From 9b99cd7e985a54c374eeb30d1d420310172d5894 Mon Sep 17 00:00:00 2001
From: nanxun <nanxun@nxsir.cn>
Date: Thu, 30 Apr 2026 16:05:59 +0800
Subject: [PATCH] fix: retry transient postgres polling saves

---
 .../LiveRoomPollingBackgroundService.cs       | 31 +++++++++++++++++--
 src/LiveRecorder.WebApi/Program.cs            | 28 +++++++++++++++--
 2 files changed, 54 insertions(+), 5 deletions(-)

diff --git a/src/LiveRecorder.Infrastructure/Services/LiveRoomPollingBackgroundService.cs b/src/LiveRecorder.Infrastructure/Services/LiveRoomPollingBackgroundService.cs
index e3fb505..ac0cb3d 100644
--- a/src/LiveRecorder.Infrastructure/Services/LiveRoomPollingBackgroundService.cs
+++ b/src/LiveRecorder.Infrastructure/Services/LiveRoomPollingBackgroundService.cs
@@ -743,15 +743,40 @@ public sealed class LiveRoomPollingBackgroundService : BackgroundService, ILiveR
                 await dbContext.SaveChangesAsync(cancellationToken);
                 return;
             }
-            catch (DbUpdateException ex) when (attempt < 5 && IsTransientDatabaseException(ex))
+            catch (Exception ex) when (attempt < 5 && IsTransientDatabaseException(ex))
             {
                 await Task.Delay(TimeSpan.FromMilliseconds(300 * Math.Pow(2, attempt - 1)), cancellationToken);
             }
         }
     }
 
-    private static bool IsTransientDatabaseException(DbUpdateException exception) =>
-        exception.InnerException is NpgsqlException npgsqlException && npgsqlException.IsTransient;
+    private static bool IsTransientDatabaseException(Exception exception)
+    {
+        if (exception is DbUpdateException dbUpdateException)
+        {
+            return dbUpdateException.InnerException is not null &&
+                   IsTransientDatabaseException(dbUpdateException.InnerException);
+        }
+
+        if (exception is NpgsqlException npgsqlException && npgsqlException.IsTransient)
+        {
+            return true;
+        }
+
+        if (exception is TimeoutException or IOException)
+        {
+            return true;
+        }
+
+        if (exception is InvalidOperationException invalidOperationException &&
+            invalidOperationException.Message.Contains("transient failure", StringComparison.OrdinalIgnoreCase))
+        {
+            return true;
+        }
+
+        return exception.InnerException is not null &&
+               IsTransientDatabaseException(exception.InnerException);
+    }
 
     private static string? Truncate(string? value, int maxLength)
     {
diff --git a/src/LiveRecorder.WebApi/Program.cs b/src/LiveRecorder.WebApi/Program.cs
index 17a5780..cdcc4f0 100644
--- a/src/LiveRecorder.WebApi/Program.cs
+++ b/src/LiveRecorder.WebApi/Program.cs
@@ -23,6 +23,7 @@ using LiveRecorder.Infrastructure.Services;
 using LiveRecorder.WebApi.Middleware;
 using Microsoft.EntityFrameworkCore;
 using Microsoft.OpenApi.Models;
+using Npgsql;
 
 var builder = WebApplication.CreateBuilder(args);
 var resetRecordingData = args.Contains("--reset-recording-data", StringComparer.OrdinalIgnoreCase);
@@ -114,9 +115,32 @@ builder.Services.AddHttpClient("bilibili", client =>
 var defaultConnection = builder.Configuration.GetConnectionString("DefaultConnection")
     ?? throw new InvalidOperationException("ConnectionStrings:DefaultConnection is required.");
 
+var connectionStringBuilder = new NpgsqlConnectionStringBuilder(defaultConnection);
+if (connectionStringBuilder.Timeout <= 0)
+{
+    connectionStringBuilder.Timeout = 15;
+}
+
+if (connectionStringBuilder.CommandTimeout <= 0)
+{
+    connectionStringBuilder.CommandTimeout = 60;
+}
+
+if (connectionStringBuilder.KeepAlive <= 0)
+{
+    connectionStringBuilder.KeepAlive = 30;
+}
+
 builder.Services.AddDbContext<LiveRecorderDbContext>(options =>
-    options.UseNpgsql(defaultConnection, npgsql =>
-        npgsql.MigrationsAssembly(typeof(LiveRecorderDbContext).Assembly.FullName)));
+    options.UseNpgsql(connectionStringBuilder.ConnectionString, npgsql =>
+    {
+        npgsql.MigrationsAssembly(typeof(LiveRecorderDbContext).Assembly.FullName);
+        npgsql.CommandTimeout(connectionStringBuilder.CommandTimeout);
+        npgsql.EnableRetryOnFailure(
+            maxRetryCount: 5,
+            maxRetryDelay: TimeSpan.FromSeconds(10),
+            errorCodesToAdd: null);
+    }));
 
 builder.Services.AddScoped<IUnitOfWork>(provider => provider.GetRequiredService<LiveRecorderDbContext>());
 builder.Services.AddScoped<IAppSettingRepository, AppSettingRepository>();