Should I give up on Integer Ids for RavenDB? - c#-4.0

Question

Update: Here's a gist that more fully demonstrates the issue https://gist.github.com/pauldambra/5051550
Ah, more update... If I make the Id property on the Mailing class a string then it all works. Should I just give up on integer ids?
I have 2 models
public class Mailing
{
public int Id { get; set; }
public string Sender { get; set; }
public string Subject { get; set; }
public DateTime Created { get; set; }
}
public class Recipient
{
public Recipient()
{
Status = RecipientStatus.Pending;
}
public RecipientStatus Status { get; set; }
public int MailingId { get; set; }
}
On my home page I want to grab the last 10 mailings. With a count of their recipients (eventually with a count of different status recipients but...)
I have made the following index
public class MailingWithRecipientCount : AbstractMultiMapIndexCreationTask<MailingWithRecipientCount.Result>
{
public class Result
{
public int MailingId { get; set; }
public string MailingSubject { get; set; }
public string MailingSender { get; set; }
public int RecipientCount { get; set; }
}
public MailingWithRecipientCount()
{
AddMap<Mailing>(mailings => from mailing in mailings
select new
{
MailingId = mailing.Id,
MailingSender = mailing.Sender,
MailingSubject = mailing.Subject,
RecipientCount = 0
});
AddMap<Recipient>(recipients => from recipient in recipients
select new
{
recipient.MailingId,
MailingSender = (string) null,
MailingSubject = (string)null,
RecipientCount = 1
});
Reduce = results => from result in results
group result by result.MailingId
into g
select new
{
MailingId = g.Key,
MailingSender = g.Select(m => m.MailingSender)
.FirstOrDefault(m => m != null),
MailingSubject = g.Select(m => m.MailingSubject)
.FirstOrDefault(m => m != null),
RecipientCount = g.Sum(r => r.RecipientCount)
};
}
}
I query using
public ActionResult Index()
{
return View(RavenSession
.Query<RavenIndexes.MailingWithRecipientCount.Result, RavenIndexes.MailingWithRecipientCount>()
.OrderByDescending(m => m.MailingId)
.Take(10)
.ToList());
}
And I get:
System.FormatException: System.FormatException : Input string was not
in a correct format. at System.Number.StringToNumber(String str,
NumberStyles options, NumberBuffer& number, NumberFormatInfo info,
Boolean parseDecimal)
Any help appreciated

Answer 1

Yes, integer ids are a pain. This is mainly because Raven always stores a full string document key, and you have to think about when you are using the key or your own id and translate appropriately. When reducing, you also need to align the int and string data types.
The minimum to get your test to pass is:
// in the "mailings" map
MailingId = mailing.Id.ToString().Split('/')[1],
// in the reduce
MailingId = g.Key.ToString(),
However - you could make your index a whole lot smaller and perform better by taking the sender and subject strings out of it. You can just put them in with a transform.
Here is a simplified complete index that does the same thing.
public class MailingWithRecipientCount : AbstractIndexCreationTask<Recipient, MailingWithRecipientCount.Result>
{
public class Result
{
public int MailingId { get; set; }
public string MailingSubject { get; set; }
public string MailingSender { get; set; }
public int RecipientCount { get; set; }
}
public MailingWithRecipientCount()
{
Map = recipients => from recipient in recipients
select new
{
recipient.MailingId,
RecipientCount = 1
};
Reduce = results => from result in results
group result by result.MailingId
into g
select new
{
MailingId = g.Key,
RecipientCount = g.Sum(r => r.RecipientCount)
};
TransformResults = (database, results) =>
from result in results
let mailing = database.Load<Mailing>("mailings/" + result.MailingId)
select new
{
result.MailingId,
MailingSubject = mailing.Subject,
MailingSender = mailing.Sender,
result.RecipientCount
};
}
}
As an aside, did you know about the RavenDB.Tests.Helpers package? It provides a simple base class RavenTestBase that you can inherit from that does most all of the legwork for you.
using (var store = NewDocumentStore())
{
// now you have an initialized, in-memory, embedded document store.
}
Also - you probably shouldn't scan the assembly for indexes in a unit test. You might introduce indexes that weren't part of what you were testing. The better route is to create the index indvidually, like this:
documentStore.ExecuteIndex(new MailingWithRecipientCount());