptionId, subscription);
913
- try {
914
- const rows = await this.fetchCollectionWithAuth(notifyPath, subscription.collectionRequest!, subscription.authContext);
915
- if (canDeliver()) callback(rows);
916
- } catch (error) {
917
- logger.error(`❌ [RealtimeService] Error in debounced driver refetch for ${subscriptionId}`, { error: error });
918
- }
919
- }, RealtimeService.REFETCH_DEBOUNCE_MS));
920
- }
921
-
922
- /**
923
- * Fetch a collection with optional RLS auth context.
924
- * When authContext is provided, the fetch runs inside a transaction
925
- * with set_config calls so PostgreSQL RLS policies are enforced.
926
- */
927
- private async fetchCollectionWithAuth(
928
- notifyPath: string,
929
- collectionRequest: StoredCollectionRequest,
930
- authContext?: SubscriptionAuthContext
931
- ): Promise<Record<string, unknown>[]> {
932
- if (this.driver) {
933
- const collection = this.registry.getCollectionByPath(notifyPath);
934
- const fetchFn = async () => this.driver!.fetchCollection({
935
- path: notifyPath,
936
- collection: collection,
937
- filter: collectionRequest.filter as FetchCollectionProps["filter"],
938
- logical: collectionRequest.logical,
939
- orderBy: collectionRequest.orderBy,
940
- order: collectionRequest.order,
941
- limit: collectionRequest.limit,
942
- offset: collectionRequest.offset,
943
- startAfter: collectionRequest.startAfter,
944
- searchString: collectionRequest.searchString,
945
- searchExplain: collectionRequest.searchExplain
946
- });
947
-
948
- // Always wrap in a transaction with session vars, defaulting to anonymous context if missing.
949
- // Refetches are reads: apply the same GUCs + reader-role downgrade as the
950
- // driver's read path, so realtime cannot leak rows the initial fetch hid.
951
- const activeAuth = authContext || { uid: ANONYMOUS_USER_ID,
952
- roles: ["anon"] };
953
- return await this.db.transaction(async (tx) => {
954
- await applyAuthContext(tx, { uid: activeAuth.uid, roles: activeAuth.roles }, this.rlsUserRole);
955
- const txEntityService = new DataService(tx, this.registry);
956
- let fetchedEntities;
957
- if (collectionRequest.searchString) {
958
- fetchedEntities = await txEntityService.searchRows(
959
- notifyPath,
960
- collectionRequest.searchString,
961
- {
962
- filter: collectionRequest.filter as FilterValues<string>,
963
- // The subscription stored a group; the search branch
964
- // did not pass it on, so a filtered live search
965
- // widened to every row matching the text.
966
- logical: collectionRequest.logical,
967
- orderBy: collectionRequest.orderBy,
968
- order: collectionRequest.order,
969
- limit: collectionRequest.limit,
970
- databaseId: collectionRequest.databaseId,
971
- searchExplain: collectionRequest.searchExplain
972
- }
973
- );
974
- } else {
975
- fetchedEntities = await txEntityService.fetchCollection(notifyPath, {
976
- filter: collectionRequest.filter as FilterValues<string>,
977
- logical: collectionRequest.logical,
978
- orderBy: collectionRequest.orderBy,
979
- order: collectionRequest.order,
980
- limit: collectionRequest.limit,
981
- offset: collectionRequest.offset,
982
- startAfter: collectionRequest.startAfter,
983
- databaseId: collectionRequest.databaseId
984
- });
985
- }
986
-
987
- // Re-apply `afterRead` lifecycle hooks to ensure consistent data structures
988
- // between the initial driver fetch and this RLS-bound refetch.
989
- const registryCollection = this.registry.getCollectionByPath(notifyPath);
990
- const resolvedCollection = collection ? { ...collection,
991
- ...registryCollection } as CollectionConfig : registryCollection as CollectionConfig;
992
-
993
- const callbacks = resolvedCollection?.callbacks;
994
- const globalCallbacks = this.registry?.getGlobalCallbacks();
995
- const propertyCallbacks = resolvedCollection?.properties ? buildPropertyCallbacks(resolvedCollection.properties) : undefined;
996
-
997
- if (globalCallbacks?.afterRead || callbacks?.afterRead || propertyCallbacks?.afterRead) {
998
- const contextForCallback = {
999
- user: { uid: activeAuth.uid,
1000
- roles: activeAuth.roles },
1001
- driver: this.driver,
1002
- data: (this.driver && "data" in this.driver) ? (this.driver as DataDriverWithData).data : undefined
1003
- } as unknown as RebaseCallContext;
1004
-
1005
- return await Promise.all(fetchedEntities.map(async (fetchedRow) => {
1006
- let processedEntity = fetchedRow;
1007
- // 1. Global callbacks first
1008
- if (globalCallbacks?.afterRead) {
1009
- processedEntity = await globalCallbacks.afterRead({
1010
- collection: resolvedCollection,
1011
- path: notifyPath,
1012
- row: processedEntity,
1013
- context: contextForCallback
1014
- }) ?? processedEntity;
1015
- }
1016
- // 2. Collection callbacks second
1017
- if (callbacks?.afterRead) {
1018
- processedEntity = await callbacks.afterRead({
1019
- collection: resolvedCollection,
1020
- path: notifyPath,
1021
- row: processedEntity,
1022
- context: contextForCallback
1023
- }) ?? processedEntity;
1024
- }
1025
- // 3. Property callbacks third
1026
- if (propertyCallbacks?.afterRead) {
1027
- processedEntity = await propertyCallbacks.afterRead({
1028
- collection: resolvedCollection,
1029
- path: notifyPath,
1030
- row: processedEntity,
1031
- context: contextForCallback
1032
- }) ?? processedEntity;
1033
- }
1034
- return processedEntity;
1035
- }));
1036
- }
1037
-
1038
- return fetchedEntities;
1039
- });
1040
- }
1041
-
1042
- // No driver — use dataService directly (no auth wrapping possible).
1043
- // The `logical` group is carried here as well: this branch answers the
1044
- // same subscription as the one above, and a fallback that drops a
1045
- // condition returns *more* rows than the path it stands in for.
1046
- if (collectionRequest.searchString) {
1047
- return await this.dataService.searchRows(
1048
- notifyPath,
1049
- collectionRequest.searchString,
1050
- {
1051
- filter: collectionRequest.filter as FilterValues<string>,
1052
- logical: collectionRequest.logical,
1053
- orderBy: collectionRequest.orderBy,
1054
- order: collectionRequest.order,
1055
- limit: collectionRequest.limit,
1056
- databaseId: collectionRequest.databaseId,
1057
- searchExplain: collectionRequest.searchExplain
1058
- }
1059
- );
1060
- }
1061
- return await this.dataService.fetchCollection(notifyPath, {
1062
- filter: collectionRequest.filter as FilterValues<string>,
1063
- logical: collectionRequest.logical,
1064
- orderBy: collectionRequest.orderBy,
1065
- order: collectionRequest.order,
1066
- limit: collectionRequest.limit,
1067
- offset: collectionRequest.offset,
1068
- startAfter: collectionRequest.startAfter,
1069
- databaseId: collectionRequest.databaseId
1070
- });
1071
- }
1072
-
1073
- /**
1074
- * Debounce an row refetch for a WebSocket subscription.
1075
- */
1076
- private debouncedSingleRefetch(
1077
- subscriptionId: string,
1078
- notifyPath: string,
1079
- id: string,
1080
- subscription: Subscription
1081
- ) {
1082
- const timerKey = `wse_${subscriptionId}`;
1083
- const existing = this.refetchTimers.get(timerKey);
1084
- if (existing) clearTimeout(existing);
1085
-
1086
- this.refetchTimers.set(timerKey, setTimeout(async () => {
1087
- this.refetchTimers.delete(timerKey);
1088
- if (this._subscriptions.get(subscriptionId) !== subscription) return;
1089
- const canDeliver = this.beginDelivery(subscriptionId, subscription);
1090
- try {
1091
- const row = await this.fetchEntityWithAuth(notifyPath, id, subscription.authContext);
1092
- if (canDeliver()) {
1093
- this.sendSingleUpdate(subscription.clientId, subscriptionId, row || null);
1094
- }
1095
- } catch (error) {
1096
- const sanitized = sanitizeErrorForClient(error, notifyPath);
1097
- this.sendError(subscription.clientId, sanitized.message, subscriptionId, sanitized.code);
1098
- }
1099
- }, RealtimeService.REFETCH_DEBOUNCE_MS));
1100
- }
1101
-
1102
- /**
1103
- * Debounce an row refetch for a Driver callback subscription.
1104
- */
1105
- private debouncedSingleDriverRefetch(
1106
- subscriptionId: string,
1107
- notifyPath: string,
1108
- id: string,
1109
- subscription: Subscription,
1110
- callback: (data: Record<string, unknown>[] | Record<string, unknown> | null) => void
1111
- ) {
1112
- const timerKey = `drve_${subscriptionId}`;
1113
- const existing = this.refetchTimers.get(timerKey);
1114
- if (existing) clearTimeout(existing);
1115
-
1116
- this.refetchTimers.set(timerKey, setTimeout(async () => {
1117
- this.refetchTimers.delete(timerKey);
1118
- if (this._subscriptions.get(subscriptionId) !== subscription) return;
1119
- const canDeliver = this.beginDelivery(subscriptionId, subscription);
1120
- try {
1121
- const row = await this.fetchEntityWithAuth(notifyPath, id, subscription.authContext);
1122
- if (canDeliver()) callback(row || null);
1123
- } catch (error) {
1124
- logger.error(`❌ [RealtimeService] Error in debounced row driver refetch for ${subscriptionId}`, { error: error });
1125
- }
1126
- }, RealtimeService.REFETCH_DEBOUNCE_MS));
1127
- }
1128
-
1129
- /**
1130
- * Fetch a single row with optional RLS auth context.
1131
- */
1132
- private async fetchEntityWithAuth(
1133
- notifyPath: string,
1134
- id: string | number,
1135
- authContext?: SubscriptionAuthContext
1136
- ): Promise<Record<string, unknown> | undefined> {
1137
- if (this.driver) {
1138
- const collection = this.registry.getCollectionByPath(notifyPath);
1139
- const fetchFn = async () => this.driver!.fetchOne({
1140
- path: notifyPath,
1141
- id,
1142
- collection
1143
- });
1144
-
1145
- // Always wrap in a transaction with session vars, defaulting to anonymous context if missing.
1146
- // Same read isolation as collection refetches: GUCs + reader-role downgrade.
1147
- const activeAuth = authContext || { uid: ANONYMOUS_USER_ID,
1148
- roles: ["anon"] };
1149
- return await this.db.transaction(async (tx) => {
1150
- await applyAuthContext(tx, { uid: activeAuth.uid, roles: activeAuth.roles }, this.rlsUserRole);
1151
- const txEntityService = new DataService(tx, this.registry);
1152
- let processedEntity = await txEntityService.fetchOne(notifyPath, id, collection?.databaseId);
1153
-
1154
- if (processedEntity) {
1155
- const registryCollection = this.registry.getCollectionByPath(notifyPath);
1156
- const resolvedCollection = collection ? { ...collection,
1157
- ...registryCollection } as CollectionConfig : registryCollection as CollectionConfig;
1158
-
1159
- const callbacks = resolvedCollection?.callbacks;
1160
- const globalCallbacks = this.registry?.getGlobalCallbacks();
1161
- const propertyCallbacks = resolvedCollection?.properties ? buildPropertyCallbacks(resolvedCollection.properties) : undefined;
1162
-
1163
- if (globalCallbacks?.afterRead || callbacks?.afterRead || propertyCallbacks?.afterRead) {
1164
- const contextForCallback = {
1165
- user: { uid: activeAuth.uid,
1166
- roles: activeAuth.roles },
1167
- driver: this.driver,
1168
- data: (this.driver && "data" in this.driver) ? (this.driver as DataDriverWithData).data : undefined
1169
- } as unknown as RebaseCallContext;
1170
-
1171
- // 1. Global callbacks first
1172
- if (globalCallbacks?.afterRead) {
1173
- processedEntity = await globalCallbacks.afterRead({
1174
- collection: resolvedCollection,
1175
- path: notifyPath,
1176
- row: processedEntity,
1177
- context: contextForCallback
1178
- }) ?? processedEntity;
1179
- }
1180
- // 2. Collection callbacks second
1181
- if (callbacks?.afterRead) {
1182
- processedEntity = await callbacks.afterRead({
1183
- collection: resolvedCollection,
1184
- path: notifyPath,
1185
- row: processedEntity,
1186
- context: contextForCallback
1187
- }) ?? processedEntity;
1188
- }
1189
- // 3. Property callbacks third
1190
- if (propertyCallbacks?.afterRead) {
1191
- processedEntity = await propertyCallbacks.afterRead({
1192
- collection: resolvedCollection,
1193
- path: notifyPath,
1194
- row: processedEntity,
1195
- context: contextForCallback
1196
- }) ?? processedEntity;
1197
- }
1198
- }
1199
- }
1200
-
1201
- return processedEntity;
1202
- });
1203
- }
1204
-
1205
- return await this.dataService.fetchOne(notifyPath, id);
1206
- }
1207
-
1208
- private sendCollectionUpdate(clientId: string, subscriptionId: string, rows: Record<string, unknown>[], path: string) {
1209
- const message: CollectionUpdateMessage = {
1210
- type: "collection_update",
1211
- subscriptionId,
1212
- rows: rows,
1213
- pks: this.primaryKeysForPath(path)
1214
- };
1215
- this.sendMessage(clientId, message);
1216
- }
1217
-
1218
- private sendSingleUpdate(clientId: string, subscriptionId: string, row: Record<string, unknown> | null) {
1219
- const message: SingleUpdateMessage = {
1220
- type: "single_update",
1221
- subscriptionId,
1222
- row: row
1223
- };
1224
- this.sendMessage(clientId, message);
1225
- }
1226
-
1227
- /**
1228
- * Send a lightweight row-level patch to a collection subscriber.
1229
- * The client can merge this into its cached data for instant feedback.
1230
- *
1231
- * The key columns ride along: the patch names a row by address, and the
1232
- * client has to find that row among the ones it cached — which carry
1233
- * columns and no address. The SDK holds no collection config to derive one
1234
- * from, so this is the only place the mapping can come from.
1235
- */
1236
- /** The key columns of the collection at `path`, if they can be resolved. */
1237
- private primaryKeysForPath(path: string): PrimaryKeyInfo[] | undefined {
1238
- try {
1239
- const collection = this.registry.getCollectionByPath(path);
1240
- if (!collection) return undefined;
1241
- const keys = getPrimaryKeys(collection, this.registry);
1242
- return keys.length > 0 ? keys : undefined;
1243
- } catch {
1244
- // `getCollectionByPath` throws on a path it cannot walk — and this
1245
- // is called for parent paths too, which include entity paths like
1246
- // `posts/1` that name no collection. Telling the subscriber nothing
1247
- // is right here; letting it throw would drop the notification.
1248
- return undefined;
1249
- }
1250
- }
1251
-
1252
- private sendError(clientId: string, error: string, subscriptionId?: string, code?: string) {
1253
- const message = {
1254
- type: "error" as const,
1255
- subscriptionId,
1256
- payload: {
1257
- error: code ? { message: error, code } : error
1258
- },
1259
- error
1260
- };
1261
- this.sendMessage(clientId, message);
1262
- }
1263
-
1264
- private sendMessage(clientId: string, message: CollectionUpdateMessage | SingleUpdateMessage | CollectionPatchMessage | { type: string; subscriptionId?: string; error?: string; payload?: unknown }) {
1265
- const client = this.clients.get(clientId);
1266
- if (client && client.readyState === WebSocket.OPEN) {
1267
- client.send(JSON.stringify(message));
1268
- }
1269
- }
1270
-
1271
- /**
1272
- * Extract parent paths from a nested path like "posts/70/tags"
1273
- * Returns ["posts", "posts/70"] for the example above
1274
- */
1275
- private getParentPaths(path: string): string[] {
1276
- const segments = path.split("/").filter(s => s.length > 0);
1277
- const parentPaths: string[] = [];
1278
-
1279
- // Build parent paths progressively
1280
- for (let i = 1; i < segments.length; i += 2) {
1281
- const parentPath = segments.slice(0, i).join("/");
1282
- if (parentPath) {
1283
- parentPaths.push(parentPath);
1284
- }
1285
-
1286
- // If there's an row ID, add the path including the row
1287
- if (i + 1 < segments.length) {
1288
- const pathWithEntity = segments.slice(0, i + 1).join("/");
1289
- parentPaths.push(pathWithEntity);
1290
- }
1291
- }
1292
-
1293
- return parentPaths;
1294
- }
1295
-
1296
- // =============================================================================
1297
- // Broadcast Channels
1298
- // =============================================================================
1299
-
1300
- /**
1301
- * Install a channel authorizer — see {@link ChannelAuthorizer}.
1302
- *
1303
- * Nothing in the framework calls this yet: it is the seam a rules API will
1304
- * be built on, kept deliberately separate from the membership floor so the
1305
- * floor holds whether or not anyone uses it.
1306
- */
1307
- setChannelAuthorizer(authorizer: ChannelAuthorizer | undefined): void {
1308
- this.channelAuthorizer = authorizer;
1309
- }
1310
-
1311
- /** Which action each channel frame is asking to perform. */
1312
- private static readonly CHANNEL_ACTIONS: Record<string, ChannelAction> = {
1313
- join_channel: "join",
1314
- broadcast: "broadcast",
1315
- channel_history: "history",
1316
- presence_track: "join",
1317
- presence_state: "presence"
1318
- };
1319
-
1320
- /**
1321
- * The one door every channel frame comes through.
1322
- *
1323
- * Returns synchronously — and so dispatches synchronously — unless an
1324
- * authorizer is installed. That matters: a client sends `join_channel`,
1325
- * `presence_state` and `channel_history` back to back on connect, and the
1326
- * socket's message handler processes each frame up to its first `await`,
1327
- * so a gate that always yielded would let the reads overtake the join that
1328
- * is about to authorize them.
1329
- */
1330
- private handleChannelMessage(
1331
- clientId: string,
1332
- type: string,
1333
- payload: Record<string, unknown> | undefined,
1334
- authContext?: SubscriptionAuthContext
1335
- ): void | Promise<void> {
1336
- const channel = payload?.channel as string;
1337
-
1338
- // Leaving and untracking only ever remove the caller's own state, so
1339
- // they need no permission — refusing them could only strand a client.
1340
- if (type === "leave_channel") {
1341
- this.leaveChannel(clientId, channel);
1342
- return;
1343
- }
1344
- if (type === "presence_untrack") {
1345
- this.removePresence(clientId, channel);
1346
- return;
1347
- }
1348
-
1349
- const action = RealtimeService.CHANNEL_ACTIONS[type];
1350
- const allowed = this.authorizeChannelAction(clientId, channel, action, authContext);
1351
- if (allowed === false) return;
1352
- if (allowed === true) return this.dispatchChannelMessage(clientId, type, channel, payload);
1353
- return allowed.then((ok) => {
1354
- if (ok) return this.dispatchChannelMessage(clientId, type, channel, payload);
1355
- });
1356
- }
1357
-
1358
- /** Perform an already-authorized channel frame. */
1359
- private dispatchChannelMessage(
1360
- clientId: string,
1361
- type: string,
1362
- channel: string,
1363
- payload: Record<string, unknown> | undefined
1364
- ): void | Promise<void> {
1365
- switch (type) {
1366
- case "join_channel":
1367
- this.joinChannel(clientId, channel);
1368
- return;
1369
- case "broadcast":
1370
- this.broadcastToChannel(clientId, channel, payload?.event as string, payload?.payload);
1371
- return;
1372
- case "channel_history":
1373
- return this.handleChannelHistoryRequest(
1374
- clientId,
1375
- channel,
1376
- payload?.sinceSeq as number | undefined,
1377
- payload?.limit as number | undefined
1378
- );
1379
- case "presence_track":
1380
- // Auto-join the channel so presence works without a separate join
1381
- this.joinChannel(clientId, channel);
1382
- this.trackPresence(clientId, channel, payload?.state as Record<string, unknown> ?? {});
1383
- return;
1384
- case "presence_state":
1385
- this.sendPresenceState(clientId, channel);
1386
- return;
1387
- }
1388
- }
1389
-
1390
- /**
1391
- * Decide whether a client may perform an action on a channel.
1392
- *
1393
- * **Membership is the floor.** Reading a channel's presence roster, replaying
1394
- * its retained history and broadcasting into it all require that this client
1395
- * has joined it. That is a low bar — joining is open to anyone who can name
1396
- * the channel — but it is not the bar that was there before, which was none
1397
- * at all: `channel_history` and `presence_state` answered any socket about
1398
- * any channel, and a broadcast fanned out to members the sender had never
1399
- * joined. Two internal tables (`rebase.channel_presence`,
1400
- * `rebase.channel_messages`) are held outside RLS on the strength of this
1401
- * check, so it fails closed: an authorizer that throws refuses the frame.
1402
- *
1403
- * Anything richer than membership belongs in a {@link ChannelAuthorizer};
1404
- * this method is where it is consulted, and the only place.
1405
- */
1406
- private authorizeChannelAction(
1407
- clientId: string,
1408
- channel: string,
1409
- action: ChannelAction,
1410
- authContext?: SubscriptionAuthContext
1411
- ): boolean | Promise<boolean> {
1412
- // Joining is what establishes membership, so it cannot require it.
1413
- if (action !== "join" && !this.channels.get(channel)?.has(clientId)) {
1414
- this.denyChannelAction(clientId, channel, action, "not a member of the channel");
1415
- return false;
1416
- }
1417
-
1418
- const authorizer = this.channelAuthorizer;
1419
- if (!authorizer) return true;
1420
-
1421
- let verdict: boolean | Promise<boolean>;
1422
- try {
1423
- verdict = authorizer({ channel, action, clientId, user: authContext });
1424
- } catch (error) {
1425
- logger.error(`❌ [Channels] Authorizer threw for ${action} on "${channel}" — refusing`, { error });
1426
- this.denyChannelAction(clientId, channel, action, "channel authorization failed");
1427
- return false;
1428
- }
1429
-
1430
- if (typeof verdict === "boolean") {
1431
- if (!verdict) this.denyChannelAction(clientId, channel, action, "refused by the channel authorizer");
1432
- return verdict;
1433
- }
1434
-
1435
- return verdict.then(
1436
- (ok) => {
1437
- if (!ok) this.denyChannelAction(clientId, channel, action, "refused by the channel authorizer");
1438
- return ok;
1439
- },
1440
- (error) => {
1441
- logger.error(`❌ [Channels] Authorizer rejected for ${action} on "${channel}" — refusing`, { error });
1442
- this.denyChannelAction(clientId, channel, action, "channel authorization failed");
1443
- return false;
1444
- }
1445
- );
1446
- }
1447
-
1448
- /** Tell the client why its channel frame went nowhere, and say so in the log. */
1449
- private denyChannelAction(clientId: string, channel: string, action: ChannelAction, reason: string): void {
1450
- this.debugLog(`🚫 [Channels] Refused ${action} on "${channel}" for ${clientId}: ${reason}`);
1451
- this.sendError(
1452
- clientId,
1453
- `Refused ${action} on channel "${channel}": ${reason}`,
1454
- undefined,
1455
- "CHANNEL_FORBIDDEN"
1456
- );
1457
- }
1458
-
1459
- /** Join a broadcast channel */
1460
- joinChannel(clientId: string, channel: string): void {
1461
- if (!this.channels.has(channel)) {
1462
- this.channels.set(channel, new Set());
1463
- }
1464
- this.channels.get(channel)!.add(clientId);
1465
- this.warnIfMemoryBusOnMultiplePods();
1466
- this.debugLog(`📡 [Broadcast] Client ${clientId} joined channel: ${channel}`);
1467
- }
1468
-
1469
- /**
1470
- * Say something the first time channels are used on a deployment that is
1471
- * demonstrably multi-pod while the bus is still the in-memory default.
1472
- *
1473
- * Every other warning in this subsystem covers a *configured* bus failing —
1474
- * the case where the operator already knew a bus mattered. The common
1475
- * misconfiguration is the opposite one: scaled to two replicas, never
1476
- * touched `realtime.bus`, and broadcast and presence quietly serve a
1477
- * fraction of the room. The evidence is already in the process, so use it.
1478
- */
1479
- private warnIfMemoryBusOnMultiplePods(): void {
1480
- if (this.memoryBusWarned) return;
1481
- if (this.bus.kind !== "memory" || !this.foreignInstanceSeen) return;
1482
- this.memoryBusWarned = true;
1483
- logger.warn(
1484
- "⚠️ [ChannelBus] Channels are in use with the in-memory bus, but notifications from another " +
1485
- "instance have been seen — this deployment runs more than one process. Broadcast and presence " +
1486
- "reach only the clients connected to this one. Set `realtime.bus` (or REALTIME_CHANNEL_BUS=postgres) " +
1487
- "to make channels cross-instance."
1488
- );
1489
- }
1490
-
1491
- /** Leave a broadcast channel */
1492
- leaveChannel(clientId: string, channel: string): void {
1493
- const members = this.channels.get(channel);
1494
- if (members) {
1495
- members.delete(clientId);
1496
- if (members.size === 0) this.channels.delete(channel);
1497
- }
1498
- // Also remove presence
1499
- this.removePresence(clientId, channel);
1500
- }
1501
-
1502
- /**
1503
- * Broadcast a message to all clients in a channel except the sender.
1504
- *
1505
- * On a channel with no retention rule this is what it always was: a
1506
- * synchronous fan-out to whoever is connected, with no sequence number, no
1507
- * SQL and no await — the body below runs to completion before returning.
1508
- *
1509
- * On a retained channel the message is durably numbered first and only then
1510
- * delivered, through a per-channel queue so that delivery order matches
1511
- * sequence order. That ordering is the whole point: a client that catches up
1512
- * with `sinceSeq` has to arrive at the same state as one that never
1513
- * disconnected.
1514
- */
1515
- broadcastToChannel(clientId: string, channel: string, event: string, payload: unknown): void {
1516
- const retention = this.channelHistory?.retentionFor(channel);
1517
- if (!retention) {
1518
- this.fanOutBroadcast(clientId, channel, event, payload);
1519
- // Other instances get the same frame, but never before the clients
1520
- // on this one: the local fan-out above is synchronous and the
1521
- // publish is not, which is also what keeps the ephemeral path free
1522
- // of any await for a single-instance deployment.
1523
- this.publishBroadcast(clientId, channel, event, payload);
1524
- return;
1525
- }
1526
-
1527
- const previous = this.channelSendQueues.get(channel) ?? Promise.resolve();
1528
- const next = previous
1529
- // A failed predecessor must not poison the chain — the next message
1530
- // on this channel is independent and still deserves to be sent.
1531
- .catch(() => { /* already reported below */ })
1532
- .then(() => this.persistAndFanOut(clientId, channel, event, payload, retention));
1533
-
1534
- this.channelSendQueues.set(channel, next);
1535
- void next.finally(() => {
1536
- // Only clear if nothing has queued behind us in the meantime.
1537
- if (this.channelSendQueues.get(channel) === next) this.channelSendQueues.delete(channel);
1538
- });
1539
- }
1540
-
1541
- /**
1542
- * Number a broadcast, store it, then deliver it.
1543
- *
1544
- * A message that cannot be stored is **not** delivered. Delivering it would
1545
- * put it in front of live subscribers while leaving it absent from every
1546
- * future replay — the two views of the channel would disagree permanently,
1547
- * and no later message could repair the gap. Failing loudly to the sender
1548
- * instead lets it retry, which for an operation stream is the only outcome
1549
- * that keeps clients convergent.
1550
- */
1551
- private async persistAndFanOut(
1552
- clientId: string,
1553
- channel: string,
1554
- event: string,
1555
- payload: unknown,
1556
- retention: ResolvedRetention
1557
- ): Promise<void> {
1558
- let seq: number;
1559
- try {
1560
- ({ seq } = await this.channelHistory!.append(channel, event, payload, clientId));
1561
- } catch (error) {
1562
- logger.error(`❌ [ChannelHistory] Could not persist broadcast on "${channel}" — message dropped`, { error });
1563
- this.sendError(
1564
- clientId,
1565
- `Could not persist broadcast on retained channel "${channel}"`,
1566
- undefined,
1567
- "CHANNEL_HISTORY_WRITE_FAILED"
1568
- );
1569
- return;
1570
- }
1571
-
1572
- this.fanOutBroadcast(clientId, channel, event, payload, seq);
1573
- this.publishBroadcast(clientId, channel, event, payload, seq);
1574
-
1575
- try {
1576
- await this.channelHistory!.prune(channel, retention);
1577
- } catch (error) {
1578
- // Retention is a housekeeping concern; the message is already
1579
- // delivered and durable, so a failed prune must not surface as a
1580
- // broadcast failure. It will be retried on the next message.
1581
- logger.warn(`⚠️ [ChannelHistory] Prune failed for "${channel}"`, { error });
1582
- }
1583
- }
1584
-
1585
- /** Deliver a broadcast frame to every member of a channel but the sender. */
1586
- private fanOutBroadcast(clientId: string, channel: string, event: string, payload: unknown, seq?: number): void {
1587
- const members = this.channels.get(channel);
1588
- if (!members) return;
1589
-
1590
- const message = JSON.stringify({
1591
- type: "broadcast",
1592
- channel,
1593
- event,
1594
- payload,
1595
- ...(seq !== undefined ? { seq } : {})
1596
- });
1597
-
1598
- for (const memberId of members) {
1599
- if (memberId === clientId) continue; // Don't echo back to sender
1600
- const ws = this.clients.get(memberId);
1601
- if (ws && ws.readyState === WebSocket.OPEN) {
1602
- ws.send(message);
1603
- }
1604
- }
1605
- }
1606
-
1607
- // =============================================================================
1608
- // Cross-Instance Channel Bus
1609
- // =============================================================================
1610
-
1611
- /**
1612
- * Install the transport that carries channel frames between instances.
1613
- *
1614
- * Called once at boot. A bus that cannot start is reported and replaced with
1615
- * the memory bus: losing cross-instance fan-out degrades collaboration to
1616
- * what it was before this existed, whereas refusing to boot takes the whole
1617
- * backend down for it.
1618
- */
1619
- async configureChannelBus(bus: ChannelBus): Promise<void> {
1620
- if (bus.kind === "memory") {
1621
- this.bus = bus;
1622
- return;
1623
- }
1624
-
1625
- try {
1626
- await bus.start((frame) => this.handleBusFrame(frame));
1627
- } catch (error) {
1628
- logger.warn(
1629
- `⚠️ [ChannelBus] Could not start the "${bus.kind}" channel bus — channel broadcast and presence ` +
1630
- "stay per-instance. Clients served by different replicas will not see each other.",
1631
- { error }
1632
- );
1633
- await bus.stop().catch(() => { /* best effort */ });
1634
- this.bus = new MemoryChannelBus();
1635
- return;
1636
- }
1637
-
1638
- this.bus = bus;
1639
-
1640
- // Presence needs shared *state*, not just shared fan-out — see
1641
- // `channel-presence.ts`. It comes up with the bus and only with it.
1642
- try {
1643
- const store = new ChannelPresenceStore(this.db, this.instanceId);
1644
- await store.ensureTables();
1645
- this.presenceStore = store;
1646
- this.ensurePresenceSweep();
1647
- } catch (error) {
1648
- logger.warn(
1649
- "⚠️ [ChannelBus] Could not create the shared presence table — presence rosters will only list " +
1650
- "clients connected to this instance (broadcast is unaffected).",
1651
- { error }
1652
- );
1653
- this.presenceStore = undefined;
1654
- }
1655
-
1656
- logger.info(
1657
- `📡 [ChannelBus] Cross-instance channels active via ${bus.kind} (instanceId: ${this.instanceId}).`
1658
- );
1659
- }
1660
-
1661
- /** Which transport is in use — `"memory"` means per-instance only. */
1662
- public getChannelBusKind(): ChannelBus["kind"] {
1663
- return this.bus.kind;
1664
- }
1665
-
1666
- /**
1667
- * Send a broadcast to the other instances.
1668
- *
1669
- * Fire-and-forget by design: the clients on this instance have already been
1670
- * served, and a bus that is briefly unreachable must not turn a broadcast
1671
- * into an error for the sender.
1672
- */
1673
- private publishBroadcast(clientId: string, channel: string, event: string, payload: unknown, seq?: number): void {
1674
- if (this.bus.kind === "memory") return;
1675
-
1676
- const frame: ChannelBusFrame = {
1677
- kind: "broadcast",
1678
- sid: this.instanceId,
1679
- channel,
1680
- event,
1681
- from: clientId,
1682
- ...(seq !== undefined ? { seq } : {}),
1683
- payload
1684
- };
1685
-
1686
- // Postgres caps a NOTIFY payload at 8 KB. A retained message is already
1687
- // durable and addressable, so it travels as a pointer and each receiver
1688
- // reads the body back — the same shape as the entity path, which
1689
- // notifies an address and refetches the row.
1690
- if (frameByteLength(frame) > this.bus.maxFrameBytes) {
1691
- if (seq === undefined) {
1692
- this.reportOversizedBroadcast(clientId, channel);
1693
- return;
1694
- }
1695
- void this.publishFrame({
1696
- kind: "broadcast_ref",
1697
- sid: this.instanceId,
1698
- channel,
1699
- from: clientId,
1700
- seq
1701
- });
1702
- return;
1703
- }
1704
-
1705
- void this.publishFrame(frame);
1706
- }
1707
-
1708
- private async publishFrame(frame: ChannelBusFrame): Promise<void> {
1709
- try {
1710
- await this.bus.publish(frame);
1711
- } catch (error) {
1712
- logger.error("❌ [ChannelBus] Failed to publish frame — other instances did not receive it", {
1713
- detail: `${frame.kind} on "${frame.channel}"`,
1714
- error
1715
- });
1716
- }
1717
- }
1718
-
1719
- /**
1720
- * Tell the sender that a message was delivered locally but nowhere else.
1721
- *
1722
- * Staying quiet here would be the worst option available: on one instance
1723
- * the app works, on two it works for half the users, and nothing in the
1724
- * logs connects the two. The fix is a one-liner in config — give the
1725
- * channel a retention rule and the message travels as a pointer instead —
1726
- * so the message says exactly that.
1727
- */
1728
- private reportOversizedBroadcast(clientId: string, channel: string): void {
1729
- const remedy =
1730
- `Add a retention rule for "${channel}" (realtime.channels) — retained messages travel by reference ` +
1731
- "and have no size limit.";
1732
-
1733
- if (!this.oversizedBroadcastWarned.has(channel)) {
1734
- this.oversizedBroadcastWarned.add(channel);
1735
- logger.warn(
1736
- `⚠️ [ChannelBus] A broadcast on ephemeral channel "${channel}" exceeds the ` +
1737
- `${this.bus.maxFrameBytes}-byte limit of the ${this.bus.kind} bus and reached only this instance. ` +
1738
- remedy
1739
- );
1740
- }
1741
- this.sendError(
1742
- clientId,
1743
- `Broadcast on "${channel}" was too large to reach other instances. ${remedy}`,
1744
- undefined,
1745
- "CHANNEL_BUS_PAYLOAD_TOO_LARGE"
1746
- );
1747
- }
1748
-
1749
- /**
1750
- * Deliver a frame published by another instance to this one's clients.
1751
- *
1752
- * Frames we published ourselves are dropped on arrival — the local fan-out
1753
- * happened before the publish — exactly as the entity-change handler skips
1754
- * its own `sid`.
1755
- */
1756
- private async handleBusFrame(frame: ChannelBusFrame): Promise<void> {
1757
- if (frame.sid === this.instanceId) return;
1758
-
1759
- switch (frame.kind) {
1760
- case "broadcast":
1761
- this.fanOutBroadcast(frame.from ?? "", frame.channel, frame.event, frame.payload, frame.seq);
1762
- return;
1763
-
1764
- case "broadcast_ref": {
1765
- // Nothing to read back for: skip the query rather than pay for
1766
- // a message no client here is waiting for.
1767
- if (!this.channels.get(frame.channel)?.size) return;
1768
-
1769
- const entry = await this.channelHistory?.getBySeq(frame.channel, frame.seq);
1770
- if (!entry) {
1771
- logger.warn(
1772
- `⚠️ [ChannelBus] Message ${frame.seq} on "${frame.channel}" is no longer retained — ` +
1773
- "clients on this instance will need to replay (channel_history) to catch up."
1774
- );
1775
- return;
1776
- }
1777
- this.fanOutBroadcast(frame.from ?? "", frame.channel, entry.event, entry.payload, entry.seq);
1778
- return;
1779
- }
1780
-
1781
- case "presence_diff":
1782
- this.deliverPresenceDiff(frame.channel, frame.joins, frame.leaves);
1783
- return;
1784
- }
1785
- }
1786
-
1787
- // =============================================================================
1788
- // Channel History
1789
- // =============================================================================
1790
-
1791
- /**
1792
- * Install retention rules and create the tables they need.
1793
- *
1794
- * Safe to call with no rules (and safe not to call at all): the store stays
1795
- * inert, no schema is created, and broadcast keeps its original
1796
- * fire-and-forget path.
1797
- */
1798
- async configureChannelHistory(
1799
- rules: ChannelRetentionRule[] | undefined,
1800
- options?: { provision?: boolean }
1801
- ): Promise<void> {
1802
- // The store is built in every process, whether or not this one creates
1803
- // the tables: retaining a message is what a process does when it
1804
- // *publishes* to a retained channel, and a function handler publishes as
1805
- // readily as a websocket client does. Only the DDL is owned.
1806
- this.channelHistory = new ChannelHistoryStore(this.db, rules ?? []);
1807
- if (!this.channelHistory.enabled) return;
1808
- if (options?.provision === false) return;
1809
- await this.channelHistory.ensureTables();
1810
- }
1811
-
1812
- /** Whether any channel is configured to retain messages. */
1813
- public isChannelHistoryEnabled(): boolean {
1814
- return this.channelHistory?.enabled ?? false;
1815
- }
1816
-
1817
- /**
1818
- * Answer a client's catch-up request.
1819
- *
1820
- * A channel with no retention rule is answered with `retained: false`
1821
- * rather than an empty list, so the client can tell "you missed nothing"
1822
- * apart from "this channel never keeps anything" — the second means its
1823
- * reconnect strategy has to be a full resync, and silence would leave it
1824
- * guessing.
1825
- */
1826
- private async handleChannelHistoryRequest(
1827
- clientId: string,
1828
- channel: string,
1829
- sinceSeq?: number,
1830
- limit?: number
1831
- ): Promise<void> {
1832
- if (!channel) return;
1833
-
1834
- const retention = this.channelHistory?.retentionFor(channel);
1835
- if (!retention) {
1836
- this.sendChannelHistory(clientId, channel, [], false);
1837
- return;
1838
- }
1839
-
1840
- try {
1841
- const { messages, latestSeq } = await this.channelHistory!.replay(channel, sinceSeq, limit);
1842
- this.sendChannelHistory(clientId, channel, messages, true, latestSeq);
1843
- } catch (error) {
1844
- logger.error(`❌ [ChannelHistory] Replay failed for "${channel}"`, { error });
1845
- this.sendError(clientId, `Could not replay history for channel "${channel}"`, undefined, "CHANNEL_HISTORY_READ_FAILED");
1846
- }
1847
- }
1848
-
1849
- private sendChannelHistory(
1850
- clientId: string,
1851
- channel: string,
1852
- messages: ChannelHistoryEntry[],
1853
- retained: boolean,
1854
- latestSeq?: number
1855
- ): void {
1856
- const ws = this.clients.get(clientId);
1857
- if (ws && ws.readyState === WebSocket.OPEN) {
1858
- ws.send(JSON.stringify({
1859
- type: "channel_history",
1860
- channel,
1861
- messages,
1862
- retained,
1863
- ...(latestSeq !== undefined ? { latestSeq } : {})
1864
- }));
1865
- }
1866
- }
1867
-
1868
- // =============================================================================
1869
- // Presence
1870
- // =============================================================================
1871
-
1872
- /**
1873
- * Track presence in a channel.
1874
- *
1875
- * The client re-sends this every ~20s as a heartbeat against the 30s
1876
- * timeout, so most calls carry the state that is already recorded. Those
1877
- * refresh `last_seen` and stop there: re-announcing an unchanged state to
1878
- * every instance would put a bus message per client per heartbeat on the
1879
- * wire to tell everyone nothing happened.
1880
- */
1881
- trackPresence(clientId: string, channel: string, state: Record<string, unknown>): void {
1882
- if (!this.presence.has(channel)) {
1883
- this.presence.set(channel, new Map());
1884
- }
1885
-
1886
- const channelPresence = this.presence.get(channel)!;
1887
- const previous = channelPresence.get(clientId);
1888
- const changed = !previous || JSON.stringify(previous.state) !== JSON.stringify(state);
1889
- channelPresence.set(clientId, { state,
1890
- lastSeen: Date.now() });
1891
-
1892
- // Refresh the shared roster on every heartbeat — that timestamp is what
1893
- // tells other instances this client is still here.
1894
- void this.presenceStoreOp(() => this.presenceStore!.track(channel, clientId, state), "track");
1895
-
1896
- // Broadcast join / state update to channel
1897
- this.deliverPresenceDiff(channel, { [clientId]: state }, {});
1898
- if (changed) {
1899
- this.publishPresenceDiff(channel, { [clientId]: state }, {});
1900
- }
1901
-
1902
- // Start cleanup interval if not running
1903
- this.ensurePresenceCleanup();
1904
- }
1905
-
1906
- /**
1907
- * Remove presence from a channel.
1908
- *
1909
- * `skipStore` is for the socket-close path, which clears every channel at
1910
- * once and then deletes the client's rows in a single statement instead of
1911
- * one per channel.
1912
- */
1913
- removePresence(clientId: string, channel: string, options?: { skipStore?: boolean }): void {
1914
- const channelPresence = this.presence.get(channel);
1915
- if (!channelPresence) return;
1916
-
1917
- const entry = channelPresence.get(clientId);
1918
- if (entry) {
1919
- channelPresence.delete(clientId);
1920
- this.deliverPresenceDiff(channel, {}, { [clientId]: entry.state });
1921
- this.publishPresenceDiff(channel, {}, { [clientId]: entry.state });
1922
- if (!options?.skipStore) {
1923
- void this.presenceStoreOp(() => this.presenceStore!.remove(channel, clientId), "remove");
1924
- }
1925
- }
1926
-
1927
- if (channelPresence.size === 0) {
1928
- this.presence.delete(channel);
1929
- }
1930
- }
1931
-
1932
- /**
1933
- * Send the full roster for a channel to one client.
1934
- *
1935
- * Answered from the shared table when there is one, because "who is in this
1936
- * document?" has a single answer that must not depend on which replica the
1937
- * asker happens to be connected to. Without a bus there is nothing to share
1938
- * and the local map *is* the roster — that path stays synchronous, which is
1939
- * what it always was.
1940
- */
1941
- sendPresenceState(clientId: string, channel: string): void {
1942
- if (!this.presenceStore) {
1943
- this.sendPresenceStateMessage(clientId, channel, this.localPresences(channel));
1944
- return;
1945
- }
1946
-
1947
- void this.presenceStore.roster(channel)
1948
- .then((presences) => {
1949
- this.sendPresenceStateMessage(clientId, channel, presences);
1950
- })
1951
- .catch((error) => {
1952
- // A roster the asker can act on beats none: fall back to the
1953
- // clients we can see rather than leaving the request unanswered.
1954
- logger.warn(`⚠️ [Presence] Could not read the shared roster for "${channel}" — answering with this instance's clients only.`, { error });
1955
- this.sendPresenceStateMessage(clientId, channel, this.localPresences(channel));
1956
- });
1957
- }
1958
-
1959
- /** Presence of the clients connected to this instance. */
1960
- private localPresences(channel: string): Record<string, Record<string, unknown>> {
1961
- const channelPresence = this.presence.get(channel);
1962
- const presences: Record<string, Record<string, unknown>> = {};
1963
- if (channelPresence) {
1964
- for (const [id, { state }] of channelPresence) {
1965
- presences[id] = state;
1966
- }
1967
- }
1968
- return presences;
1969
- }
1970
-
1971
- private sendPresenceStateMessage(
1972
- clientId: string,
1973
- channel: string,
1974
- presences: Record<string, Record<string, unknown>>
1975
- ): void {
1976
- const ws = this.clients.get(clientId);
1977
- if (ws && ws.readyState === WebSocket.OPEN) {
1978
- ws.send(JSON.stringify({
1979
- type: "presence_state",
1980
- channel,
1981
- presences
1982
- }));
1983
- }
1984
- }
1985
-
1986
- /** Deliver a presence diff to this instance's members of the channel. */
1987
- private deliverPresenceDiff(
1988
- channel: string,
1989
- joins: Record<string, Record<string, unknown>>,
1990
- leaves: Record<string, Record<string, unknown>>
1991
- ): void {
1992
- const members = this.channels.get(channel);
1993
- if (!members) return;
1994
-
1995
- const message = JSON.stringify({
1996
- type: "presence_diff",
1997
- channel,
1998
- joins,
1999
- leaves
2000
- });
2001
-
2002
- for (const memberId of members) {
2003
- const ws = this.clients.get(memberId);
2004
- if (ws && ws.readyState === WebSocket.OPEN) {
2005
- ws.send(message);
2006
- }
2007
- }
2008
- }
2009
-
2010
- /** Tell the other instances about a presence change. */
2011
- private publishPresenceDiff(
2012
- channel: string,
2013
- joins: Record<string, Record<string, unknown>>,
2014
- leaves: Record<string, Record<string, unknown>>
2015
- ): void {
2016
- if (this.bus.kind === "memory") return;
2017
- void this.publishFrame({ kind: "presence_diff", sid: this.instanceId, channel, joins, leaves });
2018
- }
2019
-
2020
- /** Run a roster write when there is a roster, and never let it throw. */
2021
- private async presenceStoreOp(op: () => Promise<void>, label: string): Promise<void> {
2022
- if (!this.presenceStore) return;
2023
- try {
2024
- await op();
2025
- } catch (error) {
2026
- logger.warn(`⚠️ [Presence] Shared roster ${label} failed`, { error });
2027
- }
2028
- }
2029
-
2030
- /** Periodic cleanup for stale presences */
2031
- private ensurePresenceCleanup(): void {
2032
- if (this.presenceInterval) return;
2033
- this.presenceInterval = setInterval(() => {
2034
- const now = Date.now();
2035
- for (const [channel, channelPresence] of this.presence) {
2036
- for (const [clientId, entry] of channelPresence) {
2037
- if (now - entry.lastSeen > RealtimeService.PRESENCE_TIMEOUT_MS) {
2038
- this.removePresence(clientId, channel);
2039
- }
2040
- }
2041
- }
2042
- // Stop interval if no presences tracked
2043
- if (this.presence.size === 0 && this.presenceInterval) {
2044
- clearInterval(this.presenceInterval);
2045
- this.presenceInterval = undefined;
2046
- }
2047
- }, 10000); // Check every 10s
2048
- }
2049
-
2050
- /**
2051
- * Reap roster rows whose owning instance stopped heartbeating.
2052
- *
2053
- * This is the cross-instance half of the sweep above, and it doubles as
2054
- * crash recovery: a pod that dies takes its clients with it but leaves
2055
- * their rows behind, and after one TTL window they look exactly like any
2056
- * other client that went quiet. The delete returns what it removed, so
2057
- * whichever instance wins the race is the one that announces the
2058
- * departures — once for the cluster, not once per replica.
2059
- */
2060
- private ensurePresenceSweep(): void {
2061
- if (this.presenceSweepInterval || !this.presenceStore) return;
2062
-
2063
- this.presenceSweepInterval = setInterval(
2064
- () => void this.sweepStalePresence(),
2065
- RealtimeService.PRESENCE_SWEEP_INTERVAL_MS
2066
- );
2067
-
2068
- // Never hold the process open for housekeeping.
2069
- (this.presenceSweepInterval as unknown as { unref?: () => void }).unref?.();
2070
- }
2071
-
2072
- /** One pass of the stale-roster sweep. See {@link ensurePresenceSweep}. */
2073
- private async sweepStalePresence(): Promise<void> {
2074
- if (!this.presenceStore) return;
2075
- try {
2076
- const removed = await this.presenceStore.sweepStale(RealtimeService.PRESENCE_TIMEOUT_MS);
2077
- for (const row of removed) {
2078
- this.debugLog(`👻 [Presence] Reaped stale presence ${row.clientId} on "${row.channel}"`);
2079
- this.deliverPresenceDiff(row.channel, {}, { [row.clientId]: row.state });
2080
- this.publishPresenceDiff(row.channel, {}, { [row.clientId]: row.state });
2081
- }
2082
- } catch (error) {
2083
- logger.warn("⚠️ [Presence] Stale-roster sweep failed", { error });
2084
- }
2085
- }
2086
-
2087
- // =============================================================================
2088
- // Lifecycle / Cleanup
2089
- // =============================================================================
2090
-
2091
- /**
2092
- * Gracefully tear down all realtime resources.
2093
- *
2094
- * This MUST be called during process shutdown, **before** `pool.end()`.
2095
- * It ensures:
2096
- * 1. All debounced refetch timers are cancelled (prevents queries after pool closes).
2097
- * 2. All subscription state and callbacks are cleared.
2098
- * 3. The dedicated LISTEN client (outside the pool) is disconnected.
2099
- * 4. All WebSocket clients are removed (but not forcefully closed — the
2100
- * HTTP server close will handle that).
2101
- */
2102
- async destroy(): Promise<void> {
2103
- // 1. Cancel every pending debounced refetch timer
2104
- for (const [key, timer] of this.refetchTimers) {
2105
- clearTimeout(timer);
2106
- this.refetchTimers.delete(key);
2107
- }
2108
-
2109
- // 2. Clear subscriptions and callbacks
2110
- this._subscriptions.clear();
2111
- this.subscriptionCallbacks.clear();
2112
-
2113
- // 3. Clear broadcast channels and presence
2114
- this.channels.clear();
2115
- this.presence.clear();
2116
- // Pending history writes hold the pool open; let them settle before the
2117
- // caller closes it, but never let a rejected one break shutdown.
2118
- await Promise.allSettled([...this.channelSendQueues.values()]);
2119
- this.channelSendQueues.clear();
2120
- this.channelHistory?.clear();
2121
- if (this.presenceInterval) {
2122
- clearInterval(this.presenceInterval);
2123
- this.presenceInterval = undefined;
2124
- }
2125
- if (this.presenceSweepInterval) {
2126
- clearInterval(this.presenceSweepInterval);
2127
- this.presenceSweepInterval = undefined;
2128
- }
2129
- this.oversizedBroadcastWarned.clear();
2130
-
2131
- // Drop this instance's roster rows now rather than leaving every other
2132
- // replica to wait out a TTL window on ghosts — a rolling deploy would
2133
- // otherwise show 30s of departed users on every restart.
2134
- if (this.presenceStore) {
2135
- try {
2136
- await this.presenceStore.removeInstance();
2137
- } catch (error) {
2138
- logger.warn("⚠️ [Presence] Could not clear this instance's roster rows on shutdown", { error });
2139
- }
2140
- this.presenceStore = undefined;
2141
- }
2142
-
2143
- // 4. Disconnect the dedicated LISTEN client(s)
2144
- await this.stopListening();
2145
- await this.stopCdc();
2146
- await this.bus.stop().catch((error) =>
2147
- logger.warn("⚠️ [ChannelBus] Error while stopping the channel bus", { error }));
2148
- this.bus = new MemoryChannelBus();
2149
-
2150
- // 5. Drop client references (don't close — server.close drains them)
2151
- this.clients.clear();
2152
-
2153
- this.debugLog("🧹 [RealtimeService] destroy() complete — all resources released.");
2154
- }
2155
-
2156
- // =============================================================================
2157
- // Database-level Change Data Capture (CDC)
2158
- // =============================================================================
2159
-
2160
- /** Whether database-level change capture is currently the active source. */
2161
- public isCdcActive(): boolean {
2162
- return this.cdcActive;
2163
- }
2164
-
2165
- /**
2166
- * Enable database-level change capture as the realtime source.
2167
- *
2168
- * A dedicated LISTEN client consumes committed changes from the `rebase_cdc`
2169
- * channel (fed by CDC triggers — see {@link provisionTriggerCdc}) and routes
2170
- * them into the same {@link notifyUpdate} pipeline used by API mutations. The
2171
- * effect: subscribers see a change no matter how it was written — psql, a
2172
- * cron in another service, raw SQL, or the Studio SQL editor — exactly like
2173
- * Supabase Realtime tailing the WAL.
2174
- *
2175
- * Because CDC observes every commit on every instance, it also *replaces* the
2176
- * legacy per-mutation cross-instance broadcast (see the guard in
2177
- * {@link notifyUpdate}); callers should not also call {@link startListening}.
2178
- *
2179
- * @param connectionString Direct Postgres connection for the LISTEN client
2180
- * (bypass PgBouncer — LISTEN needs a session connection).
2181
- */
2182
- async enableCdc(connectionString: string): Promise<void> {
2183
- if (this.cdcActive) {
2184
- logger.warn("⚠️ [CDC] enableCdc called but CDC is already active. Ignoring.");
2185
- return;
2186
- }
2187
- this.cdcTableMap = this.buildCdcTableMap();
2188
- this.junctionLinkMap = buildJunctionLinkMap(this.registry);
2189
- this.cdcListener = new CdcListener(connectionString, (event) => this.handleCdcEvent(event));
2190
- try {
2191
- // start() validates the initial connection; if it can't be established
2192
- // it rejects here, and we leave CDC inactive so the caller can fall
2193
- // back to app-level realtime rather than silently dropping events.
2194
- await this.cdcListener.start();
2195
- } catch (err) {
2196
- await this.cdcListener.stop().catch(() => { /* best effort */ });
2197
- this.cdcListener = undefined;
2198
- this.cdcTableMap = undefined;
2199
- this.junctionLinkMap = undefined;
2200
- throw err;
2201
- }
2202
- this.cdcActive = true;
2203
- // The bootstrapper says the same thing one line later, in the
2204
- // vocabulary of the setting that produced it (REALTIME_CDC).
2205
- logger.debug(
2206
- `📡 [RealtimeService] Database-level change capture ACTIVE — writes from ANY source now emit realtime events ` +
2207
- `(${this.cdcTableMap.size} mapped table key(s)).`
2208
- );
2209
- }
2210
-
2211
- /** Stop the CDC listener and clear its state. */
2212
- async stopCdc(): Promise<void> {
2213
- this.cdcActive = false;
2214
- if (this.cdcListener) {
2215
- await this.cdcListener.stop();
2216
- this.cdcListener = undefined;
2217
- }
2218
- this.cdcTableMap = undefined;
2219
- this.junctionLinkMap = undefined;
2220
- this.recentAppEmits.clear();
2221
- }
2222
-
2223
- /**
2224
- * Build the reverse map from database table → collection. A change event
2225
- * carries `schema` + `table`; realtime subscriptions are keyed by collection
2226
- * path (slug). We index by both `schema.table` and bare `table` so the lookup
2227
- * works whether or not the collection declares an explicit schema.
2228
- */
2229
- private buildCdcTableMap(): Map<string, CollectionConfig> {
2230
- const map = new Map<string, CollectionConfig>();
2231
- for (const collection of this.registry.getCollections()) {
2232
- const table = getTableName(collection);
2233
- if (!table) continue;
2234
- const schema = (collection as { schema?: string }).schema ?? "public";
2235
- map.set(`${schema}.${table}`, collection);
2236
- // Bare-table fallback; first registration wins to keep it deterministic.
2237
- if (!map.has(table)) map.set(table, collection);
2238
- }
2239
- return map;
2240
- }
2241
-
2242
- private resolveCollectionForTable(schema: string, table: string): CollectionConfig | undefined {
2243
- if (!this.cdcTableMap) return undefined;
2244
- return this.cdcTableMap.get(`${schema}.${table}`) ?? this.cdcTableMap.get(table);
2245
- }
2246
-
2247
- /**
2248
- * Route a captured database change into the realtime pipeline.
2249
- *
2250
- * Delivery is RLS-safe by construction: the raw tuple from the WAL/trigger is
2251
- * NOT forwarded to subscribers. Instead the change is marked invalidated, so
2252
- * every matching subscription re-reads the row under its own auth context via
2253
- * {@link fetchCollectionWithAuth} / {@link fetchEntityWithAuth}. A subscriber
2254
- * therefore only ever receives rows its RLS policies permit — filtering is per
2255
- * subscriber, never per publisher.
2256
- */
2257
- private async handleCdcEvent(event: CdcChangeEvent): Promise<void> {
2258
- const collection = this.resolveCollectionForTable(event.schema, event.table);
2259
- if (!collection) {
2260
- // A junction table backs no collection, but its rows *are* a child
2261
- // list. Route the change to the lists it changes before giving up.
2262
- if (await this.handleJunctionCdcEvent(event)) return;
2263
-
2264
- // Unmapped table (not backed by a collection) — nothing to deliver.
2265
- this.debugLog(`📡 [CDC] Ignoring change on unmapped table ${event.schema}.${event.table}`);
2266
- return;
2267
- }
2268
-
2269
- const path = collection.slug;
2270
- const databaseId = (collection as { databaseId?: string }).databaseId;
2271
- const id = this.extractIdFromCdcRow(collection, event.row);
2272
-
2273
- // Deletes carry a null row (subscribers drop the id); inserts/updates carry
2274
- // an invalidation marker that forces a per-subscriber RLS-bound refetch.
2275
- const row = event.op === "DELETE" ? null : { _rebase_invalidated: true };
2276
-
2277
- await this.notifyUpdate(path, id, row, databaseId, /* broadcast */ false, /* origin */ "cdc");
2278
- }
2279
-
2280
- /**
2281
- * Deliver a change on a many-to-many junction table as a change to the child
2282
- * lists it belongs to.
2283
- *
2284
- * Linking a tag to a post writes only `posts_tags`. That table backs no
2285
- * collection, so change capture dropped the event as unmapped and the
2286
- * subscribers of `posts/1/tags` never heard about it — every other write in
2287
- * the system was realtime, and this one silently was not. The junction row
2288
- * carries both ids, so it names its own paths exactly.
2289
- *
2290
- * Notifies the nested path rather than either endpoint collection, because
2291
- * invalidation walks *parent* paths and never child ones: telling `tags` it
2292
- * changed would not reach a subscription on `posts/1/tags`.
2293
- *
2294
- * Returns whether the table was recognised as a junction.
2295
- */
2296
- private async handleJunctionCdcEvent(event: CdcChangeEvent): Promise<boolean> {
2297
- const links = this.junctionLinkMap?.get(`${event.schema}.${event.table}`)
2298
- ?? this.junctionLinkMap?.get(event.table);
2299
- if (!links?.length) return false;
2300
-
2301
- for (const link of links) {
2302
- const sourceId = event.row?.[link.sourceColumn];
2303
- const targetId = event.row?.[link.targetColumn];
2304
- if (sourceId === undefined || sourceId === null || targetId === undefined || targetId === null) {
2305
- this.debugLog(
2306
- `📡 [CDC] Junction row on ${event.table} is missing '${link.sourceColumn}'/'${link.targetColumn}' — skipping.`
2307
- );
2308
- continue;
2309
- }
2310
-
2311
- const path = `${link.parentCollection.slug}/${String(sourceId)}/${link.relationKey}`;
2312
- // An unlink removes the target from this list; a link invalidates it
2313
- // so each subscriber refetches under its own RLS context.
2314
- const row = event.op === "DELETE" ? null : { _rebase_invalidated: true };
2315
-
2316
- await this.notifyUpdate(
2317
- path,
2318
- String(targetId),
2319
- row,
2320
- (link.parentCollection as { databaseId?: string }).databaseId,
2321
- /* broadcast */ false,
2322
- /* origin */ "cdc"
2323
- );
2324
- }
2325
-
2326
- return true;
2327
- }
2328
-
2329
- /** Compute the canonical (possibly composite) id string from a captured row. */
2330
- private extractIdFromCdcRow(collection: CollectionConfig, row: Record<string, unknown>): string {
2331
- // Unaddressable falls back to a collection-level invalidation: single-row
2332
- // subs won't match, but collection subs still refetch.
2333
- return deriveRowAddress(row, collection, this.registry) || "*";
2334
- }
2335
-
2336
- // ── App/CDC de-duplication ──
2337
-
2338
- private dedupKey(path: string, id: string, databaseId?: string): string {
2339
- return `${databaseId ?? ""}::${path}::${id}`;
2340
- }
2341
-
2342
- /** Record that this instance just delivered `key` via the app path. */
2343
- private markAppEmit(key: string): void {
2344
- const now = Date.now();
2345
- this.recentAppEmits.set(key, now + RealtimeService.CDC_DEDUP_WINDOW_MS);
2346
- // Opportunistic purge so the map cannot grow unbounded under write load.
2347
- if (this.recentAppEmits.size > 1000) {
2348
- for (const [k, expiry] of this.recentAppEmits) {
2349
- if (expiry <= now) this.recentAppEmits.delete(k);
2350
- }
2351
- }
2352
- }
2353
-
2354
- /** Consume a matching app-emit record if present and unexpired; true ⇒ suppress the CDC echo. */
2355
- private consumeAppEmit(key: string): boolean {
2356
- const expiry = this.recentAppEmits.get(key);
2357
- if (expiry === undefined) return false;
2358
- this.recentAppEmits.delete(key);
2359
- return expiry > Date.now();
2360
- }
2361
-
2362
- // =============================================================================
2363
- // Cross-Instance LISTEN/NOTIFY
2364
- // =============================================================================
2365
-
2366
- /**
2367
- * Enable cross-instance realtime broadcasting via Postgres LISTEN/NOTIFY.
2368
- * Creates a dedicated pg.Client (outside the Drizzle pool) that stays
2369
- * connected and listens for change notifications from other instances.
2370
- *
2371
- * This is an **optional** feature — if never called, the backend operates
2372
- * in single-instance mode (the default, perfectly fine for most setups).
2373
- *
2374
- * @param connectionString Raw Postgres connection string for the LISTEN client.
2375
- */
2376
- async startListening(connectionString: string): Promise<void> {
2377
- if (this.broadcasting) {
2378
- logger.warn("⚠️ [RealtimeService] startListening called but already listening. Ignoring.");
2379
- return;
2380
- }
2381
-
2382
- this.listenConnectionString = connectionString;
2383
- // Set broadcasting BEFORE connecting so that scheduleReconnect()
2384
- // works correctly if the initial connection attempt fails.
2385
- this.broadcasting = true;
2386
- await this.connectListenClient();
2387
- logger.info(`📡 [RealtimeService] Cross-instance realtime enabled (instanceId: ${this.instanceId})`);
2388
- }
2389
-
2390
- /**
2391
- * Stop listening and clean up the dedicated LISTEN connection.
2392
- */
2393
- async stopListening(): Promise<void> {
2394
- this.broadcasting = false;
2395
- if (this.reconnectTimer) {
2396
- clearTimeout(this.reconnectTimer);
2397
- this.reconnectTimer = undefined;
2398
- }
2399
- if (this.listenClient) {
2400
- try {
2401
- await this.listenClient.end();
2402
- } catch { /* ignore close errors */ }
2403
- this.listenClient = undefined;
2404
- }
2405
- logger.info("📡 [RealtimeService] Cross-instance realtime disabled.");
2406
- }
2407
-
2408
- /**
2409
- * Broadcast a change notification to other instances via pg_notify.
2410
- * Uses the main Drizzle connection (pooled) — NOT the LISTEN client.
2411
- */
2412
- private async broadcastChange(path: string, id: string, databaseId?: string): Promise<void> {
2413
- const payload = JSON.stringify({
2414
- sid: this.instanceId,
2415
- p: path,
2416
- eid: id,
2417
- db: databaseId ?? null
2418
- });
2419
- await this.db.execute(drizzleSql`SELECT pg_notify(${PG_NOTIFY_CHANNEL}, ${payload})`);
2420
- }
2421
-
2422
- /**
2423
- * Create and connect the dedicated LISTEN client with auto-reconnect.
2424
- */
2425
- private async connectListenClient(): Promise<void> {
2426
- if (!this.listenConnectionString) return;
2427
-
2428
- let pending: PgClient | undefined;
2429
- try {
2430
- // See `PgNotifyListener.connect` — same shape, same reason. Until
2431
- // `this.listenClient` is assigned, nothing else in this class knows
2432
- // the connection exists, so a throw between `connect()` and that
2433
- // assignment leaks a live backend and `scheduleReconnect` opens
2434
- // another one three seconds later.
2435
- const client = new PgClient({ connectionString: this.listenConnectionString });
2436
- pending = client;
2437
-
2438
- client.on("error", (err) => {
2439
- logger.error("❌ [RealtimeService] LISTEN client error", { detail: err.message });
2440
- this.scheduleReconnect();
2441
- });
2442
-
2443
- client.on("end", () => {
2444
- if (this.broadcasting) {
2445
- logger.warn("⚠️ [RealtimeService] LISTEN client disconnected unexpectedly.");
2446
- this.scheduleReconnect();
2447
- }
2448
- });
2449
-
2450
- client.on("notification", async (msg) => {
2451
- if (!msg.payload) return;
2452
- try {
2453
- const { sid, p, eid, db } = JSON.parse(msg.payload) as {
2454
- sid: string;
2455
- p: string;
2456
- eid: string;
2457
- db: string | null;
2458
- };
2459
-
2460
- // Skip our own notifications — already processed locally
2461
- if (sid === this.instanceId) return;
2462
-
2463
- // A foreign sid is proof of a second process. Nothing here
2464
- // needs that fact, but the channel path does — see
2465
- // `warnIfMemoryBusOnMultiplePods`.
2466
- this.foreignInstanceSeen = true;
2467
-
2468
- this.debugLog(`📡 [RealtimeService] Received cross-instance notification: path=${p}, id=${eid}, from=${sid}`);
2469
-
2470
- // Refetch the row from the DB so row subscriptions
2471
- // receive the actual data instead of null (which the client
2472
- // would interpret as "deleted").
2473
- let refetchedRow: Record<string, unknown> | null = null;
2474
- try {
2475
- if (this.driver) {
2476
- const collection = this.registry.getCollectionByPath(p);
2477
- const fetched = await this.driver.fetchOne({
2478
- path: p,
2479
- id: eid,
2480
- collection: collection
2481
- });
2482
- refetchedRow = fetched ?? null;
2483
- } else {
2484
- const fetched = await this.dataService.fetchOne(
2485
- p, eid, db ?? undefined
2486
- );
2487
- refetchedRow = fetched ?? null;
2488
- }
2489
- } catch (fetchErr) {
2490
- // If the fetch fails (e.g. row was deleted), refetchedRow stays null
2491
- this.debugLog(`📡 [RealtimeService] Could not refetch row ${eid} from ${p} — treating as deleted`, fetchErr);
2492
- }
2493
-
2494
- // Trigger local fan-out with broadcast=false to avoid re-broadcasting
2495
- await this.notifyUpdate(p, eid, refetchedRow, db ?? undefined, false);
2496
- } catch (err) {
2497
- logger.error("❌ [RealtimeService] Error processing cross-instance notification", { error: err });
2498
- }
2499
- });
2500
-
2501
- await client.connect();
2502
- await client.query(`LISTEN ${PG_NOTIFY_CHANNEL}`);
2503
- this.listenClient = client;
2504
- // Adopted: `destroy()` and `scheduleReconnect` close it now.
2505
- pending = undefined;
2506
-
2507
- this.debugLog(`📡 [RealtimeService] LISTEN client connected on channel "${PG_NOTIFY_CHANNEL}"`);
2508
- } catch (err) {
2509
- if (pending) {
2510
- try { await pending.end(); } catch { /* already dead */ }
2511
- }
2512
- logger.error("❌ [RealtimeService] Failed to connect LISTEN client", { error: err });
2513
- this.scheduleReconnect();
2514
- }
2515
- }
2516
-
2517
- /**
2518
- * Schedule a reconnection attempt with a fixed 3s delay.
2519
- */
2520
- private scheduleReconnect(): void {
2521
- if (!this.broadcasting || this.reconnectTimer) return;
2522
-
2523
- const delay = 3000; // Fixed 3s delay; simple and predictable
2524
- this.debugLog(`📡 [RealtimeService] Scheduling LISTEN reconnect in ${delay}ms...`);
2525
-
2526
- this.reconnectTimer = setTimeout(async () => {
2527
- this.reconnectTimer = undefined;
2528
- if (!this.broadcasting) return;
2529
-
2530
- // Clean up old client
2531
- if (this.listenClient) {
2532
- try { await this.listenClient.end(); } catch { /* ignore */ }
2533
- this.listenClient = undefined;
2534
- }
2535
-
2536
- await this.connectListenClient();
2537
- }, delay);
2538
- }
2539
- }
2540
-
2541
- /**
2542
- * Alias for RealtimeService for consistent naming with other database implementations.
2543
- * This allows code to use PostgresRealtimeProvider alongside future MongoRealtimeProvider, etc.
2544
- */
2545
- export const PostgresRealtimeProvider = RealtimeService;