MongoServer lifecycle

[saille]

This is half question, half bug report.

I believe there might be a significant bug in MongoRepository's management of the lifecycle of MongoServer. I believe MongoServer should be singleton/static, and the same instance should be used every time. Otherwise there's some overhead in connecting to the database with each instantiation of the repository.

Here in Util.cs:
private static MongoDatabase GetDatabaseFromUrl(MongoUrl url)
{
var client = new MongoClient(url);
var server = client.GetServer(); // <---- should not be returning a new MongoServer instance!
return server.GetDatabase(url.DatabaseName); // WriteConcern defaulted to Acknowledged
}

See also http://stackoverflow.com/questions/10241641/what-is-the-right-way-to-manage-mongodb-connections-in-asp-net-mvc

I picked this up when looking at performance of queries - was seeing 200ms+ overhead with each new MongoRepository.

Apologies if I have misunderstood the code.

[RobThree]

See MongoClient Connection Pooling. What goes there essentially also goes for MongoRepository since MongoRepository is merely a layer on top of the C# Driver.

The driver will, for the most part, use the same connection settings. [...] Hence, the best practice will be to only instantiate 1 MongoClient at app startup and use that. Put it in an IoC container, create a static variable somewhere, etc... Essentially, make it a singleton. Is there any reason you are constructing a new client for every request?

[saille]

I'm just instantiating MongoRepository, with a connection string argument. MongoRepository is calling Util's GetDatabaseFromUrl() on my behalf. Callstack is:
MongoRepository(connString)
Util.GetCollectionFromConnectionString(connectionString)
Util.GetCollectionFromConnectionString(connectionString, GetCollectionName())
Util.GetDatabaseFromUrl(new MongoUrl(connectionString))
.GetCollection(collectionName)

from where it does this:
var client = new MongoClient(url);
var server = client.GetServer();
return server.GetDatabase(url.DatabaseName); // WriteConcern defaulted to Acknowledged

[RobThree]

I understand; but why not keep the repository alive (e.g. keep a reference around instead of instantiating a new one every time)?

[saille]

Thanks Rob. I did structure things that way for a while. I would prefer to see the dependency on a long lived MongoServer/MongoDatabase object, or a long lived MongoRepository instance made very explicit. The examples need to show that pattern of use - bear in mind many people don't use DI containers, and even if they do, they may inject a MongoRepository instance per request without thinking about it.

I managed to figure this out myself, but I know there's a lot of Mongo / .NET noobs who will fall into the trap of simply new'ing up a MongoRepository every time they need it. For your typical business application they may never notice the latency they have incurred.

[RobThree]

I am having trouble reproducing your findings; I have created a very simple test-project that

• Creates a new repository
• Adds a document/entity

And we do that in 10 "runs" of 1000 loops:

using MongoRepository;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;

namespace ConsoleApplication1
{
    class Program
    {
        static void Main(string[] args)
        {
            for (int runs = 0; runs < 10; runs++)
            {

                var results = new List<TimeSpan>();

                for (int i = 0; i < 1000; i++)
                {
                    var s = Stopwatch.StartNew();
                    var repo = new MongoRepository<Customer>();
                    repo.Add(new Customer { Name = Guid.NewGuid().ToString("N") });
                    results.Add(s.Elapsed);
                }

                Trace.WriteLine(string.Format("Average: {0}", TimeSpan.FromTicks((long)results.Average(r => r.Ticks))));
                Trace.WriteLine(string.Format("Total  : {0}", TimeSpan.FromTicks((long)results.Sum(r => r.Ticks))));
            }
        }
    }

    class Customer : Entity
    {
        public string Name { get; set; }
    }
}

My output is:

Average: 00:00:00.0003750
Total  : 00:00:00.3750495
Average: 00:00:00.0003583
Total  : 00:00:00.3583765
Average: 00:00:00.0001761
Total  : 00:00:00.1761708
Average: 00:00:00.0001808
Total  : 00:00:00.1808039
Average: 00:00:00.0001764
Total  : 00:00:00.1764045
Average: 00:00:00.0001751
Total  : 00:00:00.1751791
Average: 00:00:00.0001778
Total  : 00:00:00.1778107
Average: 00:00:00.0001794
Total  : 00:00:00.1794380
Average: 00:00:00.0001788
Total  : 00:00:00.1788072
Average: 00:00:00.0001770
Total  : 00:00:00.1770513

The above results are for localhost; The results below are for another host on the LAN:

Average: 00:00:00.0006785
Total  : 00:00:00.6785227
Average: 00:00:00.0005014
Total  : 00:00:00.5014012
Average: 00:00:00.0004766
Total  : 00:00:00.4766673
Average: 00:00:00.0004823
Total  : 00:00:00.4823463
Average: 00:00:00.0004819
Total  : 00:00:00.4819672
Average: 00:00:00.0004909
Total  : 00:00:00.4909650
Average: 00:00:00.0004643
Total  : 00:00:00.4643221
Average: 00:00:00.0005465
Total  : 00:00:00.5465504
Average: 00:00:00.0004765
Total  : 00:00:00.4765501
Average: 00:00:00.0004820
Total  : 00:00:00.4820312

As you can see it's slower but that is to be expected since we actually have to go through/over the network. The results are pretty similar when in new up the repo outside of the inner for-loop:

var repo = new MongoRepository<Customer>();
for (int i = 0; i < 1000; i++)
{
    var s = Stopwatch.StartNew();
    repo.Add(new Customer { Name = Guid.NewGuid().ToString("N") });
    results.Add(s.Elapsed);
}

Results:

Average: 00:00:00.0006715
Total  : 00:00:00.6715329
Average: 00:00:00.0005192
Total  : 00:00:00.5192991
Average: 00:00:00.0003974
Total  : 00:00:00.3974282
Average: 00:00:00.0005721
Total  : 00:00:00.5721572
Average: 00:00:00.0005333
Total  : 00:00:00.5333527
Average: 00:00:00.0004177
Total  : 00:00:00.4177798
Average: 00:00:00.0003809
Total  : 00:00:00.3809415
Average: 00:00:00.0004902
Total  : 00:00:00.4902817
Average: 00:00:00.0004598
Total  : 00:00:00.4598998
Average: 00:00:00.0004196
Total  : 00:00:00.4196392

There's much to say about this test-case (as it doesn't take "warmup" into account, has a abysmal way of measuring etc. etc.) but as you can see my results are very different than from what you are seeing.

Are you sure you're not doing very expensive/slow DNS lookups? Or connecting to a high-latency host? Maybe something other network related?