From 9ebbea584d686f62588dac399dfd4fabcf2e75f0 Mon Sep 17 00:00:00 2001
From: Dan Stillman <dstillman@zotero.org>
Date: Thu, 30 Jul 2026 21:14:37 -0400
Subject: [PATCH] Read stored embedding hashes in one query per chunk

Indexing re-enqueues every eligible item on each start to find what
changed, and checked each one's stored hash with its own query, so a
large library ran thousands of queries at startup.
---
 chrome/content/zotero/xpcom/embeddings.js | 33 ++++++++++++++----
 test/tests/embeddingsTest.js              | 41 +++++++++++++++++++++--
 2 files changed, 65 insertions(+), 9 deletions(-)

diff --git a/chrome/content/zotero/xpcom/embeddings.js b/chrome/content/zotero/xpcom/embeddings.js
index 40861c26f1..b48cfbbd25 100644
--- a/chrome/content/zotero/xpcom/embeddings.js
+++ b/chrome/content/zotero/xpcom/embeddings.js
@@ -1099,23 +1099,42 @@ Zotero.Embeddings.Indexing = new function () {
 	} = {}) {
 		await Zotero.Items.loadDataTypes(items, ['itemData']);
 
+		// Every start re-enqueues the whole library to find what changed, so
+		// read the stored hashes in one query per chunk rather than one per
+		// item
+		let storedHashes = new Map();
+		let itemIDs = items.map(item => item.id);
+		let chunkSize = 500;
+		for (let i = 0; i < itemIDs.length; i += chunkSize) {
+			let chunk = itemIDs.slice(i, i + chunkSize);
+			let rows = await Zotero.DB.queryAsync(
+				"SELECT itemID, sourceHash FROM embeddings.itemEmbeddings WHERE itemID IN ("
+					+ chunk.map(() => '?').join(',') + ")",
+				chunk
+			);
+			for (let row of rows) {
+				storedHashes.set(row.itemID, row.sourceHash);
+			}
+		}
+
 		let toEmbed = [];
+		let toDelete = [];
 		for (let item of items) {
 			let text = _getItemText(item);
 			if (!text) {
-				await Zotero.DB.queryAsync(
-					"DELETE FROM embeddings.itemEmbeddings WHERE itemID=?", item.id
-				);
+				if (storedHashes.has(item.id)) {
+					toDelete.push(item.id);
+				}
 				continue;
 			}
 			let hash = Zotero.Utilities.Internal.md5(text);
-			let existing = await Zotero.DB.valueQueryAsync(
-				"SELECT sourceHash FROM embeddings.itemEmbeddings WHERE itemID=?", item.id
-			);
-			if (existing !== hash) {
+			if (storedHashes.get(item.id) !== hash) {
 				toEmbed.push({ item, text, hash });
 			}
 		}
+		if (toDelete.length) {
+			await _deleteEmbeddings(toDelete);
+		}
 
 		// Process one library at a time
 		toEmbed.sort((a, b) => (a.item.libraryID - b.item.libraryID)
diff --git a/test/tests/embeddingsTest.js b/test/tests/embeddingsTest.js
index 1157ff6f03..fd98dea240 100644
--- a/test/tests/embeddingsTest.js
+++ b/test/tests/embeddingsTest.js
@@ -196,9 +196,45 @@ describe("Zotero.Embeddings", function () {
 				stub.restore();
 			}
 		});
+
+		it("should look up stored hashes without a query per item", async function () {
+			this.timeout(60000);
+			for (let i = 0; i < 5; i++) {
+				await createDataObject('item', { title: "Batched lookup " + i });
+			}
+
+			let vector = new Float32Array(4).fill(0.5);
+			let embedStub = sinon.stub(Zotero.Embeddings, 'embedPassages')
+				.callsFake(async texts => texts.map(() => vector));
+			let stubs = [
+				embedStub,
+				sinon.stub(Zotero.Embeddings, 'isEnabled').returns(true),
+				sinon.stub(Zotero.Embeddings, 'getModelVersion').returns('test-model/1'),
+				sinon.stub(Zotero.Embeddings, 'isDownloaded').resolves(true),
+				sinon.stub(Zotero.Embeddings, 'preloadModel').resolves()
+			];
+			let queries = [];
+			let queryStub = sinon.stub(Zotero.DB, 'queryAsync')
+				.callsFake(function (sql, ...rest) {
+					queries.push(sql);
+					return queryStub.wrappedMethod.call(this, sql, ...rest);
+				});
+			try {
+				await Zotero.Embeddings.Indexing.startIndexing();
+			}
+			finally {
+				queryStub.restore();
+				stubs.forEach(stub => stub.restore());
+			}
+
+			// The run has to have indexed something for this to mean anything
+			assert.isTrue(embedStub.called);
+			assert.isEmpty(queries.filter(sql => sql.includes('sourceHash')
+				&& sql.includes('itemID=?')));
+			assert.isNotEmpty(queries.filter(sql => sql.includes('itemID, sourceHash')));
+		});
 	});
-	// Downloads the active model (~130 MB), so run explicitly:
-	// ZOTERO_TEST_EMBEDDINGS_INFERENCE=1 test/runtests.sh -g "real vectors" embeddings
+
 	describe("#embed() with a real model", function () {
 		before(function () {
 			if (!Services.env.get("ZOTERO_TEST_EMBEDDINGS_INFERENCE")) {
@@ -244,6 +280,7 @@ describe("Zotero.Embeddings", function () {
 			assert.isTrue(await Zotero.Embeddings.isDownloaded());
 		});
 	});
+
 	describe("memory pressure", function () {
 		afterEach(function () {
 			Services.obs.notifyObservers(null, 'memory-pressure-stop');