fix(web): Search performance improvements (#615)

CHANGELOG.md

@@ -7,11 +7,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Changed
+- Bumped the default requested search result count from 5k to 100k after optimization pass. [#615](https://github.com/sourcebot-dev/sourcebot/pull/615)
+
 ### Fixed
 - Fixed incorrect shutdown of PostHog SDK in the worker. [#609](https://github.com/sourcebot-dev/sourcebot/pull/609)
 - Fixed race condition in job schedulers. [#607](https://github.com/sourcebot-dev/sourcebot/pull/607)
 - Fixed connection sync jobs getting stuck in pending or in progress after restarting the worker. [#612](https://github.com/sourcebot-dev/sourcebot/pull/612)
 - Fixed issue where connections would always sync on startup, regardless if they changed or not. [#613](https://github.com/sourcebot-dev/sourcebot/pull/613)
+- Fixed performance bottleneck in search api. Result is a order of magnitutde improvement to average search time according to benchmarks. [#615](https://github.com/sourcebot-dev/sourcebot/pull/615) 
 
 ### Added
 - Added force resync buttons for connections and repositories. [#610](https://github.com/sourcebot-dev/sourcebot/pull/610)

packages/web/src/app/[domain]/search/components/searchResultsPage.tsx

@@ -35,7 +35,7 @@ import { FilterPanel } from "./filterPanel";
 import { useFilteredMatches } from "./filterPanel/useFilterMatches";
 import { SearchResultsPanel } from "./searchResultsPanel";
 
-const DEFAULT_MAX_MATCH_COUNT = 5000;
+const DEFAULT_MAX_MATCH_COUNT = 100_000;
 
 interface SearchResultsPageProps {
     searchQuery: string;

packages/web/src/app/api/(client)/client.ts

@@ -1,6 +1,5 @@
 'use client';
 
-import { getVersionResponseSchema, getReposResponseSchema } from "@/lib/schemas";
 import { ServiceError } from "@/lib/serviceError";
 import { GetVersionResponse, GetReposResponse } from "@/lib/types";
 import { isServiceError } from "@/lib/utils";
@@ -10,10 +9,6 @@ import {
     SearchRequest,
     SearchResponse,
 } from "@/features/search/types";
-import {
-    fileSourceResponseSchema,
-    searchResponseSchema,
-} from "@/features/search/schemas";
 
 export const search = async (body: SearchRequest, domain: string): Promise<SearchResponse | ServiceError> => {
     const result = await fetch("/api/search", {
@@ -29,10 +24,10 @@ export const search = async (body: SearchRequest, domain: string): Promise<Searc
         return result;
     }
 
-    return searchResponseSchema.parse(result);
+    return result as SearchResponse | ServiceError;
 }
 
-export const fetchFileSource = async (body: FileSourceRequest, domain: string): Promise<FileSourceResponse> => {
+export const fetchFileSource = async (body: FileSourceRequest, domain: string): Promise<FileSourceResponse | ServiceError> => {
     const result = await fetch("/api/source", {
         method: "POST",
         headers: {
@@ -42,7 +37,7 @@ export const fetchFileSource = async (body: FileSourceRequest, domain: string):
         body: JSON.stringify(body),
     }).then(response => response.json());
 
-    return fileSourceResponseSchema.parse(result);
+    return result as FileSourceResponse | ServiceError;
 }
 
 export const getRepos = async (): Promise<GetReposResponse> => {
@@ -53,7 +48,7 @@ export const getRepos = async (): Promise<GetReposResponse> => {
         },
     }).then(response => response.json());
 
-    return getReposResponseSchema.parse(result);
+    return result as GetReposResponse | ServiceError;
 }
 
 export const getVersion = async (): Promise<GetVersionResponse> => {
@@ -63,5 +58,5 @@ export const getVersion = async (): Promise<GetVersionResponse> => {
             "Content-Type": "application/json",
         },
     }).then(response => response.json());
-    return getVersionResponseSchema.parse(result);
+    return result as GetVersionResponse;
 }

packages/web/src/features/search/schemas.ts

@@ -141,6 +141,7 @@ export const searchResponseSchema = z.object({
     repositoryInfo: z.array(repositoryInfoSchema),
     isBranchFilteringEnabled: z.boolean(),
     isSearchExhaustive: z.boolean(),
+    __debug_timings: z.record(z.string(), z.number()).optional(),
 });
 
 export const fileSourceRequestSchema = z.object({

packages/web/src/features/search/searchApi.ts

@@ -1,16 +1,18 @@
 'use server';
 
-import { invalidZoektResponse, ServiceError } from "../../lib/serviceError";
-import { isServiceError } from "../../lib/utils";
-import { zoektFetch } from "./zoektClient";
-import { ErrorCode } from "../../lib/errorCodes";
-import { StatusCodes } from "http-status-codes";
-import { zoektSearchResponseSchema } from "./zoektSchema";
-import { SearchRequest, SearchResponse, SourceRange } from "./types";
-import { PrismaClient, Repo } from "@sourcebot/db";
 import { sew } from "@/actions";
-import { base64Decode } from "@sourcebot/shared";
 import { withOptionalAuthV2 } from "@/withAuthV2";
+import { PrismaClient, Repo } from "@sourcebot/db";
+import { base64Decode, createLogger } from "@sourcebot/shared";
+import { StatusCodes } from "http-status-codes";
+import { ErrorCode } from "../../lib/errorCodes";
+import { invalidZoektResponse, ServiceError } from "../../lib/serviceError";
+import { isServiceError, measure } from "../../lib/utils";
+import { SearchRequest, SearchResponse, SourceRange } from "./types";
+import { zoektFetch } from "./zoektClient";
+import { ZoektSearchResponse } from "./zoektSchema";
+
+const logger = createLogger("searchApi");
 
 // List of supported query prefixes in zoekt.
 // @see : https://github.com/sourcebot-dev/zoekt/blob/main/query/parse.go#L417
@@ -126,7 +128,7 @@ const getFileWebUrl = (template: string, branch: string, fileName: string): stri
     return encodeURI(url + optionalQueryParams);
 }
 
-export const search = async ({ query, matches, contextLines, whole }: SearchRequest) => sew(() =>
+export const search = async ({ query, matches, contextLines, whole }: SearchRequest): Promise<SearchResponse | ServiceError> => sew(() =>
     withOptionalAuthV2(async ({ org, prisma }) => {
         const transformedQuery = await transformZoektQuery(query, org.id, prisma);
         if (isServiceError(transformedQuery)) {
@@ -200,20 +202,22 @@ export const search = async ({ query, matches, contextLines, whole }: SearchRequ
             "X-Tenant-ID": org.id.toString()
         };
 
-        const searchResponse = await zoektFetch({
+        const { data: searchResponse, durationMs: fetchDurationMs } = await measure(
+            () => zoektFetch({
                 path: "/api/search",
                 body,
                 header,
                 method: "POST",
-        });
+            }),
+            "zoekt_fetch",
+            false
+        );
 
         if (!searchResponse.ok) {
             return invalidZoektResponse(searchResponse);
         }
 
-        const searchBody = await searchResponse.json();
-
-        const parser = zoektSearchResponseSchema.transform(async ({ Result }) => {
+        const transformZoektSearchResponse = async ({ Result }: ZoektSearchResponse) => {
             // @note (2025-05-12): in zoekt, repositories are identified by the `RepositoryID` field
             // which corresponds to the `id` in the Repo table. In order to efficiently fetch repository
             // metadata when transforming (potentially thousands) of file matches, we aggregate a unique
@@ -379,7 +383,48 @@ export const search = async ({ query, matches, contextLines, whole }: SearchRequ
                     flushReason: Result.FlushReason,
                 }
             } satisfies SearchResponse;
-        });
+        }
 
-        return parser.parseAsync(searchBody);
+        const { data: rawZoektResponse, durationMs: parseJsonDurationMs } = await measure(
+            () => searchResponse.json(),
+            "parse_json",
+            false
+        );
+
+        // @note: We do not use zod parseAsync here since in cases where the
+        // response is large (> 40MB), there can be significant performance issues.
+        const zoektResponse = rawZoektResponse as ZoektSearchResponse;
+
+        const { data: response, durationMs: transformZoektResponseDurationMs } = await measure(
+            () => transformZoektSearchResponse(zoektResponse),
+            "transform_zoekt_response",
+            false
+        );
+
+        const totalDurationMs = fetchDurationMs + parseJsonDurationMs + transformZoektResponseDurationMs;
+
+        // Debug log: timing breakdown
+        const timings = [
+            { name: "zoekt_fetch", duration: fetchDurationMs },
+            { name: "parse_json", duration: parseJsonDurationMs },
+            { name: "transform_zoekt_response", duration: transformZoektResponseDurationMs },
+        ];
+
+        logger.debug(`Search timing breakdown (query: "${query}"):`);
+        timings.forEach(({ name, duration }) => {
+            const percentage = ((duration / totalDurationMs) * 100).toFixed(1);
+            const durationStr = duration.toFixed(2).padStart(8);
+            const percentageStr = percentage.padStart(5);
+            logger.debug(`  ${name.padEnd(25)} ${durationStr}ms (${percentageStr}%)`);
+        });
+        logger.debug(`  ${"TOTAL".padEnd(25)} ${totalDurationMs.toFixed(2).padStart(8)}ms (100.0%)`);
+
+        return {
+            ...response,
+            __debug_timings: {
+                zoekt_fetch: fetchDurationMs,
+                parse_json: parseJsonDurationMs,
+                transform_zoekt_response: transformZoektResponseDurationMs,
+            }
+        } satisfies SearchResponse;
     }));

packages/web/src/features/search/zoektSchema.ts

@@ -75,6 +75,8 @@ export const zoektSearchResponseSchema = z.object({
     }),
 });
 
+export type ZoektSearchResponse = z.infer<typeof zoektSearchResponseSchema>;
+
 // @see : https://github.com/sourcebot-dev/zoekt/blob/3780e68cdb537d5a7ed2c84d9b3784f80c7c5d04/api.go#L728
 const zoektRepoStatsSchema = z.object({
     Repos: z.number(),