Friday, 4 August 2017

Troubleshooting timeout expired errors

Having .NET Applications working with SQL Server, some unforeseen timeout errors can be raised for two possible causes: bad-written SQL code (especially Ad-Hoc queries) and application issues. Under this circumstance, developers will generally blame SQL Server at first glance because they may think it is a misconfiguration related to timeout in SQL Server. In my experience troubleshooting these issues, I would say that lots of them are not mostly related to SQL Server configuration itself. Additionally, when it comes to application as the root cause, reading the SQL Server error log, you will not find any error about timeout or failed login events. What does it mean? it simply means that the application never tried to connect to SQL Server because it was still working at application level when the error was raised. Thus, the cause of timeout error is inside the application at Net SqlClient Data Provider level (see the CommandTimeout property value configured for the application connection). This 'CommandTimeout' property specifies the number of seconds that a application provider should wait for result sets before sending a timeout signal. It is well known that the default is 30 seconds, therefore, it may not be good enough due to application performance problems.
Timeout expired. The timeout period elapsed prior to completion of the operation or the server is not responding.
Timeout expired. The timeout period elapsed prior to obtaining a connection from the pool. This may have occurred because all pooled connections were in use and max pool size was reached.
In order to get it fixed, firstly, developers should check .NET code within applications. In some cases, the developers I worked with found infinite loops or slow codes which spent much time before sending SQL queries to the database engine, as a result of this, the timeout threshold was reached at application level and, obviously, the error was raised. By the way, the following error could also be related to timeout:
Unable to connect to SQL Server session database.
Having checked the application, you can increase the 'CommandTimeout' property value to 60 seconds (if it is not enough then change to 120 or 180, and also considerate using 0 (unlimited) which should be a workaround while the problem is being traced and fixed). That is all for now, let me know any remarks you may have. Thanks for reading. Stay tuned.

Tuesday, 1 August 2017

Updating so-called 'column statistics' for the whole SQL Server database

Definitely, statistics in SQL Server are vital to ensure stable database performance as a whole. Every DBA must know the importance of keeping statistics up to date, however, not everyone who is in charge of a database knows much about it, especially, those who do not have this role or are DBA by accident. So, I will just mention that statistics are objects that contain data distribution, density, selectivity, cardinality, etc. for all columns that were at least once queried or linked to indexes. This information helps the SQL optimizer make the right decision of what indexes and resources should use so as to execute and process the queries as fast as possible. Put differently, statistics allow SQL optimizer to create the best possible execution plan to run queries efficiently, otherwise, when statistics are not up to date bad execution plans may be created which will make inefficient use of resources bringing down the performance of the whole database and, even worse the whole server. So, it is a basic task to routinely update statistics at least once a month as part of your database maintenance tasks.
It is worth noting that statistics that linked to indexes (so-called 'index statistics') are full updated automatically when indexes are rebuilt. So, those statistics do not need explicit update. Nevertheless, statistics linked to columns (so-called 'column statistics') still need of an explicit update and this is what we need to do after rebuilding all the indexes. Thinking about it, today I am going to share a script to only run the update of columns statistics:

SET NOCOUNT ON     
DECLARE @schema_name varchar(max)      
DECLARE @table_name varchar(max)     
DECLARE @stat_name varchar(max)        
DECLARE @update_stat_cmd varchar(max)        
DECLARE @update_stat_msg_header   varchar(max)      
DECLARE update_stat_cursor CURSOR FOR    
          select  schema_name(o.[schema_id]),  object_name(s1.[object_id]) , s1.name   
          from (  
            select s.[object_id], s.name from sys.stats s  
            left join sys.indexes i on s.name=i.name  
            where i.name is null) s1  
          inner join sys.objects o on o.[object_id]=s1.[object_id]  
          where  o.type='U'  
          order by schema_name(o.[schema_id]),  object_name(s1.[object_id]) , s1.name   
  
       
OPEN update_stat_cursor        
FETCH NEXT FROM update_stat_cursor INTO  @schema_name, @table_name,  @stat_name         
WHILE (@@fetch_status = 0)        
  BEGIN    
       DECLARE @ini DATETIME, @fin DATETIME     
       SET @update_stat_msg_header =  '->Updating ['+ RTRIM(@schema_name)  +'].[' + RTRIM(@table_name) + ']([' +@stat_name + '])'      
       PRINT @update_stat_msg_header    
       SET @update_stat_cmd ='UPDATE STATISTICS ['+ RTRIM(@schema_name)  +'].[' + RTRIM(@table_name) + ']([' +@stat_name + '])'    
       SET @ini=GETDATE()    
       EXEC (@update_stat_cmd)      
       SET @fin=GETDATE()    
      
       FETCH NEXT FROM update_stat_cursor INTO @schema_name, @table_name,  @stat_name            
  END        
 
PRINT ' '       
PRINT '----------------------------------------------------------------------------- '        
SET @update_stat_msg_header = '*************  THERE ARE NO MORE STATISTICS TO BE UPDATED **************'         
PRINT @update_stat_msg_header        
PRINT ' '        
PRINT 'All statistics not linked to any index were rebuilt!'        
  
CLOSE update_stat_cursor        
DEALLOCATE update_stat_cursor    
SET NOCOUNT OFF

That is all for now. Thanks for reading. Let me know any remarks you may have. Stay tuned.

Tuesday, 25 July 2017

Testing database connectivity by using a Universal Data Link file

After installing a SQL instance, we may need to test the database connectivity from a client host to make sure that everything is working very well, for instance, sometimes Windows Firewall might block access to the service or there might be network issues. Moreover, in the likely event that there was no time to install SQL Server client tools such as SSMS or SQLCMD to carry out the test, you woud be a bit suprised to know that there is a simpler way to do it, that is via a Data Link file. Consequently, in this post I am going to show you how to create and use a Data Link file to test connectivity to a SQL instance. To begin with, you must open Notepad to create an empty .txt file and save it with the .udl extension as you can see in the following picture.
After doing that, you must open the .udl file and you will then see the following window with four tabs. The second tab "Connection" is to fill with the server name (or SQL instance name) and the credentials accordingly. For instance, I am testing the connectivity to a default SQL instance and using Windows Authentication. You must modify that to serve your needs.
In the first tab "Provider" we can choose the Provider to use in the test. By default, it is always "Microsoft OLE DB Provider for SQL Server". It is ideal to test other providers as well. It is worth noting that SQL Native Providers will be listed if SQL Client tools are installed locally in the client host from where the test is done. In the tab "Advanced" it is possible to set the timeout value whereas in the tab "All" we can see a summary of all the settings, and we can also edit the values of some important connection parameters such as "Language", "Connect Timeout", "Packet Size", "Data Source" and "Initial Catalog".

Finally, we just have to click on "Test Connection" in the tab "Connection" to proceed with the test. If the connectivity to the SQL instance is ok, you will see the message "Test connection succeeded", it couldn't have been simpler!
That is all for now. I hope you find this post helpful and practical. Let me know any remarks you ma have. Stay tuned.

Thursday, 20 July 2017

Detecting excessive compilation and recompilation issues

Undoubtedly, recompilation is a big topic to reckon with, more importantly, in database environments processing data that is changing rapidly over time and compounded by ad-hoc workloads which may cause CPU bottleneck, so it is of paramount importance to detect excessive compilation and recompilation issues and address them to warrant stable performance for the queries, and in this sense, there are some tools to be used to detect these issues such as Performance Monitor, Extended Events, SQL Server Profiler Trace, DMVs, etc. When it comes to using Performance Monitor we should concentrate the efforts on analising the performance counters SQL Server: SQL Statistics: Batch Requests/sec, SQL Server: SQL Statistics: SQL Compilations/sec, and SQL Server: SQL Statistics: SQL Recompilations/sec. What's more, you can save a trace file capturing the events SP:Recompile, SQL:StmtRecompile, and CursorRecompile, then you can use the following query to see all the recompilation events:

select spid, StartTime, Textdata, EventSubclass, ObjectID, DatabaseID, SQLHandle 
from fn_trace_gettable('C:\RecompilationTrace_01.trc', 1)
where EventClass in(37,75,166) -- 37 = Sp:Recompile, 75 = CursorRecompile, 166 = SQL:StmtRecompile

Furthermore, we can also capture the showplan XML for query compile, but doing that has significant performance overhead because it is captured for each compilation or recompilation. So, just do it for a very short time as long as you see a high value for the SQL Compilations/sec counter in Performance Monitor. Once you know where the problem is you can use Engine Tuning Advisor to see whether any indexing changes improve the compile time and the execution time of the query.
Talking of DMVs to diagnose recompilation issues, looking into 'sys.dm_exec_query_optimizer_info' is very helpful, in particular, look at the Elapsed Time, which is the time elapsed due to optimizations, and also Final Cost. If you see that Time Elapsed is very close to the CPU time, you might reach the conclusion that the compilation and recompilation time is attributable to that high CPU use. Another DMV to use is 'sys.dm_exec_query_stats' whose most important columns to look at are sql_handle, total worker time, plan generation number (the number of times the query has recompiled), and statement Start Offset. Here is an example to check the top 20 most procedures that have been recompiled.

select top 20 SQLText.text, sql_handle, plan_generation_num, execution_count, dbid, objectid 
from sys.dm_exec_query_stats
   cross apply sys.dm_exec_sql_text(sql_handle) as SQLText
where plan_generation_num >1
order by plan_generation_num desc

There is a plethora of recommendations to deal with recompilation and keep the performance in optimum conditions, but for now you can take into consideration the following options:
  • Check to see whether the stored procedure was created with the WITH RECOMPILE option or whether the RECOMPILE query hint was used. If a procedure was created with the WITH RECOMPILE option, since SQL Server 2005, you may be able to take advantage of a statement-level RECOMPILE hint if a particular statement within that procedure needs to be recompiled. Using this hint at the statement level avoids the need of recompiling the whole procedure each time it executes, while at the same time allowing the individual statement to be compiled. 
  • Recompilations can occur due to changes in statistics, and you can use the KEEPFIXED PLAN query hint to make recompilations occur only when there is the need to ensure correctness and not to respond to changes in statistics. For instance, in this context recompilation can only occur if the underlying table structure or its schema that is referenced by a statement changes, or if a table is marked with the sp_recompile stored procedure, all resulting in the fact that the plan no longer applies and consequently triggering the recompilation event. 
  • Using the KEEP PLAN query hint is useful to set the recompilation threshold of temporary tables to be the same as permanent tables. Take a look at the EventSubclass column which displays 'Statistics Changed' for an operation on a temporary table.
  • Turning off the automatic updates of statistics for indexes and statistics that are defined on a table or indexed view prevents recompilations that are due to statistics changes on that object. It is worth noting that turning off the auto-stats option is not always a good idea. This is because the query optimizer is no longer sensitive to data changes in those objects resulting in suboptimal query plans. To be honest, I never turned off this option because I always preferred trusting in SQL Server criteria and instead opting to work on optimising queries.
  • Keep in mind that recompilation thresholds for temporary tables are lower than for normal tables, so if the recompilations on a temporary table are due to statistics changes, you can change the temporary tables to table variables. A change in the cardinality of a table variable does not cause a recompilation. The side effect of this approach is that the query optimizer does not keep track of a table variable's cardinality because statistics are not created or maintained on table variables. This can result in less optimal query plans, however, you can test the different options and choose the best one. Generally, temporary tables provide much better performance than tables variables when lots of data is involved.
  • Recompilation might also occur as a result from SET option changes, so one can diagnose it by using SQL Server Profiler to determine which SET option changed. It is highly advisable to avoid changing SET options within stored procedures and it is much better to set them at the connection level, and for the overwhelmingly majority of cases the default SET options work well. It is also very important to ensure that SET options are not changed during the lifetime of the connection.
  • To avoid recompilations that are due to deferred compilations, do not combine DML with DDL and do not create the DDL as a result from conditional IF statements.
  • To avoid recompilation and also to avoid ambiguity between objects, batches should have qualified object names, for example, dbo.Table1, User1.MySP, etc.
That is all for now, let me know any remarks you may have.
HELLO, I'M PERCY REYES! I've been working as a senior SQL Server Database Engineer for over 20 years; I'm a three-time Microsoft Data Platform MVP. I'm a cryptographer conducting research on cryptographic Boolean functions and their applications.