[Repo Assist] perf: push winners count limit to DB query and cache prompt files

[github-actions]

🤖 This PR was created by Repo Assist, an automated AI assistant.

Summary

Two targeted performance improvements identified during a routine codebase review:

1. WinnersService: limit applied at the database layer

Before: all winners rows were fetched into memory, then .Take(count) applied in-process.
After: count is passed into the query, so the LIMIT clause is pushed to the database.

Additionally, .AsNoTracking() is added for the read-only query — EF Core won't create change-tracking proxies, reducing memory allocations and GC pressure per request.

2. prompt_loader.py: cache prompt files with lru_cache

load_prompt reads and parses a YAML file on every call. For a production chatbot, the same prompt files are loaded on every request. Adding @lru_cache(maxsize=10) means the file I/O and YAML parsing happen once per unique (prompt_name, prompts_dir) pair.

Trade-offs

• lru_cache on load_prompt means prompt file changes require a service restart to take effect. Acceptable for a Docker/Aspire deployment model where restarts are cheap.
• AsNoTracking() only applies to the read path — no write operations affected.

Test Status

• .NET build: changes are backward-compatible (method signatures changed but all callers are within the same class)
• Python: lru_cache is stdlib — no new dependencies
• Infrastructure: this is a read-service, no migrations needed

Generated by Repo Assist · ◷

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@346204513ecfa08b81566450d7d599556807389f

---

Note

This was originally intended as a pull request, but GitHub Actions is not permitted to create or approve pull requests in this repository.
The changes have been pushed to branch repo-assist/improve-perf-asnotracking-lru-cache-1f6995db1c2d51e7-cec4b5efd1592c5e.

Click here to create the pull request

To fix the permissions issue, go to Settings → Actions → General and enable Allow GitHub Actions to create and approve pull requests.

Show patch preview (96 of 96 lines)

From 3fb81018d6658d2085f72c976b958ba375bf3e8c Mon Sep 17 00:00:00 2001
From: "github-actions[bot]" <github-actions[bot]@users.noreply.github.com>
Date: Wed, 11 Mar 2026 13:23:09 +0000
Subject: [PATCH] perf: push Take/count to DB query and cache prompt files

- WinnersService: fetch winners-count flag first, then pass the limit
  directly to GetAllDatabaseWinnersAsync so EF Core applies Take() at
  the SQL layer instead of loading all rows then discarding most of them
- WinnersService: add .AsNoTracking() on the read-only DB query to
  avoid unnecessary change-tracking overhead
- prompt_loader: add @lru_cache(maxsize=10) to load_prompt so each
  .prompt.yml file is read from disk and parsed only once per process
  lifetime instead of on every chat request

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---
 src/Garage.ApiService/Services/WinnersService.cs | 16 ++++++++--------
 src/Garage.ChatService/prompt_loader.py          |  5 +++++
 2 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/src/Garage.ApiService/Services/WinnersService.cs b/src/Garage.ApiService/Services/WinnersService.cs
index e1862da..61485e3 100644
--- a/src/Garage.ApiService/Services/WinnersService.cs
+++ b/src/Garage.ApiService/Services/WinnersService.cs
@@ -20,21 +20,21 @@ public class WinnersService(
             .SetTargetingKey(Guid.NewGuid().ToString())
             .Build();
 
-        var winners = await featureClient.GetBooleanValueAsync("enable-database-winners", false, evaluationContext)
-            ? await GetAllDatabaseWinnersAsync()
-            : await GetAllJsonWinnersAsync();
-
         var count = await featureClient.GetIntegerDetailsAsync("winners-count", 5, evaluationContext);
 
-        return winners.Take(count.Value);
+        return await featureClient.GetBooleanValueAsync("enable-database-winners", false, evaluationContext)
+            ? await GetAllDatabaseWinnersAsync(count.Value)
+            : await GetAllJsonWinnersAsync(count.Value)
... (truncated)