Cluster-wide transactions on client failover #15700
-
|
Tested with 5.4.5. References to earlier discussions: When a document is loaded and changed, and a client fails over to a different node before the call to To replicate, you can use the code below. Place a breakpoint on line 30 ( After continuing, the exception is thrown. Should this be handled gracefully? using Raven.Client.Documents;
using Raven.Client.Documents.Session;
namespace RavenDB5Tests.Tryouts
{
public static class ClusterWideTransactions2
{
public static void ClusterWideTransactions2Test()
{
using (var store = new DocumentStore
{
Urls = new[] { "http://localhost:8081", "http://localhost:8082", "http://localhost:8083" },
Database = "ClusterWideTransactions2Test",
}.Initialize())
{
using (var session = store.OpenSession(new SessionOptions { TransactionMode = TransactionMode.ClusterWide }))
{
var existingDoc = session.Load<Doc>("doc-1");
if (existingDoc == null)
{
var doc = new Doc { Id = "doc-1", NumVal = 1 };
session.Store(doc);
session.SaveChanges();
}
}
using (var session = store.OpenSession(new SessionOptions { TransactionMode = TransactionMode.ClusterWide }))
{
var doc1 = session.Load<Doc>("doc-1");
doc1.NumVal++;
// if the node used for document loading is brought down here, the session.SaveChanges() below fails:
// Raven.Client.Exceptions.ClusterTransactionConcurrencyException:
// 'Failed to execute cluster transaction due to the following issues: Guard compare exchange value
// 'rvn-atomic/doc-1' index does not match the transaction index's 0 change vector on doc-1
// Concurrency check failed for putting the key 'rvn-atomic/doc-1'.Requested index: 0, actual index: 342'
session.SaveChanges();
}
}
}
public class Doc
{
public string Id { get; set; }
public int NumVal { get; set; }
}
}
} |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 3 replies
-
|
This is not expected, no. |
Beta Was this translation helpful? Give feedback.
-
|
I just tried running this scenario, and it is working for me. |
Beta Was this translation helpful? Give feedback.
-
|
Please find traffic captures from my scenario attached. Both fiddler and Wireshark (not from the same run). I run a three-node cluster on 5.4.100 (client version also 5.4.100) on localhost (Windows 10 machine). The client fails over the request, but the response is the following (formatting mine, Content-Length won't match): There is a difference between the first request and the failed over request (again, payload reformatted, Content-Length won't match): First request: Failed over request: The difference is that the failed over request doesn't contain the field Server configuration (ports vary between the nodes, otherwise identical): I run the code from the original post in Visual Studio. After hitting the breakpoint, I identify the node used to load the document and bring it down - I press In production, we are occasionally (rarely) hit by the exception in one of our processing applications, surprisingly in a place where very little time elapses between loading the document and storing it. If you can't reproduce this, I'm keenly interested in finding the difference between the environments so that the client would fail this over correctly for us. Traffic capture (Fiddler and Wireshark): ravendb-clusterwide-conflict.zip |
Beta Was this translation helpful? Give feedback.
-
|
(Didn't notice the reply that the bug was reproduced while typing the above message, so it's likely not that helpful, but anyway.) |
Beta Was this translation helpful? Give feedback.
-
|
The bug was reproduced. |
Beta Was this translation helpful? Give feedback.
The bug was reproduced.
You can track working progress here:
https://issues.hibernatingrhinos.com/issue/RavenDB-19887/Cluster-wide-transactions-on-client-failover