AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.
Rust agent-memoryai-agentsbm25claude-codeclicode-searchcursordatabaseelasticsearchfull-text-searchhybrid-searchjsonllocal-firstmcpmcp-servermodel-context-protocolragrustsearch-enginevector-search
fix(es-compat): stop three silent-wrong-answer behaviours (fail loud / de-fabricate)
xerj-org63e8aad1de04
diff --git a/engine/crates/xerj-api/src/es_compat.rs b/engine/crates/xerj-api/src/es_compat.rsindex 96ac666cd..c4bebdd81 100644--- a/engine/crates/xerj-api/src/es_compat.rs+++ b/engine/crates/xerj-api/src/es_compat.rs@@ -8829,6 +8829,29 @@ pub async fn search(}).collect();+ // Phrase suggester is not implemented. Detect it and fail loudly with a+ // clear 400 rather than silently defaulting its field to `_all`, scanning+ // for a nonexistent field, and emitting empty/garbage term-style output.+ // Completion and term suggesters are supported.+ if let Some(sobj) = body.suggest.as_ref().and_then(Value::as_object) {+ if sobj.values().any(|d| d.get("phrase").is_some()) {+ let reason = "phrase suggester is not supported by XERJ; supported suggesters are \+ `completion` and `term`";+ return (+ StatusCode::BAD_REQUEST,+ Json(json!({+ "error": {+ "root_cause": [{ "type": "illegal_argument_exception", "reason": reason }],+ "type": "illegal_argument_exception",+ "reason": reason,+ },+ "status": 400,+ })),+ )+ .into_response();+ }+ }+// Process suggest block if present.let suggest_result = if let Some(ref suggest_body) = body.suggest {// Gather field-level indexed terms from all indices for suggest processing.@@ -11930,9 +11953,14 @@ fn tokenize_for_analyze(input: &str, analyzer: &str) -> Vec<Value> {// GET /{index}/_stats// ─────────────────────────────────────────────────────────────────────────────-/// `POST /{index}/_disk_usage` — disk space breakdown by field. We-/// synthesise a minimal response sufficient to satisfy the YAML tests-/// that assert `store_size_in_bytes > 0` after a doc is indexed.+/// `POST /{index}/_disk_usage` — on-disk store size.+///+/// `store_size_in_bytes` is the real recursive byte sum of the index's+/// `data_dir`. XERJ does NOT compute the ES per-field / per-category+/// breakdown (inverted_index / doc_values / points / norms / term_vectors /+/// knn_vectors), so `fields` is honestly empty rather than fabricated with+/// hardcoded zeros. The total is genuinely useful; only the decomposition is+/// unavailable, and we don't imply otherwise.pub async fn index_disk_usage(State(state): State<AppState>,Path(index): Path<String>,@@ -11954,25 +11982,11 @@ pub async fn index_disk_usage(indices.insert(name.clone(),json!({+ // Real on-disk total only. The ES per-field / per-category+ // breakdown is not computed, so `fields` is honestly empty+ // rather than fabricated zeros (see the handler doc-comment)."store_size": format!("{}b", size),"store_size_in_bytes": size,- "all_fields": {- "total": format!("{}b", size),- "total_in_bytes": size,- "inverted_index": { "total": "0b", "total_in_bytes": 0 },- "stored_fields": format!("{}b", size),- "stored_fields_in_bytes": size,- "doc_values": "0b",- "doc_values_in_bytes": 0,- "points": "0b",- "points_in_bytes": 0,- "norms": "0b",- "norms_in_bytes": 0,- "term_vectors": "0b",- "term_vectors_in_bytes": 0,- "knn_vectors": "0b",- "knn_vectors_in_bytes": 0,- },"fields": {}}),);@@ -20226,6 +20240,38 @@ pub async fn eql_search(}};+ // EQL sequence/sample queries (multi-event correlation) are not+ // implemented. Detect them by their leading keyword and fail loudly+ // rather than silently degrading to a single-condition query — or, for a+ // bare `sequence ... by ...` with no inner `where`, to `match_all` over+ // EVERY doc. A silent wrong answer is a correctness hazard for SIEM+ // users. Single-event queries (`<category> where <condition>`) are+ // supported and continue through the normal path below.+ let first_word: String = eql+ .trim_start()+ .chars()+ .take_while(|c| c.is_alphanumeric() || *c == '_')+ .collect::<String>()+ .to_ascii_lowercase();+ if first_word == "sequence" || first_word == "sample" {+ let reason = format!(+ "EQL {first_word} queries (multi-event correlation) are not supported by XERJ. \+ Only single-event EQL is implemented (`<category> where <condition>`)."+ );+ return (+ StatusCode::NOT_IMPLEMENTED,+ Json(json!({+ "error": {+ "root_cause": [{ "type": "verification_exception", "reason": reason }],+ "type": "verification_exception",+ "reason": reason,+ },+ "status": 501,+ })),+ )+ .into_response();+ }+// Translate EQL -> xerj DSL, then run it through the normal search path.let inner_query = eql_to_query(&eql);let size = body.get("size").and_then(Value::as_u64).unwrap_or(10);