Database

13-Jan-15 19:07

Hi There,

Recently I faced an issue regarding deadlock. In the deadlock graph I found the isolation level of the Victim and participant Procedures as serializable. The thing is that the isolation level of the corresponding database is Default Read Committed. How the isolation level changes and will it be one of the reason for the deadlock? Kindly advice me on the same

Regards
Mahesh
"Fall Down To Rise Up!!"

Eddy Vluggen13-Jan-15 23:44

13-Jan-15 23:44

Documentation here[^].

Member 11373638 wrote:
How the isolation level changes and will it be one of the reason for the deadlock?

A deadlock occurs when a transaction is waiting for another transaction to finish, where the latter is also waiting on the first. The victim is determined by the priority.

First part is identifying the queries that cause the lock in the first place.

Bastard Programmer from Hell Suspicious | :suss:

If you can't read my code, try converting it here[^]

Member 1137363814-Jan-15 0:30

14-Jan-15 0:30

Eddy,

I did all these findings and altered the procedures. But still I am getting the deadlock. I am just struck with this Serializable isolation. Will it be also a reason for the locks?

Regards
-Mahesh

Eddy Vluggen14-Jan-15 2:56

14-Jan-15 2:56

Member 11373638 wrote:
I did all these findings

Which findings?

Member 11373638 wrote:
and altered the procedures.

What did you alter?

Member 11373638 wrote:
I am just struck with this Serializable isolation.

There is two queries that keep waiting on each other. Once you know which ones, you can check whether they use the default isolation. I don't see a reason why this should be "serialized" and not the default value.

Depending on WHAT the queries do, you can define an alternative. If it is a read, then you might opt for a dirty read - or to wait until the other query is finished. If it are two writes, then you'd either need a queue, or a way to specify whom the victim will be. It will never be possible to update a single field using two queries simultaneous, so there will always be a victim unless the logic changes.

TL;DR yes, but changing the level will not automatically fix it.

Bastard Programmer from Hell Suspicious | :suss:

If you can't read my code, try converting it here[^]

Member 1137363814-Jan-15 16:57

14-Jan-15 16:57

Eddy,

Yeah your correct, in my case one query was doing a select and other doing a update on the same record. I changed it and thought it will work fine, but i was wrong and this time also deadlock came. And good document related to deadlocks in case of Update and Select in a same table??

One more thins as per your advice changing the isolation level wont help rite!!

Regards
-Mahesh

Eddy Vluggen15-Jan-15 9:56

15-Jan-15 9:56

Member 11373638 wrote:
in my case one query was doing a select and other doing a update on the same
record.

That's good; I was afraid you'd be doing two insert/update queries, launching them after another, each in a separate thread. For a read, there's multiple ways to handle the situation, but the server will need to be told what to do.

There are isolation levels and query-hints to get the behaviour that you'd want. It'd be a huge coincident if there is always "someone" requesting a record at the moment that the PC is writing it. It is fast in writing a record (and updating the indexes, starting triggers and.. it still takes time)

So, best is to identify why someone is always reading after that update. One cause could be that the piece of code that is updating the record, is also fetching it, immediately, simply to refresh the data on the form. In that case it should block/delay the execution until the update-command returns. Even better; don't do the select at all, but refresh using the values that you already have.

Is it the same PC/terminal that executes the select after the initial update? You could verify quickly (assuming that the problem is repeatable, is it?) using a trace[^].

Member 11373638 wrote:
And good document related to deadlocks in case of Update and Select in a same
table??

I favor MSDN for documentation; besides the isolation levels, you'd also need to see the queries that get executed to determine the current and the desired behaviour. The might already contain or need additional hints[^]. The existing query might already contain a bad idea[^]. And there's a page in the ADO.NET section that may be interesting, even though we are only looking at the database. It's worth the read[^]; it mentions a third possibility - if both statements actually belong together, then they should be wrapped in a transaction. (Could be the case if there's a calculated field in there)

That's a lot of documentation to wade through and probably a lot of code to check. Best idea is to start with the bad idea link Smile | :)

Bastard Programmer from Hell Suspicious | :suss:

If you can't read my code, try converting it here[^]

Member 1137363815-Jan-15 17:15

15-Jan-15 17:15

Eddy,

Thanks for your useful information. Both the queries are executed through the same terminal. As per your suggestion I will try to use the hints and lets see whether it really helpful. Poke tongue | ;-P

Regards
-Mahesh

Eddy Vluggen16-Jan-15 7:58

Audit Tables for Many-Many Relationships

16-Jan-15 7:58

You're welcome Smile | :)

Out of curiousity, was it caused by the bad idea?

Bastard Programmer from Hell Suspicious | :suss:

If you can't read my code, try converting it here[^]

Matt U.13-Jan-15 3:40

Matt U.

13-Jan-15 3:40

We're using Entity Framework 4, ASP.NET MVC 4 and SQL Server 2008 R2. We have some join tables for several entities, for many-many relationships. I'm working to create audit tables/data for each of these relationships/join tables. So far in my research, I've found that triggers are common. We use Windows Authentication, and we have a User table that stores authorized domain users. For our audit tables, we'd like to know which user modified the relationship.

The problem I have is, how do we get the ID from our User table based on the user who updated a relationship using our MVC application? I've looked at some options, and the only thing I can come up with so far, in a trigger, is this:

SQL

SELECT @username = nt_username FROM sys.sysprocesses WHERE spid = @@SPID

That will retrieve my username without the domain, which is what I want. I can then compare it to our User table and get the ID where the username matches. But is this the best way to go about it?

djj55: Nice but may have a permission problem
Pete O'Hanlon: He has my permission to run it.

Eddy Vluggen13-Jan-15 7:49

13-Jan-15 7:49

Matt U. wrote:
But is this the best way to go about it?

Compared to which other way?

I like triggers; it'll do what it is supposed to do without having to do anything special in the calling code.

The nt_username[^] field from the current process will always be the Windows user name, so that would look correctly too. IIRC then the variable "SYSTEM_USER" should return the same.

There's two or three alternative approaches that come to mind, each of them not as good as triggers.

Bastard Programmer from Hell Suspicious | :suss:

If you can't read my code, try converting it here[^]

Matt U.13-Jan-15 10:00

Matt U.

13-Jan-15 10:00

Thanks for the info! I've been working to put it together using triggers and I can't imagine any way that would be simpler. I think I'll just go that route. I only asked because I'm not entirely familiar with triggers, setting up audit tables, and so on.

djj55: Nice but may have a permission problem
Pete O'Hanlon: He has my permission to run it.

Eddy Vluggen13-Jan-15 12:49

13-Jan-15 12:49

You're welcome Smile | :)