@sentry/junior-memory 0.132.0 → 0.133.0

package/dist/index.js

@@ -345,6 +345,8 @@ var DEFAULT_EXPIRED_ARCHIVE_LIMIT = 100;
 var PREFERENCE_ADJUDICATION_CANDIDATE_LIMIT = 10;
 var PREFERENCE_ADJUDICATION_VECTOR_LIMIT = 5;
 var VECTOR_SEARCH_OVERFETCH = 4;
+var LEXICAL_RANK_OVERFETCH = 4;
+var MAX_LEXICAL_RANK_CANDIDATES = 1e3;
 var MAX_MEMORY_CONTENT_CHARS = 4e3;
 var EMBEDDING_METRIC = "cosine";
 var nonEmptyStringSchema2 = z2.string().min(1);
@@ -903,16 +905,40 @@ async function searchVisibleLexicalMemories(args) {
     )
     FROM unnest(tsvector_to_array(${queryVector})) AS query_terms(term)
   )`;
-  const textsearch = juniorMemoryMemories.searchVector;
-  const textRank = sql2`ts_rank_cd(${textsearch}, ${tsquery})`;
-  const rows = await args.db.select({
-    memory: juniorMemoryMemories,
-    textRank
-  }).from(juniorMemoryMemories).where(and(predicate, sql2`${textsearch} @@ ${tsquery}`)).orderBy(
-    desc(textRank),
+  const candidateLimit = Math.min(
+    MAX_LEXICAL_RANK_CANDIDATES,
+    args.limit * LEXICAL_RANK_OVERFETCH
+  );
+  const candidates = args.db.select().from(juniorMemoryMemories).where(
+    and(predicate, sql2`${juniorMemoryMemories.searchVector} @@ ${tsquery}`)
+  ).orderBy(
     desc(juniorMemoryMemories.observedAtMs),
     asc(juniorMemoryMemories.id)
-  ).limit(args.limit);
+  ).limit(candidateLimit).as("lexical_candidates");
+  const textRank = sql2`ts_rank_cd(${candidates.searchVector}, ${tsquery})`;
+  const rows = await args.db.select({
+    memory: {
+      archiveReason: candidates.archiveReason,
+      archivedAtMs: candidates.archivedAtMs,
+      content: candidates.content,
+      createdAtMs: candidates.createdAtMs,
+      expiresAtMs: candidates.expiresAtMs,
+      id: candidates.id,
+      idempotencyKey: candidates.idempotencyKey,
+      kind: candidates.kind,
+      observedAtMs: candidates.observedAtMs,
+      scope: candidates.scope,
+      scopeKey: candidates.scopeKey,
+      searchVector: candidates.searchVector,
+      sourceKey: candidates.sourceKey,
+      sourcePlatform: candidates.sourcePlatform,
+      subjectKey: candidates.subjectKey,
+      subjectType: candidates.subjectType,
+      supersededAtMs: candidates.supersededAtMs,
+      supersededById: candidates.supersededById
+    },
+    textRank
+  }).from(candidates).orderBy(desc(textRank), desc(candidates.observedAtMs), asc(candidates.id)).limit(args.limit);
   const ranks = denseRanks(rows, (row) => Number(row.textRank));
   return rows.map((row, index2) => ({
     lexical: { rank: ranks[index2] },