1use anyhow::Result;
2use http_client::HttpClient;
3use indoc::indoc;
4use open_ai::{
5 MessageContent, OPEN_AI_API_URL, Request as OpenAiRequest, RequestMessage,
6 Response as OpenAiResponse, batches, non_streaming_completion,
7};
8use reqwest_client::ReqwestClient;
9use sqlez::bindable::Bind;
10use sqlez::bindable::StaticColumnCount;
11use sqlez_macros::sql;
12use std::hash::Hash;
13use std::hash::Hasher;
14use std::path::Path;
15use std::sync::{Arc, Mutex};
16
17pub struct PlainOpenAiClient {
18 pub http_client: Arc<dyn HttpClient>,
19 pub api_key: String,
20}
21
22impl PlainOpenAiClient {
23 pub fn new() -> Result<Self> {
24 let http_client: Arc<dyn http_client::HttpClient> = Arc::new(ReqwestClient::new());
25 let api_key = std::env::var("OPENAI_API_KEY")
26 .map_err(|_| anyhow::anyhow!("OPENAI_API_KEY environment variable not set"))?;
27 Ok(Self {
28 http_client,
29 api_key,
30 })
31 }
32
33 pub async fn generate(
34 &self,
35 model: &str,
36 max_tokens: u64,
37 messages: Vec<RequestMessage>,
38 ) -> Result<OpenAiResponse> {
39 let request = OpenAiRequest {
40 model: model.to_string(),
41 messages,
42 stream: false,
43 stream_options: None,
44 max_completion_tokens: Some(max_tokens),
45 stop: Vec::new(),
46 temperature: None,
47 tool_choice: None,
48 parallel_tool_calls: None,
49 tools: Vec::new(),
50 prompt_cache_key: None,
51 reasoning_effort: None,
52 };
53
54 let response = non_streaming_completion(
55 self.http_client.as_ref(),
56 OPEN_AI_API_URL,
57 &self.api_key,
58 request,
59 )
60 .await
61 .map_err(|e| anyhow::anyhow!("{:?}", e))?;
62
63 Ok(response)
64 }
65}
66
67pub struct BatchingOpenAiClient {
68 connection: Mutex<sqlez::connection::Connection>,
69 http_client: Arc<dyn HttpClient>,
70 api_key: String,
71}
72
73struct CacheRow {
74 request_hash: String,
75 request: Option<String>,
76 response: Option<String>,
77 batch_id: Option<String>,
78}
79
80impl StaticColumnCount for CacheRow {
81 fn column_count() -> usize {
82 4
83 }
84}
85
86impl Bind for CacheRow {
87 fn bind(&self, statement: &sqlez::statement::Statement, start_index: i32) -> Result<i32> {
88 let next_index = statement.bind(&self.request_hash, start_index)?;
89 let next_index = statement.bind(&self.request, next_index)?;
90 let next_index = statement.bind(&self.response, next_index)?;
91 let next_index = statement.bind(&self.batch_id, next_index)?;
92 Ok(next_index)
93 }
94}
95
96#[derive(serde::Serialize, serde::Deserialize)]
97struct SerializableRequest {
98 model: String,
99 max_tokens: u64,
100 messages: Vec<SerializableMessage>,
101}
102
103#[derive(serde::Serialize, serde::Deserialize)]
104struct SerializableMessage {
105 role: String,
106 content: String,
107}
108
109impl BatchingOpenAiClient {
110 fn new(cache_path: &Path) -> Result<Self> {
111 let http_client: Arc<dyn http_client::HttpClient> = Arc::new(ReqwestClient::new());
112 let api_key = std::env::var("OPENAI_API_KEY")
113 .map_err(|_| anyhow::anyhow!("OPENAI_API_KEY environment variable not set"))?;
114
115 let connection = sqlez::connection::Connection::open_file(cache_path.to_str().unwrap());
116 let mut statement = sqlez::statement::Statement::prepare(
117 &connection,
118 indoc! {"
119 CREATE TABLE IF NOT EXISTS openai_cache (
120 request_hash TEXT PRIMARY KEY,
121 request TEXT,
122 response TEXT,
123 batch_id TEXT
124 );
125 "},
126 )?;
127 statement.exec()?;
128 drop(statement);
129
130 Ok(Self {
131 connection: Mutex::new(connection),
132 http_client,
133 api_key,
134 })
135 }
136
137 pub fn lookup(
138 &self,
139 model: &str,
140 max_tokens: u64,
141 messages: &[RequestMessage],
142 seed: Option<usize>,
143 ) -> Result<Option<OpenAiResponse>> {
144 let request_hash_str = Self::request_hash(model, max_tokens, messages, seed);
145 let connection = self.connection.lock().unwrap();
146 let response: Vec<String> = connection.select_bound(
147 &sql!(SELECT response FROM openai_cache WHERE request_hash = ?1 AND response IS NOT NULL;),
148 )?(request_hash_str.as_str())?;
149 Ok(response
150 .into_iter()
151 .next()
152 .and_then(|text| serde_json::from_str(&text).ok()))
153 }
154
155 pub fn mark_for_batch(
156 &self,
157 model: &str,
158 max_tokens: u64,
159 messages: &[RequestMessage],
160 seed: Option<usize>,
161 ) -> Result<()> {
162 let request_hash = Self::request_hash(model, max_tokens, messages, seed);
163
164 let serializable_messages: Vec<SerializableMessage> = messages
165 .iter()
166 .map(|msg| SerializableMessage {
167 role: message_role_to_string(msg),
168 content: message_content_to_string(msg),
169 })
170 .collect();
171
172 let serializable_request = SerializableRequest {
173 model: model.to_string(),
174 max_tokens,
175 messages: serializable_messages,
176 };
177
178 let request = Some(serde_json::to_string(&serializable_request)?);
179 let cache_row = CacheRow {
180 request_hash,
181 request,
182 response: None,
183 batch_id: None,
184 };
185 let connection = self.connection.lock().unwrap();
186 connection.exec_bound::<CacheRow>(sql!(
187 INSERT OR IGNORE INTO openai_cache(request_hash, request, response, batch_id) VALUES (?, ?, ?, ?)))?(
188 cache_row,
189 )
190 }
191
192 async fn generate(
193 &self,
194 model: &str,
195 max_tokens: u64,
196 messages: Vec<RequestMessage>,
197 seed: Option<usize>,
198 cache_only: bool,
199 ) -> Result<Option<OpenAiResponse>> {
200 let response = self.lookup(model, max_tokens, &messages, seed)?;
201 if let Some(response) = response {
202 return Ok(Some(response));
203 }
204
205 if !cache_only {
206 self.mark_for_batch(model, max_tokens, &messages, seed)?;
207 }
208
209 Ok(None)
210 }
211
212 async fn sync_batches(&self) -> Result<()> {
213 let _batch_ids = self.upload_pending_requests().await?;
214 self.download_finished_batches().await
215 }
216
217 pub async fn import_batches(&self, batch_ids: &[String]) -> Result<()> {
218 for batch_id in batch_ids {
219 log::info!("Importing OpenAI batch {}", batch_id);
220
221 let batch_status = batches::retrieve_batch(
222 self.http_client.as_ref(),
223 OPEN_AI_API_URL,
224 &self.api_key,
225 batch_id,
226 )
227 .await
228 .map_err(|e| anyhow::anyhow!("Failed to retrieve batch {}: {:?}", batch_id, e))?;
229
230 log::info!("Batch {} status: {}", batch_id, batch_status.status);
231
232 if batch_status.status != "completed" {
233 log::warn!(
234 "Batch {} is not completed (status: {}), skipping",
235 batch_id,
236 batch_status.status
237 );
238 continue;
239 }
240
241 let output_file_id = batch_status.output_file_id.ok_or_else(|| {
242 anyhow::anyhow!("Batch {} completed but has no output file", batch_id)
243 })?;
244
245 let results_content = batches::download_file(
246 self.http_client.as_ref(),
247 OPEN_AI_API_URL,
248 &self.api_key,
249 &output_file_id,
250 )
251 .await
252 .map_err(|e| {
253 anyhow::anyhow!("Failed to download batch results for {}: {:?}", batch_id, e)
254 })?;
255
256 let results = batches::parse_batch_output(&results_content)
257 .map_err(|e| anyhow::anyhow!("Failed to parse batch output: {:?}", e))?;
258
259 let mut updates: Vec<(String, String, String)> = Vec::new();
260 let mut success_count = 0;
261 let mut error_count = 0;
262
263 for result in results {
264 let request_hash = result
265 .custom_id
266 .strip_prefix("req_hash_")
267 .unwrap_or(&result.custom_id)
268 .to_string();
269
270 if let Some(response_body) = result.response {
271 if response_body.status_code == 200 {
272 let response_json = serde_json::to_string(&response_body.body)?;
273 updates.push((request_hash, response_json, batch_id.clone()));
274 success_count += 1;
275 } else {
276 log::error!(
277 "Batch request {} failed with status {}",
278 request_hash,
279 response_body.status_code
280 );
281 let error_json = serde_json::json!({
282 "error": {
283 "type": "http_error",
284 "status_code": response_body.status_code
285 }
286 })
287 .to_string();
288 updates.push((request_hash, error_json, batch_id.clone()));
289 error_count += 1;
290 }
291 } else if let Some(error) = result.error {
292 log::error!(
293 "Batch request {} failed: {}: {}",
294 request_hash,
295 error.code,
296 error.message
297 );
298 let error_json = serde_json::json!({
299 "error": {
300 "type": error.code,
301 "message": error.message
302 }
303 })
304 .to_string();
305 updates.push((request_hash, error_json, batch_id.clone()));
306 error_count += 1;
307 }
308 }
309
310 let connection = self.connection.lock().unwrap();
311 connection.with_savepoint("batch_import", || {
312 let q = sql!(
313 INSERT OR REPLACE INTO openai_cache(request_hash, request, response, batch_id)
314 VALUES (?, (SELECT request FROM openai_cache WHERE request_hash = ?), ?, ?)
315 );
316 let mut exec = connection.exec_bound::<(&str, &str, &str, &str)>(q)?;
317 for (request_hash, response_json, batch_id) in &updates {
318 exec((
319 request_hash.as_str(),
320 request_hash.as_str(),
321 response_json.as_str(),
322 batch_id.as_str(),
323 ))?;
324 }
325 Ok(())
326 })?;
327
328 log::info!(
329 "Imported batch {}: {} successful, {} errors",
330 batch_id,
331 success_count,
332 error_count
333 );
334 }
335
336 Ok(())
337 }
338
339 async fn download_finished_batches(&self) -> Result<()> {
340 let batch_ids: Vec<String> = {
341 let connection = self.connection.lock().unwrap();
342 let q = sql!(SELECT DISTINCT batch_id FROM openai_cache WHERE batch_id IS NOT NULL AND response IS NULL);
343 connection.select(q)?()?
344 };
345
346 for batch_id in &batch_ids {
347 let batch_status = batches::retrieve_batch(
348 self.http_client.as_ref(),
349 OPEN_AI_API_URL,
350 &self.api_key,
351 batch_id,
352 )
353 .await
354 .map_err(|e| anyhow::anyhow!("{:?}", e))?;
355
356 log::info!("Batch {} status: {}", batch_id, batch_status.status);
357
358 if batch_status.status == "completed" {
359 let output_file_id = match batch_status.output_file_id {
360 Some(id) => id,
361 None => {
362 log::warn!("Batch {} completed but has no output file", batch_id);
363 continue;
364 }
365 };
366
367 let results_content = batches::download_file(
368 self.http_client.as_ref(),
369 OPEN_AI_API_URL,
370 &self.api_key,
371 &output_file_id,
372 )
373 .await
374 .map_err(|e| anyhow::anyhow!("{:?}", e))?;
375
376 let results = batches::parse_batch_output(&results_content)
377 .map_err(|e| anyhow::anyhow!("Failed to parse batch output: {:?}", e))?;
378
379 let mut updates: Vec<(String, String)> = Vec::new();
380 let mut success_count = 0;
381
382 for result in results {
383 let request_hash = result
384 .custom_id
385 .strip_prefix("req_hash_")
386 .unwrap_or(&result.custom_id)
387 .to_string();
388
389 if let Some(response_body) = result.response {
390 if response_body.status_code == 200 {
391 let response_json = serde_json::to_string(&response_body.body)?;
392 updates.push((response_json, request_hash));
393 success_count += 1;
394 } else {
395 log::error!(
396 "Batch request {} failed with status {}",
397 request_hash,
398 response_body.status_code
399 );
400 let error_json = serde_json::json!({
401 "error": {
402 "type": "http_error",
403 "status_code": response_body.status_code
404 }
405 })
406 .to_string();
407 updates.push((error_json, request_hash));
408 }
409 } else if let Some(error) = result.error {
410 log::error!(
411 "Batch request {} failed: {}: {}",
412 request_hash,
413 error.code,
414 error.message
415 );
416 let error_json = serde_json::json!({
417 "error": {
418 "type": error.code,
419 "message": error.message
420 }
421 })
422 .to_string();
423 updates.push((error_json, request_hash));
424 }
425 }
426
427 let connection = self.connection.lock().unwrap();
428 connection.with_savepoint("batch_download", || {
429 let q = sql!(UPDATE openai_cache SET response = ? WHERE request_hash = ?);
430 let mut exec = connection.exec_bound::<(&str, &str)>(q)?;
431 for (response_json, request_hash) in &updates {
432 exec((response_json.as_str(), request_hash.as_str()))?;
433 }
434 Ok(())
435 })?;
436 log::info!("Downloaded {} successful requests", success_count);
437 }
438 }
439
440 Ok(())
441 }
442
443 async fn upload_pending_requests(&self) -> Result<Vec<String>> {
444 const BATCH_CHUNK_SIZE: i32 = 16_000;
445 let mut all_batch_ids = Vec::new();
446 let mut total_uploaded = 0;
447
448 loop {
449 let rows: Vec<(String, String)> = {
450 let connection = self.connection.lock().unwrap();
451 let q = sql!(
452 SELECT request_hash, request FROM openai_cache
453 WHERE batch_id IS NULL AND response IS NULL
454 LIMIT ?
455 );
456 connection.select_bound(q)?(BATCH_CHUNK_SIZE)?
457 };
458
459 if rows.is_empty() {
460 break;
461 }
462
463 let request_hashes: Vec<String> = rows.iter().map(|(hash, _)| hash.clone()).collect();
464
465 let mut jsonl_content = String::new();
466 for (hash, request_str) in &rows {
467 let serializable_request: SerializableRequest =
468 serde_json::from_str(request_str).unwrap();
469
470 let messages: Vec<RequestMessage> = serializable_request
471 .messages
472 .into_iter()
473 .map(|msg| match msg.role.as_str() {
474 "user" => RequestMessage::User {
475 content: MessageContent::Plain(msg.content),
476 },
477 "assistant" => RequestMessage::Assistant {
478 content: Some(MessageContent::Plain(msg.content)),
479 tool_calls: Vec::new(),
480 },
481 "system" => RequestMessage::System {
482 content: MessageContent::Plain(msg.content),
483 },
484 _ => RequestMessage::User {
485 content: MessageContent::Plain(msg.content),
486 },
487 })
488 .collect();
489
490 let request = OpenAiRequest {
491 model: serializable_request.model,
492 messages,
493 stream: false,
494 stream_options: None,
495 max_completion_tokens: Some(serializable_request.max_tokens),
496 stop: Vec::new(),
497 temperature: None,
498 tool_choice: None,
499 parallel_tool_calls: None,
500 tools: Vec::new(),
501 prompt_cache_key: None,
502 reasoning_effort: None,
503 };
504
505 let custom_id = format!("req_hash_{}", hash);
506 let batch_item = batches::BatchRequestItem::new(custom_id, request);
507 let line = batch_item
508 .to_jsonl_line()
509 .map_err(|e| anyhow::anyhow!("Failed to serialize batch item: {:?}", e))?;
510 jsonl_content.push_str(&line);
511 jsonl_content.push('\n');
512 }
513
514 let filename = format!("batch_{}.jsonl", chrono::Utc::now().timestamp());
515 let file_obj = batches::upload_batch_file(
516 self.http_client.as_ref(),
517 OPEN_AI_API_URL,
518 &self.api_key,
519 &filename,
520 jsonl_content.into_bytes(),
521 )
522 .await
523 .map_err(|e| anyhow::anyhow!("Failed to upload batch file: {:?}", e))?;
524
525 let batch = batches::create_batch(
526 self.http_client.as_ref(),
527 OPEN_AI_API_URL,
528 &self.api_key,
529 batches::CreateBatchRequest::new(file_obj.id),
530 )
531 .await
532 .map_err(|e| anyhow::anyhow!("Failed to create batch: {:?}", e))?;
533
534 {
535 let connection = self.connection.lock().unwrap();
536 connection.with_savepoint("batch_upload", || {
537 let q = sql!(UPDATE openai_cache SET batch_id = ? WHERE request_hash = ?);
538 let mut exec = connection.exec_bound::<(&str, &str)>(q)?;
539 for hash in &request_hashes {
540 exec((batch.id.as_str(), hash.as_str()))?;
541 }
542 Ok(())
543 })?;
544 }
545
546 let batch_len = rows.len();
547 total_uploaded += batch_len;
548 log::info!(
549 "Uploaded batch {} with {} requests ({} total)",
550 batch.id,
551 batch_len,
552 total_uploaded
553 );
554
555 all_batch_ids.push(batch.id);
556 }
557
558 if !all_batch_ids.is_empty() {
559 log::info!(
560 "Finished uploading {} batches with {} total requests",
561 all_batch_ids.len(),
562 total_uploaded
563 );
564 }
565
566 Ok(all_batch_ids)
567 }
568
569 fn request_hash(
570 model: &str,
571 max_tokens: u64,
572 messages: &[RequestMessage],
573 seed: Option<usize>,
574 ) -> String {
575 let mut hasher = std::hash::DefaultHasher::new();
576 "openai".hash(&mut hasher);
577 model.hash(&mut hasher);
578 max_tokens.hash(&mut hasher);
579 for msg in messages {
580 message_content_to_string(msg).hash(&mut hasher);
581 }
582 if let Some(seed) = seed {
583 seed.hash(&mut hasher);
584 }
585 let request_hash = hasher.finish();
586 format!("{request_hash:016x}")
587 }
588}
589
590fn message_role_to_string(msg: &RequestMessage) -> String {
591 match msg {
592 RequestMessage::User { .. } => "user".to_string(),
593 RequestMessage::Assistant { .. } => "assistant".to_string(),
594 RequestMessage::System { .. } => "system".to_string(),
595 RequestMessage::Tool { .. } => "tool".to_string(),
596 }
597}
598
599fn message_content_to_string(msg: &RequestMessage) -> String {
600 match msg {
601 RequestMessage::User { content } => content_to_string(content),
602 RequestMessage::Assistant { content, .. } => {
603 content.as_ref().map(content_to_string).unwrap_or_default()
604 }
605 RequestMessage::System { content } => content_to_string(content),
606 RequestMessage::Tool { content, .. } => content_to_string(content),
607 }
608}
609
610fn content_to_string(content: &MessageContent) -> String {
611 match content {
612 MessageContent::Plain(text) => text.clone(),
613 MessageContent::Multipart(parts) => parts
614 .iter()
615 .filter_map(|part| match part {
616 open_ai::MessagePart::Text { text } => Some(text.clone()),
617 _ => None,
618 })
619 .collect::<Vec<String>>()
620 .join("\n"),
621 }
622}
623
624pub enum OpenAiClient {
625 Plain(PlainOpenAiClient),
626 Batch(BatchingOpenAiClient),
627 #[allow(dead_code)]
628 Dummy,
629}
630
631impl OpenAiClient {
632 pub fn plain() -> Result<Self> {
633 Ok(Self::Plain(PlainOpenAiClient::new()?))
634 }
635
636 pub fn batch(cache_path: &Path) -> Result<Self> {
637 Ok(Self::Batch(BatchingOpenAiClient::new(cache_path)?))
638 }
639
640 #[allow(dead_code)]
641 pub fn dummy() -> Self {
642 Self::Dummy
643 }
644
645 pub async fn generate(
646 &self,
647 model: &str,
648 max_tokens: u64,
649 messages: Vec<RequestMessage>,
650 seed: Option<usize>,
651 cache_only: bool,
652 ) -> Result<Option<OpenAiResponse>> {
653 match self {
654 OpenAiClient::Plain(plain_client) => plain_client
655 .generate(model, max_tokens, messages)
656 .await
657 .map(Some),
658 OpenAiClient::Batch(batching_client) => {
659 batching_client
660 .generate(model, max_tokens, messages, seed, cache_only)
661 .await
662 }
663 OpenAiClient::Dummy => panic!("Dummy OpenAI client is not expected to be used"),
664 }
665 }
666
667 pub async fn sync_batches(&self) -> Result<()> {
668 match self {
669 OpenAiClient::Plain(_) => Ok(()),
670 OpenAiClient::Batch(batching_client) => batching_client.sync_batches().await,
671 OpenAiClient::Dummy => panic!("Dummy OpenAI client is not expected to be used"),
672 }
673 }
674
675 pub async fn import_batches(&self, batch_ids: &[String]) -> Result<()> {
676 match self {
677 OpenAiClient::Plain(_) => {
678 anyhow::bail!("Import batches is only supported with batching client")
679 }
680 OpenAiClient::Batch(batching_client) => batching_client.import_batches(batch_ids).await,
681 OpenAiClient::Dummy => panic!("Dummy OpenAI client is not expected to be used"),
682 }
683 }
684}