@sentry/junior-memory 0.131.0 → 0.133.0

package/dist/db/schema.d.ts

@@ -121,6 +121,27 @@ export declare const juniorMemoryMemories: import("drizzle-orm/pg-core").PgTable
             identity: undefined;
             generated: undefined;
         }, {}, {}>;
+        searchVector: import("drizzle-orm/pg-core").PgColumn<{
+            name: "search_vector";
+            tableName: "junior_memory_memories";
+            dataType: "custom";
+            columnType: "PgCustomColumn";
+            data: string;
+            driverParam: unknown;
+            notNull: false;
+            hasDefault: true;
+            isPrimaryKey: false;
+            isAutoincrement: false;
+            hasRuntimeDefault: false;
+            enumValues: undefined;
+            baseColumn: never;
+            identity: undefined;
+            generated: {
+                type: "always";
+            };
+        }, {}, {
+            pgColumnBuilderBrand: "PgCustomColumnBuilderBrand";
+        }>;
         sourcePlatform: import("drizzle-orm/pg-core").PgColumn<{
             name: "source_platform";
             tableName: "junior_memory_memories";

package/dist/index.js

@@ -29,6 +29,7 @@ import { sql } from "drizzle-orm";
 import {
   bigint,
   check,
+  customType,
   index,
   integer,
   pgTable,
@@ -75,6 +76,11 @@ var memoryRuntimeContextSchema = z.union([
 ]);
 
 // src/db/schema.ts
+var tsvector = customType({
+  dataType() {
+    return "tsvector";
+  }
+});
 var juniorMemoryMemories = pgTable(
   "junior_memory_memories",
   {
@@ -85,6 +91,9 @@ var juniorMemoryMemories = pgTable(
     subjectType: text("subject_type", { enum: MEMORY_SUBJECT_TYPES }).notNull(),
     subjectKey: text("subject_key"),
     content: text("content").notNull(),
+    searchVector: tsvector("search_vector").generatedAlwaysAs(
+      sql`to_tsvector('english', "content")`
+    ),
     sourcePlatform: text("source_platform", {
       enum: MEMORY_SOURCE_PLATFORMS
     }).notNull(),
@@ -105,7 +114,7 @@ var juniorMemoryMemories = pgTable(
     index("junior_memory_memories_expiration_idx").on(table.expiresAtMs).where(
       sql`${table.archivedAtMs} IS NULL AND ${table.expiresAtMs} IS NOT NULL`
     ),
-    index("junior_memory_memories_search_idx").using("gin", sql`to_tsvector('english', ${table.content})`).where(
+    index("junior_memory_memories_search_idx").using("gin", table.scope, table.scopeKey, table.searchVector).where(
       sql`${table.archivedAtMs} IS NULL AND ${table.supersededAtMs} IS NULL AND ${table.supersededById} IS NULL`
     ),
     uniqueIndex("junior_memory_memories_idempotency_idx").on(table.scope, table.scopeKey, table.idempotencyKey).where(
@@ -336,6 +345,8 @@ var DEFAULT_EXPIRED_ARCHIVE_LIMIT = 100;
 var PREFERENCE_ADJUDICATION_CANDIDATE_LIMIT = 10;
 var PREFERENCE_ADJUDICATION_VECTOR_LIMIT = 5;
 var VECTOR_SEARCH_OVERFETCH = 4;
+var LEXICAL_RANK_OVERFETCH = 4;
+var MAX_LEXICAL_RANK_CANDIDATES = 1e3;
 var MAX_MEMORY_CONTENT_CHARS = 4e3;
 var EMBEDDING_METRIC = "cosine";
 var nonEmptyStringSchema2 = z2.string().min(1);
@@ -388,6 +399,7 @@ var memoryRowSchema = z2.object({
   id: z2.string().min(1),
   idempotencyKey: optionalStringSchema,
   observedAtMs: z2.coerce.number(),
+  searchVector: z2.string().optional(),
   scope: z2.enum(MEMORY_SCOPES),
   scopeKey: z2.string().min(1),
   sourceKey: z2.string().min(1),
@@ -893,16 +905,40 @@ async function searchVisibleLexicalMemories(args) {
     )
     FROM unnest(tsvector_to_array(${queryVector})) AS query_terms(term)
   )`;
-  const textsearch = sql2`to_tsvector('english', ${juniorMemoryMemories.content})`;
-  const textRank = sql2`ts_rank_cd(${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] },