← All changes | admin/class-pinecone-manager.php +99 -51 3.2.13 → trunk View file →

		@@ -493,8 +493,12 @@
493	493	$namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
494	494
495	495	// Pinecone's list endpoint returns vector IDs with pagination
496	496	// We then fetch the metadata for those specific IDs
	497	+ // NOTE (plan 793b82): /vectors/list is a GET endpoint with query parameters.
	498	+ // A POST is answered 200-with-an-empty-body, which reads as "no vectors" —
	499	+ // this call used to POST and only ever "worked" because the empty result
	500	+ // tripped the query-based fallback below.
497	501	$list_url = "https://{$host}/vectors/list";
498	502
499	503	// Calculate pagination token from page number
500	504	// Pinecone uses cursor-based pagination, so we need to handle this differently
		@@ -518,19 +522,20 @@
518	522	$list_params['paginationToken'] = $stored_tokens[$page];
519	523	}
520	524	}
521	525
522		- $response = wp_remote_post($list_url, array(
	526	+ $response = wp_remote_get($list_url . '?' . http_build_query($list_params), array(
523	527	'headers' => array(
524	528	'Api-Key' => $api_key,
525		- 'Content-Type' => 'application/json'
	529	+ 'accept' => 'application/json'
526	530	),
527		- 'body' => json_encode($list_params),
528	531	'timeout' => 15
529	532	));
530	533
531	534	if (is_wp_error($response) || wp_remote_retrieve_response_code($response) !== 200) {
532		- // Fallback to query-based approach
	535	+ // Genuine transport/API failure — fall back to the query-based approach,
	536	+ // but say so: a silent fallback is what masked the POST bug for a year.
	537	+ error_log('MxChat Pinecone: /vectors/list GET failed (' . (is_wp_error($response) ? $response->get_error_message() : wp_remote_retrieve_response_code($response)) . ') — falling back to query-based listing.');
533	538	return $this->mxchat_query_based_list($pinecone_options, $page, $per_page, $bot_id, $content_type);
534	539	}
535	540
536	541	$body = wp_remote_retrieve_body($response);
		@@ -549,15 +554,26 @@
549	554	$vector_ids[] = $vector['id'];
550	555	}
551	556	}
552	557
553		- // If list endpoint returned empty, fall back to query-based approach
	558	+ // A 200 with no vector ids is a REAL answer now (the namespace is empty),
	559	+ // not the failure signature it was under POST — return the empty listing
	560	+ // rather than routing into the fallback and masking a broken primary.
554	561	if (empty($vector_ids)) {
	562	+ return array('data' => array(), 'total' => 0);
	563	+ }
	564	+
	565	+ // Fetch metadata for these vector IDs
	566	+ $fetched = $this->mxchat_fetch_vectors_by_ids_for_list($pinecone_options, $vector_ids, $page, $per_page, $bot_id, $content_type);
	567	+
	568	+ // null = the fetch itself failed in transport (distinct from "ids had no
	569	+ // metadata matches") — log and fall back so the screen still lists.
	570	+ if ($fetched === null) {
	571	+ error_log('MxChat Pinecone: /vectors/fetch GET failed for the listing — falling back to query-based listing.');
555	572	return $this->mxchat_query_based_list($pinecone_options, $page, $per_page, $bot_id, $content_type);
556	573	}
557	574
558		- // Fetch metadata for these vector IDs
559		- return $this->mxchat_fetch_vectors_by_ids_for_list($pinecone_options, $vector_ids, $page, $per_page, $bot_id, $content_type);
	575	+ return $fetched;
560	576	}
561	577
562	578	/**
563	579	* Query-based listing fallback when list endpoint fails
		@@ -661,29 +677,31 @@
661	677	$api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? '';
662	678	$host = $pinecone_options['mxchat_pinecone_host'] ?? '';
663	679	$namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
664	680
665		- $fetch_url = "https://{$host}/vectors/fetch";
666		-
667		- $fetch_data = array(
668		- 'ids' => $vector_ids
669		- );
670		-
	681	+ // NOTE (plan 793b82): /vectors/fetch is a GET endpoint, and Pinecone expects
	682	+ // the ids repeated (ids=a&ids=b) — http_build_query would emit ids[0]=a.
	683	+ // A POST here is answered 200-with-an-empty-body, which reads as "no vectors".
	684	+ $fetch_query = array();
	685	+ foreach ($vector_ids as $fetch_vid) {
	686	+ $fetch_query[] = 'ids=' . rawurlencode($fetch_vid);
	687	+ }
671	688	if (!empty($namespace)) {
672		- $fetch_data['namespace'] = $namespace;
	689	+ $fetch_query[] = 'namespace=' . rawurlencode($namespace);
673	690	}
674	691
675		- $response = wp_remote_post($fetch_url, array(
	692	+ $response = wp_remote_get("https://{$host}/vectors/fetch?" . implode('&', $fetch_query), array(
676	693	'headers' => array(
677	694	'Api-Key' => $api_key,
678		- 'Content-Type' => 'application/json'
	695	+ 'accept' => 'application/json'
679	696	),
680		- 'body' => json_encode($fetch_data),
681	697	'timeout' => 15
682	698	));
683	699
	700	+ // Transport/API failure is distinct from "nothing matched": return null so
	701	+ // the caller can tell the difference (plan 793b82's empty-vs-failed rule).
684	702	if (is_wp_error($response) || wp_remote_retrieve_response_code($response) !== 200) {
685		- return array('data' => array(), 'total' => 0);
	703	+ return null;
686	704	}
687	705
688	706	$body = wp_remote_retrieve_body($response);
689	707	$data = json_decode($body, true);
		@@ -1104,31 +1122,45 @@
1104	1122	if (empty($vector_id_map)) {
1105	1123	return array();
1106	1124	}
1107	1125
1108		- // Batch check Pinecone using fetch API (max 1000 IDs per request)
	1126	+ // Batch check Pinecone using fetch API.
	1127	+ // NOTE (plan 793b82): /vectors/fetch is a GET endpoint with the ids
	1128	+ // repeated in the query string (ids=a&ids=b); the old POST here was
	1129	+ // answered 200-with-an-empty-body, so this scan saw NOTHING as indexed
	1130	+ // and callers re-embedded content Pinecone already had. Chunk size 100:
	1131	+ // measured on a live serverless index, ~9KB of URL is accepted and
	1132	+ // ~18KB draws HTTP 414, so 100 32-char ids (~4KB) leaves real margin.
	1133	+ $namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
1109	1134	$all_vector_ids = array_keys($vector_id_map);
1110		- $chunks = array_chunk($all_vector_ids, 1000);
	1135	+ $chunks = array_chunk($all_vector_ids, 100);
1111	1136	$processed_data = array();
1112	1137
1113	1138	foreach ($chunks as $chunk) {
1114		- $fetch_url = "https://{$host}/vectors/fetch";
	1139	+ $fetch_query = array();
	1140	+ foreach ($chunk as $fetch_vid) {
	1141	+ $fetch_query[] = 'ids=' . rawurlencode($fetch_vid);
	1142	+ }
	1143	+ if (!empty($namespace)) {
	1144	+ $fetch_query[] = 'namespace=' . rawurlencode($namespace);
	1145	+ }
1115	1146
1116		- $response = wp_remote_post($fetch_url, array(
	1147	+ $response = wp_remote_get("https://{$host}/vectors/fetch?" . implode('&', $fetch_query), array(
1117	1148	'headers' => array(
1118	1149	'Api-Key' => $api_key,
1119		- 'Content-Type' => 'application/json'
	1150	+ 'accept' => 'application/json'
1120	1151	),
1121		- 'body' => json_encode(array('ids' => $chunk)),
1122	1152	'timeout' => 30
1123	1153	));
1124	1154
1125	1155	if (is_wp_error($response)) {
	1156	+ error_log('MxChat Pinecone: processed-content scan /vectors/fetch GET failed: ' . $response->get_error_message());
1126	1157	continue;
1127	1158	}
1128	1159
1129	1160	$response_code = wp_remote_retrieve_response_code($response);
1130	1161	if ($response_code !== 200) {
	1162	+ error_log('MxChat Pinecone: processed-content scan /vectors/fetch GET returned HTTP ' . $response_code);
1131	1163	continue;
1132	1164	}
1133	1165
1134	1166	$body = wp_remote_retrieve_body($response);
		@@ -1311,8 +1343,9 @@
1311	1343	*/
1312	1344	public function fetch_pinecone_vectors_by_ids($pinecone_options, $vector_ids) {
1313	1345	$api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? '';
1314	1346	$host = $pinecone_options['mxchat_pinecone_host'] ?? '';
	1347	+ $namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
1315	1348
1316	1349	if (empty($api_key) || empty($host) || empty($vector_ids)) {
1317	1350	return array();
1318	1351	}
		@@ -1317,44 +1350,53 @@
1317	1350	return array();
1318	1351	}
1319	1352
1320	1353	try {
1321		- $fetch_url = "https://{$host}/vectors/fetch";
	1354	+ // NOTE (plan 793b82): /vectors/fetch is a GET endpoint with the ids
	1355	+ // repeated in the query string; the old POST here was answered
	1356	+ // 200-with-an-empty-body. Chunked at 100 ids to stay well under the
	1357	+ // measured HTTP 414 URL-length boundary.
	1358	+ $vectors = array();
	1359	+ foreach (array_chunk(array_values($vector_ids), 100) as $chunk) {
	1360	+ $fetch_query = array();
	1361	+ foreach ($chunk as $fetch_vid) {
	1362	+ $fetch_query[] = 'ids=' . rawurlencode($fetch_vid);
	1363	+ }
	1364	+ if (!empty($namespace)) {
	1365	+ $fetch_query[] = 'namespace=' . rawurlencode($namespace);
	1366	+ }
1322	1367
1323		- // Pinecone fetch API allows fetching specific vectors by ID
1324		- $fetch_data = array(
1325		- 'ids' => array_values($vector_ids)
1326		- );
	1368	+ $response = wp_remote_get("https://{$host}/vectors/fetch?" . implode('&', $fetch_query), array(
	1369	+ 'headers' => array(
	1370	+ 'Api-Key' => $api_key,
	1371	+ 'accept' => 'application/json'
	1372	+ ),
	1373	+ 'timeout' => 30
	1374	+ ));
1327	1375
1328		- $response = wp_remote_post($fetch_url, array(
1329		- 'headers' => array(
1330		- 'Api-Key' => $api_key,
1331		- 'Content-Type' => 'application/json'
1332		- ),
1333		- 'body' => json_encode($fetch_data),
1334		- 'timeout' => 30
1335		- ));
	1376	+ if (is_wp_error($response)) {
	1377	+ error_log('MxChat Pinecone: fetch_pinecone_vectors_by_ids GET failed: ' . $response->get_error_message());
	1378	+ continue;
	1379	+ }
1336	1380
1337		- if (is_wp_error($response)) {
1338		- return array();
1339		- }
	1381	+ if (wp_remote_retrieve_response_code($response) !== 200) {
	1382	+ error_log('MxChat Pinecone: fetch_pinecone_vectors_by_ids GET returned HTTP ' . wp_remote_retrieve_response_code($response));
	1383	+ continue;
	1384	+ }
1340	1385
1341		- $response_code = wp_remote_retrieve_response_code($response);
1342		-
1343		- if ($response_code !== 200) {
1344		- return array();
	1386	+ $data = json_decode(wp_remote_retrieve_body($response), true);
	1387	+ if (isset($data['vectors']) && is_array($data['vectors'])) {
	1388	+ $vectors += $data['vectors'];
	1389	+ }
1345	1390	}
1346	1391
1347		- $body = wp_remote_retrieve_body($response);
1348		- $data = json_decode($body, true);
1349		-
1350		- if (!isset($data['vectors'])) {
	1392	+ if (empty($vectors)) {
1351	1393	return array();
1352	1394	}
1353	1395
1354	1396	$processed_data = array();
1355	1397
1356		- foreach ($data['vectors'] as $vector_id => $vector_data) {
	1398	+ foreach ($vectors as $vector_id => $vector_data) {
1357	1399	$metadata = $vector_data['metadata'] ?? array();
1358	1400	$source_url = $metadata['source_url'] ?? '';
1359	1401
1360	1402	if (!empty($source_url)) {
		@@ -1397,8 +1439,9 @@
1397	1439	*/
1398	1440	public function mxchat_delete_all_from_pinecone($pinecone_options, $content_type_filter = '') {
1399	1441	$api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? '';
1400	1442	$host = $pinecone_options['mxchat_pinecone_host'] ?? '';
	1443	+ $namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
1401	1444
1402	1445	if (empty($api_key) || empty($host)) {
1403	1446	return array(
1404	1447	'success' => false,
		@@ -1443,9 +1486,9 @@
1443	1486	$batch_size = 100;
1444	1487	$batches = array_chunk($vector_ids, $batch_size);
1445	1488
1446	1489	foreach ($batches as $batch) {
1447		- $result = $this->mxchat_delete_pinecone_batch($batch, $api_key, $host);
	1490	+ $result = $this->mxchat_delete_pinecone_batch($batch, $api_key, $host, $namespace);
1448	1491	if ($result['success']) {
1449	1492	$total_deleted += count($batch);
1450	1493	} else {
1451	1494	$failed_batches++;
		@@ -1486,9 +1529,9 @@
1486	1529
1487	1530	/**
1488	1531	* Deletes batch of vectors from Pinecone database
1489	1532	*/
1490		- public function mxchat_delete_pinecone_batch($vector_ids, $api_key, $host) {
	1533	+ public function mxchat_delete_pinecone_batch($vector_ids, $api_key, $host, $namespace = '') {
1491	1534	// Build the API endpoint
1492	1535	$api_endpoint = "https://{$host}/vectors/delete";
1493	1536
1494	1537	// Prepare the request body with the IDs
		@@ -1494,8 +1537,13 @@
1494	1537	// Prepare the request body with the IDs
1495	1538	$request_body = array(
1496	1539	'ids' => $vector_ids
1497	1540	);
	1541	+
	1542	+ // Add namespace if provided — omitting it targets the default namespace
	1543	+ if (!empty($namespace)) {
	1544	+ $request_body['namespace'] = $namespace;
	1545	+ }
1498	1546
1499	1547	// Make the deletion request
1500	1548	$response = wp_remote_post($api_endpoint, array(
1501	1549	'headers' => array(