magicapi-ai-gateway 1.0.0

[DEPRECATED] This package has been renamed to 'noveum-ai-gateway'. Please use the new package for all future development. A high-performance AI Gateway proxy for routing requests to various AI providers, offering seamless integration and management of multiple AI providers.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406

use super::metrics::MetricsRegistry;
use super::provider_metrics::{get_metrics_extractor, ProviderMetrics, MetricsExtractor};
use super::RequestMetrics;
use axum::{
    body::{Body, Bytes},
    extract::State,
    http::{Request, Response},
    middleware::Next,
};
use futures_util::StreamExt;
use std::{sync::Arc, time::{Instant, Duration}};
use tokio::sync::mpsc;
use tokio_stream::wrappers::ReceiverStream;
use tracing::{debug, error};
use axum::body::to_bytes;
use hyper::Error;
use serde_json::Value;
use http;

// Constants for safeguards
const CHANNEL_SIZE: usize = 1000; // Increased buffer for streaming response
const MAX_ACCUMULATED_TEXT: usize = 5 * 1024 * 1024; // 5MB limit

pub async fn metrics_middleware(
    State(registry): State<Arc<MetricsRegistry>>,
    mut req: Request<Body>,
    next: Next,
) -> Response<Body> {
    let start = Instant::now();

    // Extract provider and other request info
    let provider = req
        .headers()
        .get("x-provider")
        .and_then(|h| h.to_str().ok())
        .unwrap_or("openai")
        .to_string();

    let path = req.uri().path().to_string();
    let method = req.method().to_string();

    // Extract project_id, org_id, and user_id from headers
    let project_id = req
        .headers()
        .get("x-project-id")
        .and_then(|h| h.to_str().ok())
        .map(|s| s.to_string());
        
    let org_id = req
        .headers()
        .get("x-organisation-id")
        .and_then(|h| h.to_str().ok())
        .map(|s| s.to_string());
        
    let user_id = req
        .headers()
        .get("x-user-id")
        .and_then(|h| h.to_str().ok())
        .map(|s| s.to_string());
        
    let experiment_id = req
        .headers()
        .get("x-experiment-id")
        .and_then(|h| h.to_str().ok())
        .map(|s| s.to_string());

    debug!("Received request: provider={}, path={}, method={}", provider, path, method);

    // Get metrics extractor for this provider
    let metrics_extractor = get_metrics_extractor(&provider);

    // Store original values before consuming request
    let original_method = req.method().clone();
    let original_uri = req.uri().clone();
    let original_headers = req.headers().clone();
    
    // Extract and store the original request body
    let (req_size, req_body, body) = {
        let bytes = to_bytes(req.into_body(), usize::MAX).await.unwrap_or_default();
        let size = bytes.len();
        let req_body = serde_json::from_slice(&bytes).ok();
        debug!("Request body size: {} bytes", size);
        (size, req_body, Body::from(bytes))
    };

    // Reconstruct request with original values
    let mut new_req = Request::builder()
        .method(original_method)
        .uri(original_uri)
        .body(body)
        .unwrap();
    *new_req.headers_mut() = original_headers;

    // Process the response with a timeout
    let response = tokio::time::timeout(
        Duration::from_secs(30),
        next.run(new_req)
    ).await.unwrap_or_else(|_| {
        debug!("Request timed out after 30 seconds");
        Response::builder()
            .status(http::StatusCode::GATEWAY_TIMEOUT)
            .body(Body::from("Request timed out after 30 seconds"))
            .unwrap()
    });

    let is_streaming = response
        .headers()
        .get("content-type")
        .and_then(|v| v.to_str().ok())
        .map(|v| v.contains("text/event-stream"))
        .unwrap_or(false);

    debug!("Response is streaming: {}", is_streaming);

    if is_streaming {
        handle_streaming_response(
            response,
            registry,
            provider,
            path,
            method,
            req_size,
            req_body,
            start,
            metrics_extractor,
            project_id,
            org_id,
            user_id,
            experiment_id,
        )
        .await
    } else {
        handle_regular_response(
            response,
            registry,
            provider,
            path,
            method,
            req_size,
            req_body,
            start,
            metrics_extractor,
            project_id,
            org_id,
            user_id,
            experiment_id,
        )
        .await
    }
}

async fn handle_regular_response(
    response: Response<Body>,
    registry: Arc<MetricsRegistry>,
    provider: String,
    path: String,
    method: String,
    req_size: usize,
    req_body: Option<Value>,
    start: Instant,
    metrics_extractor: Box<dyn MetricsExtractor>,
    project_id: Option<String>,
    org_id: Option<String>,
    user_id: Option<String>,
    experiment_id: Option<String>,
) -> Response<Body> {
    // Time to first byte is essentially the time taken to get the response headers
    let ttfb = start.elapsed();
    debug!("Time to first byte (TTFB): {:?}", ttfb);

    let (parts, body) = response.into_parts();
    let bytes = to_bytes(body, usize::MAX).await.unwrap_or_default();
    let resp_size = bytes.len();

    debug!("Regular response body size: {} bytes", resp_size);

    // Extract provider request ID from response headers
    let provider_request_id = parts.headers.get("x-request-id")
        .or_else(|| parts.headers.get("request-id"))  // Also check for Anthropic's request-id header
        .and_then(|v| v.to_str().ok())
        .map(String::from);
    
    if let Some(id) = &provider_request_id {
        debug!("Provider request ID: {}", id);
    }

    // Extract metrics from response body
    let (provider_metrics, resp_body) = if let Ok(json) = serde_json::from_slice::<serde_json::Value>(&bytes) {
        (metrics_extractor.extract_metrics(&json), Some(json))
    } else {
        (ProviderMetrics::default(), None)
    };

    debug!("Extracted provider metrics: {:?}", provider_metrics);

    let metrics = RequestMetrics {
        provider,
        path,
        method,
        model: provider_metrics.model,
        total_latency: start.elapsed(),
        provider_latency: provider_metrics.provider_latency,
        ttfb,  // Add the TTFB measurement
        request_size: req_size,
        response_size: resp_size,
        input_tokens: provider_metrics.input_tokens,
        output_tokens: provider_metrics.output_tokens,
        total_tokens: provider_metrics.total_tokens,
        status_code: parts.status.as_u16(),
        cost: provider_metrics.cost,
        project_id,
        org_id,
        user_id,
        experiment_id,
        provider_request_id,
        request_body: req_body,
        response_body: resp_body,
        ..Default::default()
    };

    registry.record_metrics(metrics).await;

    Response::from_parts(parts, Body::from(bytes))
}

async fn handle_streaming_response(
    response: Response<Body>,
    registry: Arc<MetricsRegistry>,
    provider: String,
    path: String,
    method: String,
    req_size: usize,
    req_body: Option<Value>,
    start: Instant,
    metrics_extractor: Box<dyn MetricsExtractor>,
    project_id: Option<String>,
    org_id: Option<String>,
    user_id: Option<String>,
    experiment_id: Option<String>,
) -> Response<Body> {
    // Time to first byte is essentially the time taken to get the response headers
    let ttfb = start.elapsed();
    debug!("Time to first byte for streaming response (TTFB): {:?}", ttfb);

    let (parts, body) = response.into_parts();
    let (tx, rx) = mpsc::channel::<Result<Bytes, Error>>(CHANNEL_SIZE);

    // Extract provider request ID from response headers
    let provider_request_id = parts.headers.get("x-request-id")
        .or_else(|| parts.headers.get("request-id"))  // Also check for Anthropic's request-id header
        .and_then(|v| v.to_str().ok())
        .map(String::from);
    
    if let Some(id) = &provider_request_id {
        debug!("Provider request ID for streaming response: {}", id);
    }

    let metrics_registry = registry.clone();
    let mut accumulated_text = String::with_capacity(MAX_ACCUMULATED_TEXT);

    // Process the stream
    tokio::spawn(async move {
        let mut response_size = 0;
        let mut accumulated_metrics = ProviderMetrics::default();
        let mut final_metrics_found = false;
        let mut resp_body = None;
        let mut streamed_chunks = Vec::new();

        let mut stream = body.into_data_stream();
        while let Some(chunk) = stream.next().await {
            if let Ok(bytes) = chunk {
                response_size += bytes.len();
                debug!("Streaming response chunk size: {} bytes", bytes.len());

                if let Ok(chunk_str) = String::from_utf8(bytes.to_vec()) {
                    if accumulated_text.len() + chunk_str.len() > MAX_ACCUMULATED_TEXT {
                        error!("Accumulated text exceeded maximum size of {} bytes", MAX_ACCUMULATED_TEXT);
                        break;
                    }
                    accumulated_text.push_str(&chunk_str);
                    
                    // Try to parse the chunk as JSON and store it
                    if let Ok(json_chunk) = serde_json::from_str::<Value>(&chunk_str) {
                        // Only store non-empty chunks
                        if !json_chunk.is_null() && !json_chunk.as_object().map_or(true, |o| o.is_empty()) {
                            streamed_chunks.push(json_chunk);
                        }
                    } else {
                        // For streaming that sends chunks broken up, try to parse 
                        // different formats (like data: {...}\n\n for SSE)
                        for line in chunk_str.lines() {
                            if line.starts_with("data: ") {
                                let data = line.trim_start_matches("data: ");
                                if data == "[DONE]" {
                                    debug!("Received [DONE] signal in streaming");
                                    continue;
                                }
                                
                                if let Ok(json_data) = serde_json::from_str::<Value>(data) {
                                    streamed_chunks.push(json_data.clone());
                                    
                                    // Try to extract metrics from this chunk
                                    if let Some(chunk_metrics) = metrics_extractor.extract_streaming_metrics(data) {
                                        debug!("Found metrics in streaming chunk: {:?}", chunk_metrics);
                                        accumulated_metrics = chunk_metrics;
                                        final_metrics_found = true;
                                    }
                                }
                            }
                        }
                    }
                }
                
                // Always forward the bytes to the client
                if let Err(e) = tx.send(Ok(bytes)).await {
                    error!("Failed to forward streaming chunk: {}", e);
                    break;
                }
            } else if let Err(e) = chunk {
                error!("Error in streaming response: {}", e);
                // For type compatibility, we'll just break the stream instead of trying to send the error
                // This avoids issues with error type conversions
                break;
            }
        }

        // Try to parse the accumulated response
        if !accumulated_text.is_empty() {
            resp_body = serde_json::from_str(&accumulated_text).ok();
        }
        
        // Track if this is a provider that requires special streaming handling
        let is_openai_streaming = provider == "openai";
        let is_groq_streaming = provider == "groq";
        let needs_special_streaming_handling = is_openai_streaming || is_groq_streaming;

        // For providers that don't always include token data in streaming responses,
        // create a minimal metrics record with what we know
        if !final_metrics_found && needs_special_streaming_handling && !streamed_chunks.is_empty() {
            // Try to extract model from the stream chunks
            let model = streamed_chunks.iter()
                .find_map(|chunk| chunk.get("model").and_then(|m| m.as_str()))
                .unwrap_or(if is_groq_streaming { "llama" } else { "unknown" })
                .to_string();
                
            debug!("Creating partial metrics for {} streaming response with model: {}", 
                   provider, model);
            
            // Rough token estimation based on accumulated text length
            // Approximation: ~4 characters per token for English text
            let estimated_output_tokens = if !accumulated_text.is_empty() {
                Some((accumulated_text.len() as f64 / 4.0).ceil() as u32)
            } else {
                None
            };
            
            debug!("Estimated output tokens from {} bytes of text: {:?}", 
                accumulated_text.len(), estimated_output_tokens);
            
            accumulated_metrics = ProviderMetrics {
                model,
                provider_latency: Duration::from_millis(0),
                output_tokens: estimated_output_tokens,
                // We don't have input tokens or total tokens
                ..Default::default()
            };
            
            final_metrics_found = true;
        }

        // Record final metrics if we found them
        if final_metrics_found {
            let metrics = RequestMetrics {
                provider,
                path,
                method,
                model: accumulated_metrics.model,
                total_latency: start.elapsed(),
                provider_latency: accumulated_metrics.provider_latency,
                ttfb,  // Add the TTFB measurement
                request_size: req_size,
                response_size,
                input_tokens: accumulated_metrics.input_tokens,
                output_tokens: accumulated_metrics.output_tokens,
                total_tokens: accumulated_metrics.total_tokens,
                status_code: parts.status.as_u16(),
                cost: accumulated_metrics.cost,
                project_id,
                org_id,
                user_id,
                experiment_id,
                provider_request_id,
                request_body: req_body,
                response_body: resp_body,
                streamed_data: if !streamed_chunks.is_empty() { Some(streamed_chunks) } else { None },
                is_streaming: true,
                ..Default::default()
            };
            metrics_registry.record_metrics(metrics).await;
        } else {
            debug!("No final metrics found in streaming response. Total text accumulated: {} bytes", accumulated_text.len());
        }
    });

    Response::from_parts(parts, Body::from_stream(ReceiverStream::new(rx)))
}