Tracing and Profiling SQL Server potx

 The rowset provider is not designed for data loss guarantee  If data is not being consumed quickly enough and its internal buffers fill it waits up to 20 seconds before it begins dr

ISQL Users Group May 2009

Tracing and Profiling

SQL Server

Yaniv.Etrogi@gmail.com

http://blogs.microsoft.co.il/blogs/yaniv_etrogi

Over the next 1 hour:

 SQL Trace Architecture and

Terminology

 Security and Permissions

 Profiler

 Server side Tracing

 Saving and Replaying Traces

 Querying Server-Side Trace Metadata

Introduction

troubleshooting are all possible by

the ability to view what's going on

inside SQL Server

view into what’s going on inside the database engine and at a very

granular level

SQL Trace Architecture and Terminology

 SQL Trace is an SQL Server database engine

technology

 SQL Server Profiler is a NET application that

uses system stored procedures exposing the

functionality of SQL Trace

Microsoft.SqlServer.Management.Trace

http://msdn.microsoft.com/en-us/library/microsoft.sqlserver.management.t race.aspx

events which are generated when

various actions occur in the database engine

– For example, a user login or the execution of a query are each actions that cause events to fire

collection of columns that contain the collected data when the event fires

selectively enables data collection as

required

– Data is never collected until someone asks for it

columns to select from

to the final destination

trace data are either a file on the

database server (or a network

share) or a rowset to a client

to ensure that if the data is not

consumed quickly enough (written

to disk or read from the rowset) it

will be queued

guarantee that no event data will be lost

writes are not occurring quickly

enough the internal buffers begin to fill

the SQLTRACE_LOCK and

IO_COMPLETION wait types

 The rowset provider is not designed for data loss guarantee

 If data is not being consumed quickly

enough and its internal buffers fill it waits

up to 20 seconds before it begins dropping events in order to free buffers for the sake

of getting things moving

 SQL Server Profiler will then send an

information message informing that some events had been lost and had not been

captured

 You can find out if you're headed in that direction by monitoring SQL Server's

TRACEWRITE wait type which is

incremented as threads are waiting for

buffers to free up

thread is started whenever at least one trace is active on the server

– SELECT * FROM sysprocesses

WHERE cmd = N'TRACE QUEUE TASK' ;

responsible for:

– Flushing file provider buffers (done every 4 seconds)

– Can be seen as a wait_type of

SQLTRACE_BUFFER_FLUSH

– Closing rowset-based traces that are considered to be expired (dropping events for more than 10 minutes)

trace data directly to a table

Security and Permissions

permission called ALTER TRACE This is

a server-level permission granted to a login principal and allows access to

start, stop, or modify a trace (in

addition to being able to generate defined events)

<LoginName>;

security features to keep passwords

secured

–SQL Trace will automatically omit data if an event

contains a call to a password-related stored procedure or statement For example, a call to CREATE LOGIN including the WITH PASSWORD option will be blanked out by SQL Trace

–SQL Trace will not return statement text or query plans

generated within an encrypted stored procedure,

user-defined function, or view

Profiler

 The General tab allows you to control how the trace will be processed by the consumer

 The default setting is to use the rowset provider

displaying the events in real time in the SQL Server

Profiler tool window

 Other available options are

 When saving to a table Profiler uses the rowset

provider and routes the data back into a table

 When saving to a server side file Profiler actually

starts two equivalent traces—one using the rowset

provider and the other using the file provider

 Saving to a client-side file does not use the file

provider at all Rather, the data is routed to the Profiler tool via the rowset provider and then saved from there

to a file.

offered by the file provider

 The Events Selection tab allows you to select events that you'd like to trace along with their associated data columns

 In order to narrow your scope and help ensure that tracing does not cause performance issues SQL Trace offers the ability to filter the events based on various criteria

 In SQL Server Profiler filtration is exposed via the

“Column Filters” tab

 Remember to check the “Show All Columns”

checkbox in order to see the complete list of columns

 Once you click the Run button data will begin

streaming immediately and be displayed at the Profiler window (this is because Profiler uses the rowset

provider)

it you may disable scrolling using the Auto Scroll icon on the SQL Server Profiler toolbar

filtered you may be in a situation that the icons in the Profiler tool bar are seen as if they were inactive thus disallowing you to stop the trace The only way out here is to stop the trace using sp_trace_setstatus

Saving and Replaying

Traces

 Profiler ships with eight predefined templates

settings that you can save to create a reusable trace definitions

 TSQL_Replay template selects a variety of columns for 15 events that are all required for Profiler to be able to play back (Replay) a collected trace at a later time

experienced on a production system by replaying events collected on the

production system at a test environment

 Collected trace data has to be saved and then reopened

before a replay can begin

 SQL Server Profiler offers the following options for saving trace data available from the File menu:

proprietary binary format This is generally the fastest way to save the data, and also the smallest in terms of bytes on disk

table in a database This option is useful if you need to manipulate or report on the data using T-SQL

only those events and columns needed for replay functionality are saved

 Once the data has been saved to a file or table the

original trace window can be closed and the file or

table reopened via SQL Server Profiler's File menu

 A trace reopened in this way will have a Replay menu

on the Profiler tool bar allowing you to start replaying the trace, stop the replay, or set a breakpoint

 During the course of the replay, the same events

used to produce the trace being replayed will be traced from the server on which you replay

client-side save No server-client-side option exists for saving playback results

 The trace will be replayed on multiple threads (2005 only), corresponding to at most the “Number Of Replay Threads” specified

that all events will be played back in exactly the order in which they

occurred as based upon the EventSequence column Multiple threads will still be used to simulate multiple spids

Profiler to reorder the order in which each spid starts to execute events,

in order to enhance playback performance However, the order of events will remain consistent with the EventSequence within a given spid

Server-Side Tracing

add/remove event/column combinations

to traces based on the TraceId

 sp_trace_setfilter is used to define

event filters based on trace columns

 sp_trace_setstatus is called to start,

stop and and delete a trace

–Traces can be started and stopped multiple times over

their lifespan

Querying Server-Side Trace

Metadata

SELECT *

FROM sys.traces WHERE [status] = 1;

This query returns the trace status, which will

be 1 (started) or 0 (stopped); the server-side path to the trace file (or NULL if the trace is using the rowset provider); the maximum file size (or again, NULL in the case of the rowset provider); information about how many

buffers of what size are in use for processing the I/O; the number of events captured; and the number of dropped events (NULL if your trace is using the file provider(loseless

guarantee) ).

SELECT

e.name AS [event], c.name AS [column]

FROM

fn_trace_geteventinfo(@TraceId) ei

INNER JOIN sys.trace_events e

ON ei.eventid = e.trace_event_id INNER JOIN sys.trace_columns c

ON ei.columnid = c.trace_column_id;

SELECT

columnid ,c.name AS [column]

,logical_operator ,comparison_operator ,value

FROM fn_trace_getfilterinfo(TraceId ) ei

INNER JOIN sys.trace_columns c

ON ei.columnid = c.trace_column_id;

ON c.trace_column_id =

b.trace_column_id

ORDER BY e.name;

INNER JOIN sys.trace_subclass_values s

ON c.trace_column_id =

s.trace_column_id

INNER JOIN sys.trace_events e

ON e.trace_event_id = s.trace_event_id WHERE e.name LIKE 'Audit Login';`

 Reading trace files

SELECT * FROM

fn_trace_gettable(‘B:\Traces\TraceErrors.trc', 1);

the number of trace files to read If the second parameter is DEFAULT then all available trace files will be read

SELECT * INTO ProfilerErrors

FROM

fn_trace_gettable(‘B:\Traces\TraceErrors.trc', DEFAULT);

from the fact that the SELCT INTO command is a BULK operation where as just performing queries against the trace loads sql server’s memory and if it

is a large set of data spilling begins thus incurring additional I/O

SELECT IDENTITY(int, 1, 1) AS RowNumber, * INTO dbo.ProfilerErrors

FROM

fn_trace_gettable(‘B:\Traces\TraceErrors.trc', DEFAULT);

 Controlling trace state

EXEC sp_trace_setstatus @TraceId, 1; @TraceId, 1; start start

EXEC sp_trace_setstatus @TraceId, 0; @TraceId, 0; stop stop

EXEC sp_trace_setstatus @TraceId, 2; @TraceId, 2; –remove –remove

 A trace in a stopped state can be started without the need to recreate the trace This allows to

modify the Events/Columns selection and the

selected Filters

 A restart of the SQL Server service removes all trace definitions

–Zeros the EventSequence column

–If you need your trace always up and running you can launch it from a start up stored procedure

?

?

Inside Microsoft SQL Server 2005

http://www.insidesqlserver.com/thebooks.html

Trace Errors

http://blogs.microsoft.co.il/blogs/yaniv_etrogi/TraceErrors.zip