Last post, I talked about Customized Event and setting up trace for web applications which usually use pooled connection. Assuming you already have everything setup, SqlCommand component has been wrapped up to send an customized event with user or session information before an actually procedure call taking place. The trace is also setup to receive both procedure calls and customized events. To simplify the scenario, trace only capture the customized event and RPC Complete with column SPID, TextData, StartTime. To receive the trace data, you can either save them to the disk then read the trace file from the disk or directly stream the trace data.( see my post here for detail). After trace data being read, there will be always a customized event prior to an actual procedure call. When you are asked to get SQL Server response time for an particular application, for instance, give me average respose time of procedure ABC every 30 minutes during the day for that user, you can query the table includes the trace rows to locate all procedure ABC, then based on SPID and identity values to find previous record to get the user information of the procedure call, perform filter, aggregates…It will be fine if your table is small. The trace system generate more than 500MB data every hour, performing such query will be very challenging irrespective of how you design the indexes on the trace table. Now you may think to re-format the trace data to table with columns, EventID, SPID, TextData, StartTime, and UserInfo to simplify the query. But SQL Profiler cannot do it in that way, you will have to write an ETL for that.
SQL Profiler is a wonderful tool being wide used for system monitoring, performance monitoring, debugging, etc. ( you can count more than I do ;), I better shut up.) It provides tons of events triggered by SQL Server from different areas for various purposes. You are allowed to create customized event for your own by calling sp_trace_generateevent. There is already an example in BOL that you can trace get notified whenever a record is being inserted into a table. It quite useful when you try to centralize the error messages or some other stuff. Well, if your database is serving a busy website where the user management is application type user permission management, you are asked to keep track of every procedure calls with application user tied with, customized event will come to play. This was the one I have done before. Please feel free to tell me your story with customized events by replying my post.
SQL Server Tracing System (2) — 2 Ways to Receive Trace Data
There are one way to setup a trace programmatically but 2 ways to acquire and consume the trace data. To setup a trace, you need to to call trace supporting procedures to define events and columns. If you specify the location of the trace target, SQL Server will write trace data to files. By this approach, the SQL Server instance which reads the data files need to have file access permission to them. The another approach is that you don’t specify the target location. By this approach, the trace data will be “streamed” to the client directly.
Continue reading “SQL Server Tracing System (2) — 2 Ways to Receive Trace Data”
SQL Server Tracing System (1) — DMVs
SQL Server tracing system is used very often by DBAs and programmers for pinpointing issues in the databases. It’s supported by few system objects that describes itself including list of events, columns of each event might return, start and stop event, retrieve event data, status of each event sessions, etc. SQL Profiler utilizes those objects to retrieve and show the trace information on the screen. From debugging perspective, it’s enough for a SQL developer. DBAs, however, prefers to create trace programmatically and save the trace in the database or file system for alerting and later performance analysis. You can use SQL Profiler to trace the SQLs issued by anbother SQL Profiler to generate the trace procedure sequences. Alternatively, you can also browse trace related DMVs to find out all trace related metadata.