Firebase cold-storage approach

.github/workflows/deploy.yml

@@ -190,9 +190,11 @@ jobs:
             # deploy/update stack
             sudo env IMAGE_TAG="$IMAGE_TAG" docker stack deploy -c stack.yml --with-registry-auth --resolve-image always --prune pubpub
 
-            # show progress and cleanup
+            # show progress
             sudo docker stack services pubpub
-            sudo docker image prune -f
+
+            # Remove images not used by running containers, keeping the last 72h for rollback
+            sudo docker image prune -a --filter "until=72h" -f
 
             # wait until rollout is complete and then clear cache
             wait_rollout() {

.gitignore

@@ -63,6 +63,7 @@ dist
 workers/tasks/export/styles/printDocument.css
 pubpub-postgres-test/
 pubpub-localdb/
+firebase-export-*/
 .pubpub_repl_history*
 tsconfig.tsbuildinfo
 .jest/secret-env.js

client/components/Editor/Editor.stories.tsx

@@ -162,6 +162,7 @@ storiesOf('Editor', module)
 							updatechangeObject(evt);
 						}}
 						collaborativeOptions={{
+							pubId: 'storybook-pub-id',
 							firebaseRef: draftRef as any,
 							clientData,
 							initialDocKey: -1,
@@ -237,6 +238,7 @@ storiesOf('Editor', module)
 							}
 						}}
 						collaborativeOptions={{
+							pubId: 'storybook-pub-id',
 							firebaseRef: draftRef as any,
 							clientData,
 							initialDocKey: -1,

client/components/Editor/plugins/collaborative/document.ts

@@ -45,6 +45,7 @@ export default (
 ) => {
 	const { collaborativeOptions, isReadOnly, onError = noop } = options;
 	const {
+		pubId,
 		firebaseRef: ref,
 		onStatusChange = noop,
 		onUpdateLatestKey = noop,
@@ -100,7 +101,7 @@ export default (
 					/* If multiple of saveEveryNSteps, update checkpoint */
 					const saveEveryNSteps = 100;
 					if (snapshot.key && snapshot.key % saveEveryNSteps === 0) {
-						storeCheckpoint(ref, newState.doc, snapshot.key);
+						storeCheckpoint(pubId, newState.doc, snapshot.key);
 					}
 				}
 

client/components/Editor/types.ts

@@ -42,6 +42,7 @@ export type CollaborativeOptions = {
 	clientData: {
 		id: null | string;
 	};
+	pubId: string;
 	firebaseRef: firebase.database.Reference;
 	initialDocKey: number;
 	onStatusChange?: (status: CollaborativeEditorStatus) => unknown;

client/components/Editor/utils/firebase.ts

@@ -4,26 +4,30 @@ import type { Step } from 'prosemirror-transform';
 
 import type { CompressedChange, CompressedKeyable } from '../types';
 
-import { compressStateJSON, compressStepJSON } from 'prosemirror-compress-pubpub';
+import { compressStepJSON } from 'prosemirror-compress-pubpub';
 import uuid from 'uuid';
 
+import { apiFetch } from 'client/utils/apiFetch';
+
 export const firebaseTimestamp = { '.sv': 'timestamp' };
 
-export const storeCheckpoint = async (
-	firebaseRef: firebase.database.Reference,
-	doc: Node,
-	keyNumber: number,
-) => {
-	const checkpoint = {
-		d: compressStateJSON({ doc: doc.toJSON() }).d,
-		k: keyNumber,
-		t: firebaseTimestamp,
-	};
-	await Promise.all([
-		firebaseRef.child(`checkpoints/${keyNumber}`).set(checkpoint),
-		firebaseRef.child('checkpoint').set(checkpoint),
-		firebaseRef.child(`checkpointMap/${keyNumber}`).set(firebaseTimestamp),
-	]);
+/**
+ * Store a checkpoint by writing the doc to Postgres via the server API.
+ * Firebase checkpoints are no longer written — Postgres is the single
+ * source of truth for checkpoints.
+ */
+export const storeCheckpoint = async (pubId: string, doc: Node, keyNumber: number) => {
+	try {
+		await apiFetch.post('/api/draftCheckpoint', {
+			pubId,
+			historyKey: keyNumber,
+			doc: doc.toJSON(),
+		});
+	} catch (err) {
+		// Non-fatal: the checkpoint is an optimization, not required for correctness.
+		// The next checkpoint attempt (100 steps later) will try again.
+		console.error('Failed to store checkpoint:', err);
+	}
 };
 
 export const flattenKeyables = (

client/containers/Pub/PubDocument/PubBody.tsx

@@ -33,6 +33,7 @@ const markSentryError = (err: Error) => {
 const PubBody = (props: Props) => {
 	const { editorWrapperRef } = props;
 	const {
+		pubData,
 		noteManager,
 		updateCollabData,
 		historyData: { setLatestHistoryKey },
@@ -84,6 +85,7 @@ const PubBody = (props: Props) => {
 
 	const collaborativeOptions = includeCollabPlugin &&
 		!!firebaseDraftRef && {
+			pubId: pubData.id,
 			initialDocKey: initialHistoryKey,
 			firebaseRef: firebaseDraftRef,
 			clientData: localCollabUser,

server/apiRoutes.ts

@@ -12,6 +12,7 @@ import { router as customScriptRouter } from './customScript/api';
 import { router as devApiRouter } from './dev/api';
 import { router as discussionRouter } from './discussion/api';
 import { router as doiRouter } from './doi/api';
+import { router as draftCheckpointRouter } from './draftCheckpoint/api';
 import { router as editorRouter } from './editor/api';
 import { router as integrationDataOAuth1Router } from './integrationDataOAuth1/api';
 import { router as landingPageFeatureRouter } from './landingPageFeature/api';
@@ -46,6 +47,7 @@ const apiRouter = Router()
 	.use(customScriptRouter)
 	.use(discussionRouter)
 	.use(doiRouter)
+	.use(draftCheckpointRouter)
 	.use(editorRouter)
 	.use(integrationDataOAuth1Router)
 	.use(landingPageFeatureRouter)

server/draft/model.ts

@@ -10,7 +10,6 @@ import {
 	Model,
 	PrimaryKey,
 	Table,
-	//	HasOne,
 } from 'sequelize-typescript';
 // import { Pub } from '../models';
 
@@ -29,7 +28,4 @@ export class Draft extends Model<InferAttributes<Draft>, InferCreationAttributes
 	@AllowNull(false)
 	@Column(DataType.STRING)
 	declare firebasePath: string;
-
-	// @HasOne(() => Pub, { as: 'pub', foreignKey: 'draftId' })
-	// pub?: Pub;
 }

server/draftCheckpoint/api.ts

@@ -0,0 +1,39 @@
+import { Router } from 'express';
+
+import { Draft, Pub } from 'server/models';
+import { wrap } from 'server/wrap';
+import { expect } from 'utils/assert';
+
+import { upsertDraftCheckpoint } from './queries';
+
+export const router = Router();
+
+router.post(
+	'/api/draftCheckpoint',
+	wrap(async (req, res) => {
+		const userId = req.user?.id;
+		if (!userId) {
+			return res.status(401).json({});
+		}
+
+		const { pubId, historyKey: rawHistoryKey, doc } = req.body;
+		const historyKey =
+			typeof rawHistoryKey === 'string' ? parseInt(rawHistoryKey, 10) : rawHistoryKey;
+		if (!pubId || typeof historyKey !== 'number' || Number.isNaN(historyKey) || !doc) {
+			return res.status(400).json({ error: 'Missing pubId, historyKey, or doc' });
+		}
+
+		// Look up the draft for this pub
+		const pub = await Pub.findOne({
+			where: { id: pubId },
+			include: [{ model: Draft, as: 'draft' }],
+		});
+		if (!pub?.draft) {
+			return res.status(404).json({ error: 'Pub or draft not found' });
+		}
+
+		const checkpoint = await upsertDraftCheckpoint(pub.draft.id, historyKey, doc, Date.now());
+
+		return res.status(200).json({ id: checkpoint.id });
+	}),
+);

server/draftCheckpoint/model.ts

@@ -0,0 +1,73 @@
+import type { CreationOptional, InferAttributes, InferCreationAttributes } from 'sequelize';
+
+import type { SerializedModel } from 'types';
+
+import {
+	AllowNull,
+	BelongsTo,
+	Column,
+	DataType,
+	Default,
+	Index,
+	Model,
+	PrimaryKey,
+	Table,
+} from 'sequelize-typescript';
+
+import { Draft } from '../models';
+
+@Table
+export class DraftCheckpoint extends Model<
+	InferAttributes<DraftCheckpoint>,
+	InferCreationAttributes<DraftCheckpoint>
+> {
+	public declare toJSON: <M extends Model>(this: M) => SerializedModel<M>;
+
+	@Default(DataType.UUIDV4)
+	@PrimaryKey
+	@Column(DataType.UUID)
+	declare id: CreationOptional<string>;
+
+	@AllowNull(false)
+	@Index
+	@Column(DataType.UUID)
+	declare draftId: string;
+
+	// The history key this checkpoint represents (i.e. the doc state after applying
+	// all changes up to and including this key)
+	@AllowNull(false)
+	@Column(DataType.INTEGER)
+	declare historyKey: number;
+
+	// The compressed doc JSON (same shape as Doc.content — a ProseMirror doc JSON)
+	@AllowNull(false)
+	@Column(DataType.JSONB)
+	declare doc: Record<string, any>;
+
+	// Timestamp of the change at this history key
+	@Column(DataType.BIGINT)
+	declare timestamp: number | null;
+
+	// Firebase discussion positions at the time of cold storage, keyed by discussion ID.
+	// Stored so they can be "thawed" back into Firebase when the draft is next loaded.
+	@Column(DataType.JSONB)
+	declare discussions: Record<string, any> | null;
+
+	// Cumulative StepMap ranges from the latest release historyKey to this checkpoint's
+	// historyKey. Used to map discussion anchors during release creation when the
+	// original steps are no longer available in Firebase.
+	// Shape: Array<number[]> — each inner array is a StepMap.ranges (triples of
+	// [oldStart, oldSize, newSize]).
+	@Column(DataType.JSONB)
+	declare stepMaps: number[][] | null;
+
+	// The history key that stepMaps cover up to. After cold storage thaw + editing,
+	// the checkpoint's historyKey advances but stepMaps still only cover up to this key.
+	// At release time, Firebase changes from stepMapToKey+1 → currentKey are composed
+	// with the stored stepMaps.
+	@Column(DataType.INTEGER)
+	declare stepMapToKey: number | null;
+
+	@BelongsTo(() => Draft, { as: 'draft', foreignKey: 'draftId' })
+	declare draft?: Draft;
+}

server/draftCheckpoint/queries.ts

@@ -0,0 +1,61 @@
+import type { DocJson } from 'types';
+
+import { DraftCheckpoint } from 'server/models';
+
+/**
+ * Create or update the checkpoint for a draft.
+ * Each draft has at most one checkpoint — an upsert on draftId.
+ */
+export const upsertDraftCheckpoint = async (
+	draftId: string,
+	historyKey: number,
+	doc: DocJson,
+	timestamp: number | null = null,
+	sequelizeTransaction: any = null,
+	options: {
+		discussions?: Record<string, any> | null;
+		stepMaps?: number[][] | null;
+		stepMapToKey?: number | null;
+	} = {},
+) => {
+	// Only include optional fields in the update if they were explicitly provided.
+	// This prevents normal checkpoint writes (from the client API) from clobbering
+	// stepMaps/discussions that were set during cold storage.
+	const optionalFields: Partial<
+		Pick<DraftCheckpoint, 'discussions' | 'stepMaps' | 'stepMapToKey'>
+	> = {};
+	if ('discussions' in options) optionalFields.discussions = options.discussions ?? null;
+	if ('stepMaps' in options) optionalFields.stepMaps = options.stepMaps ?? null;
+	if ('stepMapToKey' in options) optionalFields.stepMapToKey = options.stepMapToKey ?? null;
+
+	const existing = await DraftCheckpoint.findOne({
+		where: { draftId },
+		transaction: sequelizeTransaction,
+	});
+
+	if (existing) {
+		// Only update if the new key is more recent
+		if (historyKey > existing.historyKey) {
+			await existing.update(
+				{ historyKey, doc, timestamp, ...optionalFields },
+				{ transaction: sequelizeTransaction },
+			);
+		}
+		return existing;
+	}
+
+	return DraftCheckpoint.create(
+		{ draftId, historyKey, doc, timestamp, ...optionalFields },
+		{ transaction: sequelizeTransaction },
+	);
+};
+
+/**
+ * Get the checkpoint for a draft, if one exists.
+ */
+export const getDraftCheckpoint = async (draftId: string, sequelizeTransaction: any = null) => {
+	return DraftCheckpoint.findOne({
+		where: { draftId },
+		transaction: sequelizeTransaction,
+	});
+};

server/models.ts

@@ -18,6 +18,7 @@ import { Discussion } from './discussion/model';
 import { DiscussionAnchor } from './discussionAnchor/model';
 import { Doc } from './doc/model';
 import { Draft } from './draft/model';
+import { DraftCheckpoint } from './draftCheckpoint/model';
 import { EmailChangeToken } from './emailChangeToken/model';
 import { Export } from './export/model';
 import { ExternalPublication } from './externalPublication/model';
@@ -78,6 +79,7 @@ sequelize.addModels([
 	DiscussionAnchor,
 	Doc,
 	Draft,
+	DraftCheckpoint,
 	EmailChangeToken,
 	Export,
 	ExternalPublication,
@@ -174,6 +176,7 @@ export {
 	EmailChangeToken,
 	Doc,
 	Draft,
+	DraftCheckpoint,
 	Export,
 	ExternalPublication,
 	FeatureFlag,

server/pubHistory/queries.ts

@@ -14,7 +14,7 @@ export const restorePubDraftToHistoryKey = async (options: RestorePubOptions) =>
 	const { pubId, userId, historyKey } = options;
 	assert(typeof historyKey === 'number' && historyKey >= 0);
 	const pubDraftRef = await getPubDraftRef(pubId);
-	const { doc } = await getPubDraftDoc(pubDraftRef, historyKey);
+	const { doc } = await getPubDraftDoc(pubId, historyKey);
 	const editor = await editFirebaseDraftByRef(pubDraftRef, userId);
 
 	editor.transform((tr, schema) => {