dakera_client/
client.rs

1//! Dakera client implementation
2
3use reqwest::{Client, StatusCode};
4use std::sync::{Arc, Mutex};
5use std::time::Duration;
6use tracing::{debug, instrument};
7
8use serde::Deserialize;
9
10use crate::error::{ClientError, Result, ServerErrorCode};
11use crate::types::*;
12
13/// Default timeout for requests
14const DEFAULT_TIMEOUT_SECS: u64 = 30;
15
16/// Dakera client for interacting with the vector database
17#[derive(Debug, Clone)]
18pub struct DakeraClient {
19    /// HTTP client
20    pub(crate) client: Client,
21    /// Base URL of the Dakera server
22    pub(crate) base_url: String,
23    /// Retry configuration (wired into API call sites in a follow-up; suppressed until then)
24    #[allow(dead_code)]
25    pub(crate) retry_config: RetryConfig,
26    /// OPS-1: last seen rate-limit headers (shared across clones)
27    pub(crate) last_rate_limit: Arc<Mutex<Option<RateLimitHeaders>>>,
28}
29
30impl DakeraClient {
31    /// Create a new client with the given base URL
32    ///
33    /// # Example
34    ///
35    /// ```rust,no_run
36    /// use dakera_client::DakeraClient;
37    ///
38    /// let client = DakeraClient::new("http://localhost:3000").unwrap();
39    /// ```
40    pub fn new(base_url: impl Into<String>) -> Result<Self> {
41        DakeraClientBuilder::new(base_url).build()
42    }
43
44    /// Create a new client builder for more configuration options
45    pub fn builder(base_url: impl Into<String>) -> DakeraClientBuilder {
46        DakeraClientBuilder::new(base_url)
47    }
48
49    // ========================================================================
50    // Health & Status
51    // ========================================================================
52
53    /// Check server health
54    #[instrument(skip(self))]
55    pub async fn health(&self) -> Result<HealthResponse> {
56        let url = format!("{}/health", self.base_url);
57        let response = self.client.get(&url).send().await?;
58
59        if response.status().is_success() {
60            Ok(response.json().await?)
61        } else {
62            // Health endpoint might return simple OK
63            Ok(HealthResponse {
64                healthy: true,
65                version: None,
66                uptime_seconds: None,
67            })
68        }
69    }
70
71    /// Check if server is ready
72    #[instrument(skip(self))]
73    pub async fn ready(&self) -> Result<ReadinessResponse> {
74        let url = format!("{}/health/ready", self.base_url);
75        let response = self.client.get(&url).send().await?;
76
77        if response.status().is_success() {
78            Ok(response.json().await?)
79        } else {
80            Ok(ReadinessResponse {
81                ready: false,
82                components: None,
83            })
84        }
85    }
86
87    /// Check if server is live
88    #[instrument(skip(self))]
89    pub async fn live(&self) -> Result<bool> {
90        let url = format!("{}/health/live", self.base_url);
91        let response = self.client.get(&url).send().await?;
92        Ok(response.status().is_success())
93    }
94
95    // ========================================================================
96    // Namespace Operations
97    // ========================================================================
98
99    /// List all namespaces
100    #[instrument(skip(self))]
101    pub async fn list_namespaces(&self) -> Result<Vec<String>> {
102        let url = format!("{}/v1/namespaces", self.base_url);
103        let response = self.client.get(&url).send().await?;
104        self.handle_response::<ListNamespacesResponse>(response)
105            .await
106            .map(|r| r.namespaces)
107    }
108
109    /// Get namespace information
110    #[instrument(skip(self))]
111    pub async fn get_namespace(&self, namespace: &str) -> Result<NamespaceInfo> {
112        let url = format!("{}/v1/namespaces/{}", self.base_url, namespace);
113        let response = self.client.get(&url).send().await?;
114        self.handle_response(response).await
115    }
116
117    /// Create a new namespace
118    #[instrument(skip(self, request))]
119    pub async fn create_namespace(
120        &self,
121        namespace: &str,
122        request: CreateNamespaceRequest,
123    ) -> Result<NamespaceInfo> {
124        let url = format!("{}/v1/namespaces/{}", self.base_url, namespace);
125        let response = self.client.post(&url).json(&request).send().await?;
126        self.handle_response(response).await
127    }
128
129    /// Create or update a namespace configuration (upsert semantics — v0.6.0).
130    ///
131    /// Creates the namespace if it does not exist, or updates its distance-metric
132    /// configuration if it already exists.  Dimension changes are rejected to
133    /// prevent silent data corruption.  Requires `Scope::Write`.
134    #[instrument(skip(self, request), fields(namespace = %namespace))]
135    pub async fn configure_namespace(
136        &self,
137        namespace: &str,
138        request: ConfigureNamespaceRequest,
139    ) -> Result<ConfigureNamespaceResponse> {
140        let url = format!("{}/v1/namespaces/{}", self.base_url, namespace);
141        let response = self.client.put(&url).json(&request).send().await?;
142        self.handle_response(response).await
143    }
144
145    // ========================================================================
146    // Vector Operations
147    // ========================================================================
148
149    /// Upsert vectors into a namespace
150    #[instrument(skip(self, request), fields(vector_count = request.vectors.len()))]
151    pub async fn upsert(&self, namespace: &str, request: UpsertRequest) -> Result<UpsertResponse> {
152        let url = format!("{}/v1/namespaces/{}/vectors", self.base_url, namespace);
153        debug!(
154            "Upserting {} vectors to {}",
155            request.vectors.len(),
156            namespace
157        );
158
159        let response = self.client.post(&url).json(&request).send().await?;
160        self.handle_response(response).await
161    }
162
163    /// Upsert a single vector (convenience method)
164    #[instrument(skip(self, vector))]
165    pub async fn upsert_one(&self, namespace: &str, vector: Vector) -> Result<UpsertResponse> {
166        self.upsert(namespace, UpsertRequest::single(vector)).await
167    }
168
169    /// Upsert vectors in column format (Turbopuffer-inspired)
170    ///
171    /// This format is more efficient for bulk upserts as it avoids repeating
172    /// field names for each vector. All arrays must have equal length.
173    ///
174    /// # Example
175    ///
176    /// ```rust,no_run
177    /// use dakera_client::{DakeraClient, ColumnUpsertRequest};
178    ///
179    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
180    /// let client = DakeraClient::new("http://localhost:3000")?;
181    ///
182    /// let request = ColumnUpsertRequest::new(
183    ///     vec!["id1".to_string(), "id2".to_string(), "id3".to_string()],
184    ///     vec![
185    ///         vec![0.1, 0.2, 0.3],
186    ///         vec![0.4, 0.5, 0.6],
187    ///         vec![0.7, 0.8, 0.9],
188    ///     ],
189    /// )
190    /// .with_attribute("category", vec![
191    ///     serde_json::json!("A"),
192    ///     serde_json::json!("B"),
193    ///     serde_json::json!("A"),
194    /// ]);
195    ///
196    /// let response = client.upsert_columns("my-namespace", request).await?;
197    /// println!("Upserted {} vectors", response.upserted_count);
198    /// # Ok(())
199    /// # }
200    /// ```
201    #[instrument(skip(self, request), fields(namespace = %namespace, count = request.ids.len()))]
202    pub async fn upsert_columns(
203        &self,
204        namespace: &str,
205        request: ColumnUpsertRequest,
206    ) -> Result<UpsertResponse> {
207        let url = format!(
208            "{}/v1/namespaces/{}/upsert-columns",
209            self.base_url, namespace
210        );
211        debug!(
212            "Upserting {} vectors in column format to {}",
213            request.ids.len(),
214            namespace
215        );
216
217        let response = self.client.post(&url).json(&request).send().await?;
218        self.handle_response(response).await
219    }
220
221    /// Query for similar vectors
222    #[instrument(skip(self, request), fields(top_k = request.top_k))]
223    pub async fn query(&self, namespace: &str, request: QueryRequest) -> Result<QueryResponse> {
224        let url = format!("{}/v1/namespaces/{}/query", self.base_url, namespace);
225        debug!(
226            "Querying namespace {} for top {} results",
227            namespace, request.top_k
228        );
229
230        let response = self.client.post(&url).json(&request).send().await?;
231        self.handle_response(response).await
232    }
233
234    /// Simple query with just a vector and top_k (convenience method)
235    #[instrument(skip(self, vector))]
236    pub async fn query_simple(
237        &self,
238        namespace: &str,
239        vector: Vec<f32>,
240        top_k: u32,
241    ) -> Result<QueryResponse> {
242        self.query(namespace, QueryRequest::new(vector, top_k))
243            .await
244    }
245
246    /// Execute multiple queries in a single request
247    ///
248    /// This allows executing multiple vector similarity queries in parallel,
249    /// which is more efficient than making separate requests.
250    ///
251    /// # Example
252    ///
253    /// ```rust,no_run
254    /// use dakera_client::{DakeraClient, BatchQueryRequest, BatchQueryItem};
255    ///
256    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
257    /// let client = DakeraClient::new("http://localhost:3000")?;
258    ///
259    /// let request = BatchQueryRequest::new(vec![
260    ///     BatchQueryItem::new(vec![0.1, 0.2, 0.3], 5).with_id("query1"),
261    ///     BatchQueryItem::new(vec![0.4, 0.5, 0.6], 10).with_id("query2"),
262    /// ]);
263    ///
264    /// let response = client.batch_query("my-namespace", request).await?;
265    /// println!("Executed {} queries in {}ms", response.query_count, response.total_latency_ms);
266    /// # Ok(())
267    /// # }
268    /// ```
269    #[instrument(skip(self, request), fields(namespace = %namespace, query_count = request.queries.len()))]
270    pub async fn batch_query(
271        &self,
272        namespace: &str,
273        request: BatchQueryRequest,
274    ) -> Result<BatchQueryResponse> {
275        let url = format!("{}/v1/namespaces/{}/batch-query", self.base_url, namespace);
276        debug!(
277            "Batch querying namespace {} with {} queries",
278            namespace,
279            request.queries.len()
280        );
281
282        let response = self.client.post(&url).json(&request).send().await?;
283        self.handle_response(response).await
284    }
285
286    /// Delete vectors by ID
287    #[instrument(skip(self, request), fields(id_count = request.ids.len()))]
288    pub async fn delete(&self, namespace: &str, request: DeleteRequest) -> Result<DeleteResponse> {
289        let url = format!(
290            "{}/v1/namespaces/{}/vectors/delete",
291            self.base_url, namespace
292        );
293        debug!("Deleting {} vectors from {}", request.ids.len(), namespace);
294
295        let response = self.client.post(&url).json(&request).send().await?;
296        self.handle_response(response).await
297    }
298
299    /// Delete a single vector by ID (convenience method)
300    #[instrument(skip(self))]
301    pub async fn delete_one(&self, namespace: &str, id: &str) -> Result<DeleteResponse> {
302        self.delete(namespace, DeleteRequest::single(id)).await
303    }
304
305    // ========================================================================
306    // Full-Text Search Operations
307    // ========================================================================
308
309    /// Index documents for full-text search
310    #[instrument(skip(self, request), fields(doc_count = request.documents.len()))]
311    pub async fn index_documents(
312        &self,
313        namespace: &str,
314        request: IndexDocumentsRequest,
315    ) -> Result<IndexDocumentsResponse> {
316        let url = format!(
317            "{}/v1/namespaces/{}/fulltext/index",
318            self.base_url, namespace
319        );
320        debug!(
321            "Indexing {} documents in {}",
322            request.documents.len(),
323            namespace
324        );
325
326        let response = self.client.post(&url).json(&request).send().await?;
327        self.handle_response(response).await
328    }
329
330    /// Index a single document (convenience method)
331    #[instrument(skip(self, document))]
332    pub async fn index_document(
333        &self,
334        namespace: &str,
335        document: Document,
336    ) -> Result<IndexDocumentsResponse> {
337        self.index_documents(
338            namespace,
339            IndexDocumentsRequest {
340                documents: vec![document],
341            },
342        )
343        .await
344    }
345
346    /// Perform full-text search
347    #[instrument(skip(self, request))]
348    pub async fn fulltext_search(
349        &self,
350        namespace: &str,
351        request: FullTextSearchRequest,
352    ) -> Result<FullTextSearchResponse> {
353        let url = format!(
354            "{}/v1/namespaces/{}/fulltext/search",
355            self.base_url, namespace
356        );
357        debug!("Full-text search in {} for: {}", namespace, request.query);
358
359        let response = self.client.post(&url).json(&request).send().await?;
360        self.handle_response(response).await
361    }
362
363    /// Simple full-text search (convenience method)
364    #[instrument(skip(self))]
365    pub async fn search_text(
366        &self,
367        namespace: &str,
368        query: &str,
369        top_k: u32,
370    ) -> Result<FullTextSearchResponse> {
371        self.fulltext_search(namespace, FullTextSearchRequest::new(query, top_k))
372            .await
373    }
374
375    /// Get full-text index statistics
376    #[instrument(skip(self))]
377    pub async fn fulltext_stats(&self, namespace: &str) -> Result<FullTextStats> {
378        let url = format!(
379            "{}/v1/namespaces/{}/fulltext/stats",
380            self.base_url, namespace
381        );
382        let response = self.client.get(&url).send().await?;
383        self.handle_response(response).await
384    }
385
386    /// Delete documents from full-text index
387    #[instrument(skip(self, request))]
388    pub async fn fulltext_delete(
389        &self,
390        namespace: &str,
391        request: DeleteRequest,
392    ) -> Result<DeleteResponse> {
393        let url = format!(
394            "{}/v1/namespaces/{}/fulltext/delete",
395            self.base_url, namespace
396        );
397        let response = self.client.post(&url).json(&request).send().await?;
398        self.handle_response(response).await
399    }
400
401    // ========================================================================
402    // Hybrid Search Operations
403    // ========================================================================
404
405    /// Perform hybrid search (vector + full-text)
406    #[instrument(skip(self, request), fields(top_k = request.top_k))]
407    pub async fn hybrid_search(
408        &self,
409        namespace: &str,
410        request: HybridSearchRequest,
411    ) -> Result<HybridSearchResponse> {
412        let url = format!("{}/v1/namespaces/{}/hybrid", self.base_url, namespace);
413        debug!(
414            "Hybrid search in {} with vector_weight={}",
415            namespace, request.vector_weight
416        );
417
418        let response = self.client.post(&url).json(&request).send().await?;
419        self.handle_response(response).await
420    }
421
422    // ========================================================================
423    // Multi-Vector Search Operations
424    // ========================================================================
425
426    /// Multi-vector search with positive/negative vectors and MMR
427    ///
428    /// This performs semantic search using multiple positive vectors (to search towards)
429    /// and optional negative vectors (to search away from). Supports MMR (Maximal Marginal
430    /// Relevance) for result diversity.
431    ///
432    /// # Example
433    ///
434    /// ```rust,no_run
435    /// use dakera_client::{DakeraClient, MultiVectorSearchRequest};
436    ///
437    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
438    /// let client = DakeraClient::new("http://localhost:3000")?;
439    ///
440    /// // Search towards multiple concepts, away from others
441    /// let request = MultiVectorSearchRequest::new(vec![
442    ///     vec![0.1, 0.2, 0.3],  // positive vector 1
443    ///     vec![0.4, 0.5, 0.6],  // positive vector 2
444    /// ])
445    /// .with_negative_vectors(vec![
446    ///     vec![0.7, 0.8, 0.9],  // negative vector
447    /// ])
448    /// .with_top_k(10)
449    /// .with_mmr(0.7);  // Enable MMR with lambda=0.7
450    ///
451    /// let response = client.multi_vector_search("my-namespace", request).await?;
452    /// for result in response.results {
453    ///     println!("ID: {}, Score: {}", result.id, result.score);
454    /// }
455    /// # Ok(())
456    /// # }
457    /// ```
458    #[instrument(skip(self, request), fields(namespace = %namespace))]
459    pub async fn multi_vector_search(
460        &self,
461        namespace: &str,
462        request: MultiVectorSearchRequest,
463    ) -> Result<MultiVectorSearchResponse> {
464        let url = format!("{}/v1/namespaces/{}/multi-vector", self.base_url, namespace);
465        debug!(
466            "Multi-vector search in {} with {} positive vectors",
467            namespace,
468            request.positive_vectors.len()
469        );
470
471        let response = self.client.post(&url).json(&request).send().await?;
472        self.handle_response(response).await
473    }
474
475    // ========================================================================
476    // Aggregation Operations
477    // ========================================================================
478
479    /// Aggregate vectors with grouping (Turbopuffer-inspired)
480    ///
481    /// This performs aggregation queries on vector metadata, supporting
482    /// count, sum, avg, min, and max operations with optional grouping.
483    ///
484    /// # Example
485    ///
486    /// ```rust,no_run
487    /// use dakera_client::{DakeraClient, AggregationRequest};
488    ///
489    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
490    /// let client = DakeraClient::new("http://localhost:3000")?;
491    ///
492    /// // Count all vectors and sum scores, grouped by category
493    /// let request = AggregationRequest::new()
494    ///     .with_count("total_count")
495    ///     .with_sum("total_score", "score")
496    ///     .with_avg("avg_score", "score")
497    ///     .with_group_by("category");
498    ///
499    /// let response = client.aggregate("my-namespace", request).await?;
500    /// if let Some(groups) = response.aggregation_groups {
501    ///     for group in groups {
502    ///         println!("Group: {:?}", group.group_key);
503    ///     }
504    /// }
505    /// # Ok(())
506    /// # }
507    /// ```
508    #[instrument(skip(self, request), fields(namespace = %namespace))]
509    pub async fn aggregate(
510        &self,
511        namespace: &str,
512        request: AggregationRequest,
513    ) -> Result<AggregationResponse> {
514        let url = format!("{}/v1/namespaces/{}/aggregate", self.base_url, namespace);
515        debug!(
516            "Aggregating in namespace {} with {} aggregations",
517            namespace,
518            request.aggregate_by.len()
519        );
520
521        let response = self.client.post(&url).json(&request).send().await?;
522        self.handle_response(response).await
523    }
524
525    // ========================================================================
526    // Unified Query Operations
527    // ========================================================================
528
529    /// Unified query with flexible ranking options (Turbopuffer-inspired)
530    ///
531    /// This provides a unified API for vector search (ANN/kNN), full-text search (BM25),
532    /// and attribute ordering. Supports combining multiple ranking functions with
533    /// Sum, Max, and Product operators.
534    ///
535    /// # Example
536    ///
537    /// ```rust,no_run
538    /// use dakera_client::{DakeraClient, UnifiedQueryRequest, SortDirection};
539    ///
540    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
541    /// let client = DakeraClient::new("http://localhost:3000")?;
542    ///
543    /// // Vector ANN search
544    /// let request = UnifiedQueryRequest::vector_search(vec![0.1, 0.2, 0.3], 10);
545    /// let response = client.unified_query("my-namespace", request).await?;
546    ///
547    /// // Full-text BM25 search
548    /// let request = UnifiedQueryRequest::fulltext_search("content", "hello world", 10);
549    /// let response = client.unified_query("my-namespace", request).await?;
550    ///
551    /// // Attribute ordering with filter
552    /// let request = UnifiedQueryRequest::attribute_order("timestamp", SortDirection::Desc, 10)
553    ///     .with_filter(serde_json::json!({"category": {"$eq": "science"}}));
554    /// let response = client.unified_query("my-namespace", request).await?;
555    ///
556    /// for result in response.results {
557    ///     println!("ID: {}, Score: {:?}", result.id, result.dist);
558    /// }
559    /// # Ok(())
560    /// # }
561    /// ```
562    #[instrument(skip(self, request), fields(namespace = %namespace))]
563    pub async fn unified_query(
564        &self,
565        namespace: &str,
566        request: UnifiedQueryRequest,
567    ) -> Result<UnifiedQueryResponse> {
568        let url = format!(
569            "{}/v1/namespaces/{}/unified-query",
570            self.base_url, namespace
571        );
572        debug!(
573            "Unified query in namespace {} with top_k={}",
574            namespace, request.top_k
575        );
576
577        let response = self.client.post(&url).json(&request).send().await?;
578        self.handle_response(response).await
579    }
580
581    /// Simple vector search using the unified query API (convenience method)
582    ///
583    /// This is a shortcut for `unified_query` with a vector ANN search.
584    #[instrument(skip(self, vector))]
585    pub async fn unified_vector_search(
586        &self,
587        namespace: &str,
588        vector: Vec<f32>,
589        top_k: usize,
590    ) -> Result<UnifiedQueryResponse> {
591        self.unified_query(namespace, UnifiedQueryRequest::vector_search(vector, top_k))
592            .await
593    }
594
595    /// Simple full-text search using the unified query API (convenience method)
596    ///
597    /// This is a shortcut for `unified_query` with a BM25 full-text search.
598    #[instrument(skip(self))]
599    pub async fn unified_text_search(
600        &self,
601        namespace: &str,
602        field: &str,
603        query: &str,
604        top_k: usize,
605    ) -> Result<UnifiedQueryResponse> {
606        self.unified_query(
607            namespace,
608            UnifiedQueryRequest::fulltext_search(field, query, top_k),
609        )
610        .await
611    }
612
613    // ========================================================================
614    // Query Explain Operations
615    // ========================================================================
616
617    /// Explain query execution plan (similar to SQL EXPLAIN)
618    ///
619    /// This provides detailed information about how a query will be executed,
620    /// including index selection, execution stages, cost estimates, and
621    /// performance recommendations.
622    ///
623    /// # Example
624    ///
625    /// ```rust,no_run
626    /// use dakera_client::{DakeraClient, QueryExplainRequest};
627    ///
628    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
629    /// let client = DakeraClient::new("http://localhost:3000")?;
630    ///
631    /// // Explain a vector search query
632    /// let request = QueryExplainRequest::vector_search(vec![0.1, 0.2, 0.3], 10)
633    ///     .with_verbose();
634    /// let plan = client.explain_query("my-namespace", request).await?;
635    ///
636    /// println!("Query plan: {}", plan.summary);
637    /// println!("Estimated time: {}ms", plan.cost_estimate.estimated_time_ms);
638    ///
639    /// for stage in &plan.stages {
640    ///     println!("Stage {}: {} - {}", stage.order, stage.name, stage.description);
641    /// }
642    ///
643    /// for rec in &plan.recommendations {
644    ///     println!("Recommendation ({}): {}", rec.priority, rec.description);
645    /// }
646    /// # Ok(())
647    /// # }
648    /// ```
649    #[instrument(skip(self, request), fields(namespace = %namespace))]
650    pub async fn explain_query(
651        &self,
652        namespace: &str,
653        request: QueryExplainRequest,
654    ) -> Result<QueryExplainResponse> {
655        let url = format!("{}/v1/namespaces/{}/explain", self.base_url, namespace);
656        debug!(
657            "Explaining query in namespace {} (query_type={:?}, top_k={})",
658            namespace, request.query_type, request.top_k
659        );
660
661        let response = self.client.post(&url).json(&request).send().await?;
662        self.handle_response(response).await
663    }
664
665    // ========================================================================
666    // Cache Warming Operations
667    // ========================================================================
668
669    /// Warm cache for vectors in a namespace
670    ///
671    /// This pre-loads vectors into cache tiers for faster subsequent access.
672    /// Supports priority levels and can run in the background.
673    ///
674    /// # Example
675    ///
676    /// ```rust,no_run
677    /// use dakera_client::{DakeraClient, WarmCacheRequest, WarmingPriority};
678    ///
679    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
680    /// let client = DakeraClient::new("http://localhost:3000")?;
681    ///
682    /// // Warm entire namespace with high priority
683    /// let response = client.warm_cache(
684    ///     WarmCacheRequest::new("my-namespace")
685    ///         .with_priority(WarmingPriority::High)
686    /// ).await?;
687    ///
688    /// println!("Warmed {} entries", response.entries_warmed);
689    /// # Ok(())
690    /// # }
691    /// ```
692    #[instrument(skip(self, request), fields(namespace = %request.namespace, priority = ?request.priority))]
693    pub async fn warm_cache(&self, request: WarmCacheRequest) -> Result<WarmCacheResponse> {
694        let url = format!(
695            "{}/v1/namespaces/{}/cache/warm",
696            self.base_url, request.namespace
697        );
698        debug!(
699            "Warming cache for namespace {} with priority {:?}",
700            request.namespace, request.priority
701        );
702
703        let response = self.client.post(&url).json(&request).send().await?;
704        self.handle_response(response).await
705    }
706
707    /// Warm specific vectors by ID (convenience method)
708    #[instrument(skip(self, vector_ids))]
709    pub async fn warm_vectors(
710        &self,
711        namespace: &str,
712        vector_ids: Vec<String>,
713    ) -> Result<WarmCacheResponse> {
714        self.warm_cache(WarmCacheRequest::new(namespace).with_vector_ids(vector_ids))
715            .await
716    }
717
718    // ========================================================================
719    // Export Operations
720    // ========================================================================
721
722    /// Export vectors from a namespace with pagination
723    ///
724    /// This exports all vectors from a namespace, supporting pagination for
725    /// large datasets. Use the `next_cursor` from the response to fetch
726    /// subsequent pages.
727    ///
728    /// # Example
729    ///
730    /// ```rust,no_run
731    /// use dakera_client::{DakeraClient, ExportRequest};
732    ///
733    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
734    /// let client = DakeraClient::new("http://localhost:3000")?;
735    ///
736    /// // Export first page of vectors
737    /// let mut request = ExportRequest::new().with_top_k(1000);
738    /// let response = client.export("my-namespace", request).await?;
739    ///
740    /// println!("Exported {} vectors", response.returned_count);
741    ///
742    /// // Fetch next page if available
743    /// if let Some(cursor) = response.next_cursor {
744    ///     let next_request = ExportRequest::new().with_cursor(cursor);
745    ///     let next_response = client.export("my-namespace", next_request).await?;
746    /// }
747    /// # Ok(())
748    /// # }
749    /// ```
750    #[instrument(skip(self, request), fields(namespace = %namespace))]
751    pub async fn export(&self, namespace: &str, request: ExportRequest) -> Result<ExportResponse> {
752        let url = format!("{}/v1/namespaces/{}/export", self.base_url, namespace);
753        debug!(
754            "Exporting vectors from namespace {} (top_k={}, cursor={:?})",
755            namespace, request.top_k, request.cursor
756        );
757
758        let response = self.client.post(&url).json(&request).send().await?;
759        self.handle_response(response).await
760    }
761
762    /// Export all vectors from a namespace (convenience method)
763    ///
764    /// This is a simple wrapper that exports with default settings.
765    #[instrument(skip(self))]
766    pub async fn export_all(&self, namespace: &str) -> Result<ExportResponse> {
767        self.export(namespace, ExportRequest::new()).await
768    }
769
770    // ========================================================================
771    // Operations
772    // ========================================================================
773
774    /// Get system diagnostics
775    #[instrument(skip(self))]
776    pub async fn diagnostics(&self) -> Result<SystemDiagnostics> {
777        let url = format!("{}/ops/diagnostics", self.base_url);
778        let response = self.client.get(&url).send().await?;
779        self.handle_response(response).await
780    }
781
782    /// List background jobs
783    #[instrument(skip(self))]
784    pub async fn list_jobs(&self) -> Result<Vec<JobInfo>> {
785        let url = format!("{}/ops/jobs", self.base_url);
786        let response = self.client.get(&url).send().await?;
787        self.handle_response(response).await
788    }
789
790    /// Get a specific job status
791    #[instrument(skip(self))]
792    pub async fn get_job(&self, job_id: &str) -> Result<Option<JobInfo>> {
793        let url = format!("{}/ops/jobs/{}", self.base_url, job_id);
794        let response = self.client.get(&url).send().await?;
795
796        if response.status() == StatusCode::NOT_FOUND {
797            return Ok(None);
798        }
799
800        self.handle_response(response).await.map(Some)
801    }
802
803    /// Trigger index compaction
804    #[instrument(skip(self, request))]
805    pub async fn compact(&self, request: CompactionRequest) -> Result<CompactionResponse> {
806        let url = format!("{}/ops/compact", self.base_url);
807        let response = self.client.post(&url).json(&request).send().await?;
808        self.handle_response(response).await
809    }
810
811    /// Request graceful shutdown
812    #[instrument(skip(self))]
813    pub async fn shutdown(&self) -> Result<()> {
814        let url = format!("{}/ops/shutdown", self.base_url);
815        let response = self.client.post(&url).send().await?;
816
817        if response.status().is_success() {
818            Ok(())
819        } else {
820            let status = response.status().as_u16();
821            let text = response.text().await.unwrap_or_default();
822            Err(ClientError::Server {
823                status,
824                message: text,
825                code: None,
826            })
827        }
828    }
829
830    // ========================================================================
831    // Fetch by ID
832    // ========================================================================
833
834    /// Fetch vectors by their IDs
835    #[instrument(skip(self, request), fields(id_count = request.ids.len()))]
836    pub async fn fetch(&self, namespace: &str, request: FetchRequest) -> Result<FetchResponse> {
837        let url = format!("{}/v1/namespaces/{}/fetch", self.base_url, namespace);
838        debug!("Fetching {} vectors from {}", request.ids.len(), namespace);
839        let response = self.client.post(&url).json(&request).send().await?;
840        self.handle_response(response).await
841    }
842
843    /// Fetch vectors by IDs (convenience method)
844    #[instrument(skip(self))]
845    pub async fn fetch_by_ids(&self, namespace: &str, ids: &[&str]) -> Result<Vec<Vector>> {
846        let request = FetchRequest::new(ids.iter().map(|s| s.to_string()).collect());
847        self.fetch(namespace, request).await.map(|r| r.vectors)
848    }
849
850    // ========================================================================
851    // Text Auto-Embedding Operations
852    // ========================================================================
853
854    /// Upsert text documents with automatic server-side embedding generation
855    #[instrument(skip(self, request), fields(doc_count = request.documents.len()))]
856    pub async fn upsert_text(
857        &self,
858        namespace: &str,
859        request: UpsertTextRequest,
860    ) -> Result<TextUpsertResponse> {
861        let url = format!("{}/v1/namespaces/{}/upsert-text", self.base_url, namespace);
862        debug!(
863            "Upserting {} text documents to {}",
864            request.documents.len(),
865            namespace
866        );
867        let response = self.client.post(&url).json(&request).send().await?;
868        self.handle_response(response).await
869    }
870
871    /// Query using natural language text with automatic server-side embedding
872    #[instrument(skip(self, request), fields(top_k = request.top_k))]
873    pub async fn query_text(
874        &self,
875        namespace: &str,
876        request: QueryTextRequest,
877    ) -> Result<TextQueryResponse> {
878        let url = format!("{}/v1/namespaces/{}/query-text", self.base_url, namespace);
879        debug!("Text query in {} for: {}", namespace, request.text);
880        let response = self.client.post(&url).json(&request).send().await?;
881        self.handle_response(response).await
882    }
883
884    /// Query text (convenience method)
885    #[instrument(skip(self))]
886    pub async fn query_text_simple(
887        &self,
888        namespace: &str,
889        text: &str,
890        top_k: u32,
891    ) -> Result<TextQueryResponse> {
892        self.query_text(namespace, QueryTextRequest::new(text, top_k))
893            .await
894    }
895
896    /// Execute multiple text queries with automatic embedding in a single request
897    #[instrument(skip(self, request), fields(query_count = request.queries.len()))]
898    pub async fn batch_query_text(
899        &self,
900        namespace: &str,
901        request: BatchQueryTextRequest,
902    ) -> Result<BatchQueryTextResponse> {
903        let url = format!(
904            "{}/v1/namespaces/{}/batch-query-text",
905            self.base_url, namespace
906        );
907        debug!(
908            "Batch text query in {} with {} queries",
909            namespace,
910            request.queries.len()
911        );
912        let response = self.client.post(&url).json(&request).send().await?;
913        self.handle_response(response).await
914    }
915
916    // ========================================================================
917    // CE-4: GLiNER Entity Extraction
918    // ========================================================================
919
920    /// Configure namespace-level entity extraction settings (CE-4).
921    ///
922    /// Sends `PATCH /v1/namespaces/{namespace}/config` with the provided
923    /// [`NamespaceNerConfig`].
924    #[instrument(skip(self, config))]
925    pub async fn configure_namespace_ner(
926        &self,
927        namespace: &str,
928        config: NamespaceNerConfig,
929    ) -> Result<serde_json::Value> {
930        let url = format!("{}/v1/namespaces/{}/config", self.base_url, namespace);
931        let response = self.client.patch(&url).json(&config).send().await?;
932        self.handle_response(response).await
933    }
934
935    /// Extract entities from arbitrary text using the GLiNER pipeline (CE-4).
936    ///
937    /// Sends `POST /v1/memories/extract` with the supplied text and optional
938    /// entity type list.
939    #[instrument(skip(self, text, entity_types))]
940    pub async fn extract_entities(
941        &self,
942        text: &str,
943        entity_types: Option<Vec<String>>,
944    ) -> Result<EntityExtractionResponse> {
945        let url = format!("{}/v1/memories/extract", self.base_url);
946        let body = serde_json::json!({
947            "text": text,
948            "entity_types": entity_types,
949        });
950        let response = self.client.post(&url).json(&body).send().await?;
951        self.handle_response(response).await
952    }
953
954    /// Retrieve entity tags associated with a stored memory (CE-4).
955    ///
956    /// Sends `GET /v1/memory/entities/{memory_id}`.
957    #[instrument(skip(self))]
958    pub async fn memory_entities(&self, memory_id: &str) -> Result<MemoryEntitiesResponse> {
959        let url = format!("{}/v1/memory/entities/{}", self.base_url, memory_id);
960        let response = self.client.get(&url).send().await?;
961        self.handle_response(response).await
962    }
963
964    // ========================================================================
965    // Private Helpers
966    // ========================================================================
967
968    /// Rate-limit headers from the most recent API response (OPS-1).
969    ///
970    /// Returns `None` until the first successful request has been made.
971    pub fn last_rate_limit_headers(&self) -> Option<RateLimitHeaders> {
972        self.last_rate_limit.lock().ok()?.clone()
973    }
974
975    /// Handle response and deserialize JSON
976    pub(crate) async fn handle_response<T: serde::de::DeserializeOwned>(
977        &self,
978        response: reqwest::Response,
979    ) -> Result<T> {
980        let status = response.status();
981
982        // OPS-1: capture rate-limit headers before consuming the response body
983        if let Ok(mut guard) = self.last_rate_limit.lock() {
984            *guard = Some(RateLimitHeaders::from_response(&response));
985        }
986
987        if status.is_success() {
988            Ok(response.json().await?)
989        } else {
990            let status_code = status.as_u16();
991            // Extract Retry-After before consuming response
992            let retry_after = response
993                .headers()
994                .get("Retry-After")
995                .and_then(|v| v.to_str().ok())
996                .and_then(|s| s.parse::<u64>().ok());
997            let text = response.text().await.unwrap_or_default();
998
999            if status_code == 429 {
1000                return Err(ClientError::RateLimitExceeded { retry_after });
1001            }
1002
1003            #[derive(Deserialize)]
1004            struct ErrorBody {
1005                error: Option<String>,
1006                code: Option<ServerErrorCode>,
1007            }
1008
1009            let (message, code) = if let Ok(body) = serde_json::from_str::<ErrorBody>(&text) {
1010                (body.error.unwrap_or_else(|| text.clone()), body.code)
1011            } else {
1012                (text, None)
1013            };
1014
1015            match status_code {
1016                401 => Err(ClientError::Server {
1017                    status: 401,
1018                    message,
1019                    code,
1020                }),
1021                403 => Err(ClientError::Authorization {
1022                    status: 403,
1023                    message,
1024                    code,
1025                }),
1026                404 => match &code {
1027                    Some(ServerErrorCode::NamespaceNotFound) => {
1028                        Err(ClientError::NamespaceNotFound(message))
1029                    }
1030                    Some(ServerErrorCode::VectorNotFound) => {
1031                        Err(ClientError::VectorNotFound(message))
1032                    }
1033                    _ => Err(ClientError::Server {
1034                        status: 404,
1035                        message,
1036                        code,
1037                    }),
1038                },
1039                _ => Err(ClientError::Server {
1040                    status: status_code,
1041                    message,
1042                    code,
1043                }),
1044            }
1045        }
1046    }
1047
1048    /// Execute a fallible async operation with retry logic and exponential backoff.
1049    ///
1050    /// Retries on transient errors (5xx, rate-limit, connection/timeout).
1051    /// Respects the `Retry-After` header when the server returns HTTP 429.
1052    /// Does NOT retry on 4xx client errors (except 429).
1053    ///
1054    /// NOTE: API call-site wiring is deferred to a follow-up (infrastructure PR).
1055    #[allow(dead_code)]
1056    pub(crate) async fn execute_with_retry<F, Fut, T>(&self, f: F) -> Result<T>
1057    where
1058        F: Fn() -> Fut,
1059        Fut: std::future::Future<Output = Result<T>>,
1060    {
1061        let rc = &self.retry_config;
1062
1063        for attempt in 0..rc.max_retries {
1064            match f().await {
1065                Ok(v) => return Ok(v),
1066                Err(e) => {
1067                    let is_last = attempt == rc.max_retries - 1;
1068                    if is_last || !e.is_retryable() {
1069                        return Err(e);
1070                    }
1071
1072                    let wait = match &e {
1073                        ClientError::RateLimitExceeded {
1074                            retry_after: Some(secs),
1075                        } => Duration::from_secs(*secs),
1076                        _ => {
1077                            let base_ms = rc.base_delay.as_millis() as f64;
1078                            let backoff_ms = base_ms * 2f64.powi(attempt as i32);
1079                            let capped_ms = backoff_ms.min(rc.max_delay.as_millis() as f64);
1080                            let final_ms = if rc.jitter {
1081                                // Simple deterministic jitter: vary between 50% and 150%
1082                                let seed = (attempt as u64).wrapping_mul(6364136223846793005);
1083                                let factor = 0.5 + (seed % 1000) as f64 / 1000.0;
1084                                capped_ms * factor
1085                            } else {
1086                                capped_ms
1087                            };
1088                            Duration::from_millis(final_ms as u64)
1089                        }
1090                    };
1091
1092                    tokio::time::sleep(wait).await;
1093                }
1094            }
1095        }
1096
1097        // Unreachable: the loop always returns on the last attempt
1098        Err(ClientError::Config("retry loop exhausted".to_string()))
1099    }
1100}
1101
1102/// Builder for DakeraClient
1103#[derive(Debug)]
1104pub struct DakeraClientBuilder {
1105    base_url: String,
1106    timeout: Duration,
1107    connect_timeout: Option<Duration>,
1108    retry_config: RetryConfig,
1109    user_agent: Option<String>,
1110}
1111
1112impl DakeraClientBuilder {
1113    /// Create a new builder
1114    pub fn new(base_url: impl Into<String>) -> Self {
1115        Self {
1116            base_url: base_url.into(),
1117            timeout: Duration::from_secs(DEFAULT_TIMEOUT_SECS),
1118            connect_timeout: None,
1119            retry_config: RetryConfig::default(),
1120            user_agent: None,
1121        }
1122    }
1123
1124    /// Set the request timeout
1125    pub fn timeout(mut self, timeout: Duration) -> Self {
1126        self.timeout = timeout;
1127        self
1128    }
1129
1130    /// Set the request timeout in seconds
1131    pub fn timeout_secs(mut self, secs: u64) -> Self {
1132        self.timeout = Duration::from_secs(secs);
1133        self
1134    }
1135
1136    /// Set the connection establishment timeout (defaults to `timeout`).
1137    pub fn connect_timeout(mut self, timeout: Duration) -> Self {
1138        self.connect_timeout = Some(timeout);
1139        self
1140    }
1141
1142    /// Set fine-grained retry configuration.
1143    pub fn retry_config(mut self, config: RetryConfig) -> Self {
1144        self.retry_config = config;
1145        self
1146    }
1147
1148    /// Set the maximum number of retry attempts.
1149    pub fn max_retries(mut self, max_retries: u32) -> Self {
1150        self.retry_config.max_retries = max_retries;
1151        self
1152    }
1153
1154    /// Set a custom user agent
1155    pub fn user_agent(mut self, user_agent: impl Into<String>) -> Self {
1156        self.user_agent = Some(user_agent.into());
1157        self
1158    }
1159
1160    /// Build the client
1161    pub fn build(self) -> Result<DakeraClient> {
1162        // Normalize base URL (remove trailing slash)
1163        let base_url = self.base_url.trim_end_matches('/').to_string();
1164
1165        // Validate URL
1166        if !base_url.starts_with("http://") && !base_url.starts_with("https://") {
1167            return Err(ClientError::InvalidUrl(
1168                "URL must start with http:// or https://".to_string(),
1169            ));
1170        }
1171
1172        let user_agent = self
1173            .user_agent
1174            .unwrap_or_else(|| format!("dakera-client/{}", env!("CARGO_PKG_VERSION")));
1175
1176        let connect_timeout = self.connect_timeout.unwrap_or(self.timeout);
1177
1178        let client = Client::builder()
1179            .timeout(self.timeout)
1180            .connect_timeout(connect_timeout)
1181            .user_agent(user_agent)
1182            .build()
1183            .map_err(|e| ClientError::Config(e.to_string()))?;
1184
1185        Ok(DakeraClient {
1186            client,
1187            base_url,
1188            retry_config: self.retry_config,
1189            last_rate_limit: Arc::new(Mutex::new(None)),
1190        })
1191    }
1192}
1193
1194// ============================================================================
1195// SSE Streaming (CE-1)
1196// ============================================================================
1197
1198impl DakeraClient {
1199    /// Subscribe to namespace-scoped SSE events.
1200    ///
1201    /// Opens a long-lived connection to `GET /v1/namespaces/{namespace}/events`
1202    /// and returns a [`tokio::sync::mpsc::Receiver`] that yields
1203    /// [`DakeraEvent`] results as they arrive.  The background task exits when
1204    /// the server closes the stream or the receiver is dropped.
1205    ///
1206    /// Requires a Read-scoped API key.
1207    ///
1208    /// # Example
1209    ///
1210    /// ```rust,no_run
1211    /// use dakera_client::DakeraClient;
1212    ///
1213    /// #[tokio::main]
1214    /// async fn main() -> Result<(), Box<dyn std::error::Error>> {
1215    ///     let client = DakeraClient::new("http://localhost:3000")?;
1216    ///     let mut rx = client.stream_namespace_events("my-ns").await?;
1217    ///     while let Some(result) = rx.recv().await {
1218    ///         println!("{:?}", result?);
1219    ///     }
1220    ///     Ok(())
1221    /// }
1222    /// ```
1223    pub async fn stream_namespace_events(
1224        &self,
1225        namespace: &str,
1226    ) -> Result<tokio::sync::mpsc::Receiver<Result<crate::events::DakeraEvent>>> {
1227        let url = format!(
1228            "{}/v1/namespaces/{}/events",
1229            self.base_url,
1230            urlencoding::encode(namespace)
1231        );
1232        self.stream_sse(url).await
1233    }
1234
1235    /// Subscribe to the global SSE event stream (all namespaces).
1236    ///
1237    /// Opens a long-lived connection to `GET /ops/events` and returns a
1238    /// [`tokio::sync::mpsc::Receiver`] that yields [`DakeraEvent`] results.
1239    ///
1240    /// Requires an Admin-scoped API key.
1241    pub async fn stream_global_events(
1242        &self,
1243    ) -> Result<tokio::sync::mpsc::Receiver<Result<crate::events::DakeraEvent>>> {
1244        let url = format!("{}/ops/events", self.base_url);
1245        self.stream_sse(url).await
1246    }
1247
1248    /// Subscribe to the memory lifecycle SSE event stream (DASH-B).
1249    ///
1250    /// Opens a long-lived connection to `GET /v1/events/stream` and returns a
1251    /// [`tokio::sync::mpsc::Receiver`] that yields [`MemoryEvent`] results as
1252    /// they arrive.  The background task exits when the server closes the stream
1253    /// or the receiver is dropped.
1254    ///
1255    /// Requires a Read-scoped API key.
1256    pub async fn stream_memory_events(
1257        &self,
1258    ) -> Result<tokio::sync::mpsc::Receiver<Result<crate::events::MemoryEvent>>> {
1259        let url = format!("{}/v1/events/stream", self.base_url);
1260        self.stream_sse(url).await
1261    }
1262
1263    /// Low-level generic SSE streaming helper.
1264    async fn stream_sse<T>(&self, url: String) -> Result<tokio::sync::mpsc::Receiver<Result<T>>>
1265    where
1266        T: serde::de::DeserializeOwned + Send + 'static,
1267    {
1268        use futures_util::StreamExt;
1269
1270        let response = self
1271            .client
1272            .get(&url)
1273            .header("Accept", "text/event-stream")
1274            .header("Cache-Control", "no-cache")
1275            .send()
1276            .await?;
1277
1278        if !response.status().is_success() {
1279            let status = response.status().as_u16();
1280            let body = response.text().await.unwrap_or_default();
1281            return Err(ClientError::Server {
1282                status,
1283                message: body,
1284                code: None,
1285            });
1286        }
1287
1288        let (tx, rx) = tokio::sync::mpsc::channel(64);
1289
1290        tokio::spawn(async move {
1291            let mut byte_stream = response.bytes_stream();
1292            let mut remaining = String::new();
1293            let mut data_lines: Vec<String> = Vec::new();
1294
1295            while let Some(chunk) = byte_stream.next().await {
1296                match chunk {
1297                    Ok(bytes) => {
1298                        remaining.push_str(&String::from_utf8_lossy(&bytes));
1299                        while let Some(pos) = remaining.find('\n') {
1300                            let raw = &remaining[..pos];
1301                            let line = raw.trim_end_matches('\r').to_string();
1302                            remaining = remaining[pos + 1..].to_string();
1303
1304                            if line.starts_with(':') {
1305                                // SSE comment / heartbeat — skip
1306                            } else if let Some(data) = line.strip_prefix("data:") {
1307                                data_lines.push(data.trim_start().to_string());
1308                            } else if line.is_empty() {
1309                                if !data_lines.is_empty() {
1310                                    let payload = data_lines.join("\n");
1311                                    data_lines.clear();
1312                                    let result = serde_json::from_str::<T>(&payload)
1313                                        .map_err(ClientError::Json);
1314                                    if tx.send(result).await.is_err() {
1315                                        return; // receiver dropped
1316                                    }
1317                                }
1318                            } else {
1319                                // Unrecognised field (e.g. "event:") — ignore
1320                            }
1321                        }
1322                    }
1323                    Err(e) => {
1324                        let _ = tx.send(Err(ClientError::Http(e))).await;
1325                        return;
1326                    }
1327                }
1328            }
1329        });
1330
1331        Ok(rx)
1332    }
1333}
1334
1335#[cfg(test)]
1336mod tests {
1337    use super::*;
1338
1339    #[test]
1340    fn test_client_builder() {
1341        let client = DakeraClient::new("http://localhost:3000");
1342        assert!(client.is_ok());
1343    }
1344
1345    #[test]
1346    fn test_client_builder_with_options() {
1347        let client = DakeraClient::builder("http://localhost:3000")
1348            .timeout_secs(60)
1349            .user_agent("test-client/1.0")
1350            .build();
1351        assert!(client.is_ok());
1352    }
1353
1354    #[test]
1355    fn test_client_builder_invalid_url() {
1356        let client = DakeraClient::new("invalid-url");
1357        assert!(client.is_err());
1358    }
1359
1360    #[test]
1361    fn test_client_builder_trailing_slash() {
1362        let client = DakeraClient::new("http://localhost:3000/").unwrap();
1363        assert!(!client.base_url.ends_with('/'));
1364    }
1365
1366    #[test]
1367    fn test_vector_creation() {
1368        let v = Vector::new("test", vec![0.1, 0.2, 0.3]);
1369        assert_eq!(v.id, "test");
1370        assert_eq!(v.values.len(), 3);
1371        assert!(v.metadata.is_none());
1372    }
1373
1374    #[test]
1375    fn test_query_request_builder() {
1376        let req = QueryRequest::new(vec![0.1, 0.2], 10)
1377            .with_filter(serde_json::json!({"category": "test"}))
1378            .include_metadata(false);
1379
1380        assert_eq!(req.top_k, 10);
1381        assert!(req.filter.is_some());
1382        assert!(!req.include_metadata);
1383    }
1384
1385    #[test]
1386    fn test_hybrid_search_request() {
1387        let req = HybridSearchRequest::new(vec![0.1], "test query", 5).with_vector_weight(0.7);
1388
1389        assert_eq!(req.vector_weight, 0.7);
1390        assert_eq!(req.text, "test query");
1391        assert!(req.vector.is_some());
1392    }
1393
1394    #[test]
1395    fn test_hybrid_search_weight_clamping() {
1396        let req = HybridSearchRequest::new(vec![0.1], "test", 5).with_vector_weight(1.5); // Should be clamped to 1.0
1397
1398        assert_eq!(req.vector_weight, 1.0);
1399    }
1400
1401    #[test]
1402    fn test_hybrid_search_text_only() {
1403        let req = HybridSearchRequest::text_only("bm25 query", 10);
1404
1405        assert!(req.vector.is_none());
1406        assert_eq!(req.text, "bm25 query");
1407        assert_eq!(req.top_k, 10);
1408        // Verify vector is not serialised
1409        let json = serde_json::to_value(&req).unwrap();
1410        assert!(json.get("vector").is_none());
1411    }
1412
1413    #[test]
1414    fn test_text_document_builder() {
1415        let doc = TextDocument::new("doc1", "Hello world").with_ttl(3600);
1416
1417        assert_eq!(doc.id, "doc1");
1418        assert_eq!(doc.text, "Hello world");
1419        assert_eq!(doc.ttl_seconds, Some(3600));
1420        assert!(doc.metadata.is_none());
1421    }
1422
1423    #[test]
1424    fn test_upsert_text_request_builder() {
1425        let docs = vec![
1426            TextDocument::new("doc1", "Hello"),
1427            TextDocument::new("doc2", "World"),
1428        ];
1429        let req = UpsertTextRequest::new(docs).with_model(EmbeddingModel::BgeSmall);
1430
1431        assert_eq!(req.documents.len(), 2);
1432        assert_eq!(req.model, Some(EmbeddingModel::BgeSmall));
1433    }
1434
1435    #[test]
1436    fn test_query_text_request_builder() {
1437        let req = QueryTextRequest::new("semantic search query", 5)
1438            .with_filter(serde_json::json!({"category": "docs"}))
1439            .include_vectors(true)
1440            .with_model(EmbeddingModel::E5Small);
1441
1442        assert_eq!(req.text, "semantic search query");
1443        assert_eq!(req.top_k, 5);
1444        assert!(req.filter.is_some());
1445        assert!(req.include_vectors);
1446        assert_eq!(req.model, Some(EmbeddingModel::E5Small));
1447    }
1448
1449    #[test]
1450    fn test_fetch_request_builder() {
1451        let req = FetchRequest::new(vec!["id1".to_string(), "id2".to_string()]);
1452
1453        assert_eq!(req.ids.len(), 2);
1454        assert!(req.include_values);
1455        assert!(req.include_metadata);
1456    }
1457
1458    #[test]
1459    fn test_create_namespace_request_builder() {
1460        let req = CreateNamespaceRequest::new()
1461            .with_dimensions(384)
1462            .with_index_type("hnsw");
1463
1464        assert_eq!(req.dimensions, Some(384));
1465        assert_eq!(req.index_type.as_deref(), Some("hnsw"));
1466    }
1467
1468    #[test]
1469    fn test_batch_query_text_request() {
1470        let req =
1471            BatchQueryTextRequest::new(vec!["query one".to_string(), "query two".to_string()], 10);
1472
1473        assert_eq!(req.queries.len(), 2);
1474        assert_eq!(req.top_k, 10);
1475        assert!(!req.include_vectors);
1476        assert!(req.model.is_none());
1477    }
1478
1479    // =========================================================================
1480    // RetryConfig tests
1481    // =========================================================================
1482
1483    #[test]
1484    fn test_retry_config_defaults() {
1485        let rc = RetryConfig::default();
1486        assert_eq!(rc.max_retries, 3);
1487        assert_eq!(rc.base_delay, Duration::from_millis(100));
1488        assert_eq!(rc.max_delay, Duration::from_secs(60));
1489        assert!(rc.jitter);
1490    }
1491
1492    #[test]
1493    fn test_builder_connect_timeout() {
1494        let client = DakeraClient::builder("http://localhost:3000")
1495            .connect_timeout(Duration::from_secs(5))
1496            .timeout_secs(30)
1497            .build()
1498            .unwrap();
1499        // Client was built successfully with separate connect timeout
1500        assert!(client.base_url.starts_with("http"));
1501    }
1502
1503    #[test]
1504    fn test_builder_max_retries() {
1505        let client = DakeraClient::builder("http://localhost:3000")
1506            .max_retries(5)
1507            .build()
1508            .unwrap();
1509        assert_eq!(client.retry_config.max_retries, 5);
1510    }
1511
1512    #[test]
1513    fn test_builder_retry_config() {
1514        let rc = RetryConfig {
1515            max_retries: 7,
1516            base_delay: Duration::from_millis(200),
1517            max_delay: Duration::from_secs(30),
1518            jitter: false,
1519        };
1520        let client = DakeraClient::builder("http://localhost:3000")
1521            .retry_config(rc)
1522            .build()
1523            .unwrap();
1524        assert_eq!(client.retry_config.max_retries, 7);
1525        assert!(!client.retry_config.jitter);
1526    }
1527
1528    #[test]
1529    fn test_rate_limit_error_retryable() {
1530        let e = ClientError::RateLimitExceeded { retry_after: None };
1531        assert!(e.is_retryable());
1532    }
1533
1534    #[test]
1535    fn test_rate_limit_error_with_retry_after_zero() {
1536        // retry_after: Some(0) should still be Some, not treated as missing
1537        let e = ClientError::RateLimitExceeded {
1538            retry_after: Some(0),
1539        };
1540        assert!(e.is_retryable());
1541        if let ClientError::RateLimitExceeded {
1542            retry_after: Some(secs),
1543        } = &e
1544        {
1545            assert_eq!(*secs, 0u64);
1546        } else {
1547            panic!("unexpected variant");
1548        }
1549    }
1550
1551    #[tokio::test]
1552    async fn test_execute_with_retry_succeeds_immediately() {
1553        let client = DakeraClient::builder("http://localhost:3000")
1554            .max_retries(3)
1555            .build()
1556            .unwrap();
1557
1558        let call_count = std::sync::Arc::new(std::sync::atomic::AtomicU32::new(0));
1559        let cc = call_count.clone();
1560        let result = client
1561            .execute_with_retry(|| {
1562                let cc = cc.clone();
1563                async move {
1564                    cc.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
1565                    Ok::<u32, ClientError>(42)
1566                }
1567            })
1568            .await;
1569        assert_eq!(result.unwrap(), 42);
1570        assert_eq!(call_count.load(std::sync::atomic::Ordering::SeqCst), 1);
1571    }
1572
1573    #[tokio::test]
1574    async fn test_execute_with_retry_no_retry_on_4xx() {
1575        let client = DakeraClient::builder("http://localhost:3000")
1576            .max_retries(3)
1577            .build()
1578            .unwrap();
1579
1580        let call_count = std::sync::Arc::new(std::sync::atomic::AtomicU32::new(0));
1581        let cc = call_count.clone();
1582        let result = client
1583            .execute_with_retry(|| {
1584                let cc = cc.clone();
1585                async move {
1586                    cc.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
1587                    Err::<u32, ClientError>(ClientError::Server {
1588                        status: 400,
1589                        message: "bad request".to_string(),
1590                        code: None,
1591                    })
1592                }
1593            })
1594            .await;
1595        assert!(result.is_err());
1596        // Should not retry on 4xx
1597        assert_eq!(call_count.load(std::sync::atomic::Ordering::SeqCst), 1);
1598    }
1599
1600    #[tokio::test]
1601    async fn test_execute_with_retry_retries_on_5xx() {
1602        let client = DakeraClient::builder("http://localhost:3000")
1603            .retry_config(RetryConfig {
1604                max_retries: 3,
1605                base_delay: Duration::from_millis(0),
1606                max_delay: Duration::from_millis(0),
1607                jitter: false,
1608            })
1609            .build()
1610            .unwrap();
1611
1612        let call_count = std::sync::Arc::new(std::sync::atomic::AtomicU32::new(0));
1613        let cc = call_count.clone();
1614        let result = client
1615            .execute_with_retry(|| {
1616                let cc = cc.clone();
1617                async move {
1618                    let n = cc.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
1619                    if n < 2 {
1620                        Err::<u32, ClientError>(ClientError::Server {
1621                            status: 503,
1622                            message: "unavailable".to_string(),
1623                            code: None,
1624                        })
1625                    } else {
1626                        Ok(99)
1627                    }
1628                }
1629            })
1630            .await;
1631        assert_eq!(result.unwrap(), 99);
1632        assert_eq!(call_count.load(std::sync::atomic::Ordering::SeqCst), 3);
1633    }
1634
1635    // =========================================================================
1636    // CE-2: Batch Recall / Forget (v0.7.0)
1637    // =========================================================================
1638
1639    #[test]
1640    fn test_batch_recall_request_new() {
1641        use crate::memory::BatchRecallRequest;
1642        let req = BatchRecallRequest::new("agent-1");
1643        assert_eq!(req.agent_id, "agent-1");
1644        assert_eq!(req.limit, 100);
1645    }
1646
1647    #[test]
1648    fn test_batch_recall_request_builder() {
1649        use crate::memory::{BatchMemoryFilter, BatchRecallRequest};
1650        let filter = BatchMemoryFilter::default()
1651            .with_tags(vec!["qa".to_string()])
1652            .with_min_importance(0.7);
1653        let req = BatchRecallRequest::new("agent-1")
1654            .with_filter(filter)
1655            .with_limit(50);
1656        assert_eq!(req.agent_id, "agent-1");
1657        assert_eq!(req.limit, 50);
1658        assert_eq!(
1659            req.filter.tags.as_deref(),
1660            Some(["qa".to_string()].as_slice())
1661        );
1662        assert_eq!(req.filter.min_importance, Some(0.7));
1663    }
1664
1665    #[test]
1666    fn test_batch_recall_request_serialization() {
1667        use crate::memory::{BatchMemoryFilter, BatchRecallRequest};
1668        let filter = BatchMemoryFilter::default().with_min_importance(0.5);
1669        let req = BatchRecallRequest::new("agent-1")
1670            .with_filter(filter)
1671            .with_limit(25);
1672        let json = serde_json::to_value(&req).unwrap();
1673        assert_eq!(json["agent_id"], "agent-1");
1674        assert_eq!(json["limit"], 25);
1675        assert_eq!(json["filter"]["min_importance"], 0.5);
1676    }
1677
1678    #[test]
1679    fn test_batch_forget_request_new() {
1680        use crate::memory::{BatchForgetRequest, BatchMemoryFilter};
1681        let filter = BatchMemoryFilter::default().with_min_importance(0.1);
1682        let req = BatchForgetRequest::new("agent-1", filter);
1683        assert_eq!(req.agent_id, "agent-1");
1684        assert_eq!(req.filter.min_importance, Some(0.1));
1685    }
1686
1687    #[test]
1688    fn test_batch_forget_request_serialization() {
1689        use crate::memory::{BatchForgetRequest, BatchMemoryFilter};
1690        let filter = BatchMemoryFilter {
1691            created_before: Some(1_700_000_000),
1692            ..Default::default()
1693        };
1694        let req = BatchForgetRequest::new("agent-1", filter);
1695        let json = serde_json::to_value(&req).unwrap();
1696        assert_eq!(json["agent_id"], "agent-1");
1697        assert_eq!(json["filter"]["created_before"], 1_700_000_000u64);
1698    }
1699
1700    #[test]
1701    fn test_batch_recall_response_deserialization() {
1702        use crate::memory::BatchRecallResponse;
1703        let json = serde_json::json!({
1704            "memories": [],
1705            "total": 42,
1706            "filtered": 7
1707        });
1708        let resp: BatchRecallResponse = serde_json::from_value(json).unwrap();
1709        assert_eq!(resp.total, 42);
1710        assert_eq!(resp.filtered, 7);
1711        assert!(resp.memories.is_empty());
1712    }
1713
1714    #[test]
1715    fn test_batch_forget_response_deserialization() {
1716        use crate::memory::BatchForgetResponse;
1717        let json = serde_json::json!({ "deleted_count": 13 });
1718        let resp: BatchForgetResponse = serde_json::from_value(json).unwrap();
1719        assert_eq!(resp.deleted_count, 13);
1720    }
1721
1722    // =========================================================================
1723    // OPS-1: RateLimitHeaders (v0.7.0)
1724    // =========================================================================
1725
1726    #[test]
1727    fn test_rate_limit_headers_default_all_none() {
1728        use crate::types::RateLimitHeaders;
1729        let rl = RateLimitHeaders {
1730            limit: None,
1731            remaining: None,
1732            reset: None,
1733            quota_used: None,
1734            quota_limit: None,
1735        };
1736        assert!(rl.limit.is_none());
1737        assert!(rl.remaining.is_none());
1738        assert!(rl.reset.is_none());
1739        assert!(rl.quota_used.is_none());
1740        assert!(rl.quota_limit.is_none());
1741    }
1742
1743    #[test]
1744    fn test_rate_limit_headers_populated() {
1745        use crate::types::RateLimitHeaders;
1746        let rl = RateLimitHeaders {
1747            limit: Some(1000),
1748            remaining: Some(750),
1749            reset: Some(1_700_000_060),
1750            quota_used: Some(500),
1751            quota_limit: Some(10_000),
1752        };
1753        assert_eq!(rl.limit, Some(1000));
1754        assert_eq!(rl.remaining, Some(750));
1755        assert_eq!(rl.reset, Some(1_700_000_060));
1756        assert_eq!(rl.quota_used, Some(500));
1757        assert_eq!(rl.quota_limit, Some(10_000));
1758    }
1759
1760    #[test]
1761    fn test_last_rate_limit_headers_initially_none() {
1762        let client = DakeraClient::new("http://localhost:3000").unwrap();
1763        assert!(client.last_rate_limit_headers().is_none());
1764    }
1765
1766    // =========================================================================
1767    // CE-4: GLiNER Entity Extraction
1768    // =========================================================================
1769
1770    #[test]
1771    fn test_namespace_ner_config_default() {
1772        use crate::types::NamespaceNerConfig;
1773        let cfg = NamespaceNerConfig::default();
1774        assert!(!cfg.extract_entities);
1775        assert!(cfg.entity_types.is_none());
1776    }
1777
1778    #[test]
1779    fn test_namespace_ner_config_serialization_skip_none() {
1780        use crate::types::NamespaceNerConfig;
1781        let cfg = NamespaceNerConfig {
1782            extract_entities: true,
1783            entity_types: None,
1784        };
1785        let json = serde_json::to_value(&cfg).unwrap();
1786        assert_eq!(json["extract_entities"], true);
1787        // entity_types should be omitted when None
1788        assert!(json.get("entity_types").is_none());
1789    }
1790
1791    #[test]
1792    fn test_namespace_ner_config_serialization_with_types() {
1793        use crate::types::NamespaceNerConfig;
1794        let cfg = NamespaceNerConfig {
1795            extract_entities: true,
1796            entity_types: Some(vec!["PERSON".to_string(), "ORG".to_string()]),
1797        };
1798        let json = serde_json::to_value(&cfg).unwrap();
1799        assert_eq!(json["extract_entities"], true);
1800        assert_eq!(json["entity_types"][0], "PERSON");
1801        assert_eq!(json["entity_types"][1], "ORG");
1802    }
1803
1804    #[test]
1805    fn test_extracted_entity_deserialization() {
1806        use crate::types::ExtractedEntity;
1807        let json = serde_json::json!({
1808            "entity_type": "PERSON",
1809            "value": "Alice",
1810            "score": 0.95
1811        });
1812        let entity: ExtractedEntity = serde_json::from_value(json).unwrap();
1813        assert_eq!(entity.entity_type, "PERSON");
1814        assert_eq!(entity.value, "Alice");
1815        assert!((entity.score - 0.95).abs() < f64::EPSILON);
1816    }
1817
1818    #[test]
1819    fn test_entity_extraction_response_deserialization() {
1820        use crate::types::EntityExtractionResponse;
1821        let json = serde_json::json!({
1822            "entities": [
1823                { "entity_type": "PERSON", "value": "Bob", "score": 0.9 },
1824                { "entity_type": "ORG",    "value": "Acme", "score": 0.87 }
1825            ]
1826        });
1827        let resp: EntityExtractionResponse = serde_json::from_value(json).unwrap();
1828        assert_eq!(resp.entities.len(), 2);
1829        assert_eq!(resp.entities[0].entity_type, "PERSON");
1830        assert_eq!(resp.entities[1].value, "Acme");
1831    }
1832
1833    #[test]
1834    fn test_memory_entities_response_deserialization() {
1835        use crate::types::MemoryEntitiesResponse;
1836        let json = serde_json::json!({
1837            "memory_id": "mem-abc-123",
1838            "entities": [
1839                { "entity_type": "LOC", "value": "London", "score": 0.88 }
1840            ]
1841        });
1842        let resp: MemoryEntitiesResponse = serde_json::from_value(json).unwrap();
1843        assert_eq!(resp.memory_id, "mem-abc-123");
1844        assert_eq!(resp.entities.len(), 1);
1845        assert_eq!(resp.entities[0].entity_type, "LOC");
1846        assert_eq!(resp.entities[0].value, "London");
1847    }
1848
1849    #[test]
1850    fn test_configure_namespace_ner_url_pattern() {
1851        // Verify the client is constructable and base_url is correct
1852        let client = DakeraClient::new("http://localhost:3000").unwrap();
1853        let expected = "http://localhost:3000/v1/namespaces/my-ns/config";
1854        let actual = format!("{}/v1/namespaces/{}/config", client.base_url, "my-ns");
1855        assert_eq!(actual, expected);
1856    }
1857
1858    #[test]
1859    fn test_extract_entities_url_pattern() {
1860        let client = DakeraClient::new("http://localhost:3000").unwrap();
1861        let expected = "http://localhost:3000/v1/memories/extract";
1862        let actual = format!("{}/v1/memories/extract", client.base_url);
1863        assert_eq!(actual, expected);
1864    }
1865
1866    #[test]
1867    fn test_memory_entities_url_pattern() {
1868        let client = DakeraClient::new("http://localhost:3000").unwrap();
1869        let memory_id = "mem-xyz-789";
1870        let expected = "http://localhost:3000/v1/memory/entities/mem-xyz-789";
1871        let actual = format!("{}/v1/memory/entities/{}", client.base_url, memory_id);
1872        assert_eq!(actual, expected);
1873    }
1874}
dakera_client/client.rs

dakera_client/
client.rs
