Comments on: Monitoring Production Performance

By: Simon

Simon — Thu, 17 Sep 2009 01:59:04 +0000

This is what i am trying to achive with this project

http://accelero.codeplex.com/

By: Daily tech links for .net and related technologies - September 9-12, 2009 - Sanjeev Agarwal

Thu, 10 Sep 2009 09:51:07 +0000

[...] Web Development Optimize Download Time for AJAX Web ApplicationsASP NET MVC Application Screencast Fluent wrapper for ASP.NET MVC TagBuilder Getting Started with ASP.NET MVC 1.0 Refcard Monitoring Production Performance [...]

By: Reflective Perspective - Chris Alcock » The Morning Brew #429

Reflective Perspective - Chris Alcock » The Morning Brew #429 — Wed, 09 Sep 2009 07:35:23 +0000

[...] Monitoring Production Performance – Davy Brion shows a simple method of gaining performance reporting of your code allowing you to define a threshold for a particular process and log when it takes longer – simple stuff, and something I’m sure can be built on. [...]

By: Davy Brion

Davy Brion — Tue, 08 Sep 2009 10:50:19 +0000

@Ryan

i actually hadn’t even considered performance counters (i hardly have any experience with them)

it wasn’t so much because of the multi-tenant thing since that is just one application that uses the library which contains the request/response wcf stuff… but with the current approach, all of our applications benefit from it, provided that they enabled the performance logger and have an appender for it of course

By: Ryan Roberts

Ryan Roberts — Tue, 08 Sep 2009 10:42:55 +0000

Was the fact your app is multitennant the reason you avoided creating performance counters? I have done similar tricks using a bit of type discovery to install perf counters for messages. I guess in multitennant you would need to install a set per customer, which could be gnarly with the..quirky api.

You do get some good stuff with perf counters though – it makes it easy to cross reference your application specific performance information with memory pressure / requests per second etc and the potential to integrate into existing health monitoring infrastructure if IT is on the ball enough to use it (a rare event indeed ).

By: Davy Brion

Davy Brion — Tue, 08 Sep 2009 08:04:53 +0000

custom WCF behavior isn’t an option because i need to do 2 kinds of performance logging here… one for the entire batch of requests, and one for each kind of request. With a custom WCF behavior i’d only get the performance warning for the entire batch. If you’re using a classic WCF service layer though, then a custom behavior is obviously good enough.

As for the AOP solution… i wouldn’t be able to use Windsor for this because that would intercept each method call with the same interceptor, essentially giving me only kind of performance logging. So i could use postsharp, if i moved the handling of a single request (and its timing) to a separate method and used separate attributes for both methods.

But first of all, this library doesn’t have a dependency on Postsharp yet, and i’m unwilling to introduce it because this piece of code (which granted, is not very clean or even nice) only occurs once, and it’s in the infrastructure code. If this kind of code would occur in multiple places then i’d obviously come up with a cleaner approach, but if it’s only in one place? I don’t really see the point then… i’d even argue that it would purely be intellectual masturbation

By: Jan Van Ryswyck

Jan Van Ryswyck — Tue, 08 Sep 2009 05:37:16 +0000

What made you choose for integrating the logging code into the processor itself instead of taking the AOP route or implementing a custom WCF behavior?