I don't want to get off on a rant here, but....

mod_ssl attacking Subversion clients, demanding client certificates

2012-01-30T16:10:05Z

Over the weekend upgraded to Subversion 1.7.2, Apache 2.2.21 (which contains mod_ssl 2.2.21). Everything worked great browsing the repository from a browser. Problems started as soon as svn command line or TortoiseSVN were used. Client Certificate prompts all over the place, sometimes cancelling worked, sometimes it caused the attempt to fail, general annoyance and stupidity across the board.

Verified 100 times that "SSLVerifyClient none" was set, moved it to vhost and directory levels as well, no dice. I could break browser access by setting it to require. Nothing worked to config it away, so I put back the old 2.2.15 mod_ssl file and bam, everything works like a charm again. It looks like there were some recent mod_ssl changes around optional at the server level prevented required at a lower level... it seems this went too far for some clients. Since 2.2.21 has been in the wild for a long time I'm guessing this only impacts the SVN HTTP library, since browsers work fine, and that would have caused a whole lot of rioting on the internet if browsers broke from the change.

You can have the Session but you have to know the secret knock

2011-08-12T21:29:40Z

Was very perplexed by this issue, context was fine, but context.Session was null, until finding this StackOverflow post.

One better though if you don't need to write context.Session is to implement: System.Web.SessionState.IReadOnlySessionState instead, which is probably somehow cheaper.

At least they support options

2011-06-02T12:02:12Z

"... will offer a choice of database services, including MySQL and the NoSQL system MongoDB. It also will offer MongoDB and Redis open source systems..." from Information Week's print article "VMWare Platform Takes It Deeper Into Cloud."

At least they have fixed it in the online version, though it's published under a different title. It's always funny to me that they can be so wrong technically while being so wrong editorially as well.

Troubleshooting slow ASP pages

2011-03-18T21:50:28Z

Since I've always wondered about this and never bothered to figure it out before....

In IIS 6 logs, the timetaken value represents the entire time that IIS was touching the request. This at the very least includes time after IIS got the request, but is waiting to hand it off to ASP/ASP.Net for processing. It may also include time spend sending the bytes back to the client, but I don't have big enough data or slow enough networks to really answer that question well.

This does make a certain amount of sense. It's an web server log, not a web framework log, but it makes trying to troubleshoot "slow" pages really, really hard. We can't run multiple workers because it breaks classic ASP Session. So everything just queues up every time there's a long running request. So given a page with a large timetaken in the log, did that page really run for a long time? Or was it sitting on a queue waiting on another page that was running slowly? Sure the first page in order that was slow is prob. the root cause of all the following slowness, but how does one determine that root page?

What would be ideal would be separate timequeued, timeprocessing, timenetwork type columns... maybe those exist in IIS 7.... of course that would require getting our classic ASP certified for Windows 2008.

Dear Information Week please just let me go

2012-01-24T20:57:37Z

Since cancelling my free subscription Information Week can no longer afford to fact check... oh wait, they never did that before either. Their latest flub that jumped off the page at me "...Java's JSON-based..." wait, what? Java's JSON? Pretty sure that's not the case and a 2 second Google confirms, "JSON (JavaScript Object Notation) is a lightweight data-interchange format..." wow, that was hard. Of course it's not the first time, and I'm sure it won't be the last.

I've ignored the renewal emails, for months, but the quality of reporting is so low, it doesn't really shock me that they can't manage to end my subscription either. I guess correctness doesn't matter nearly as much as number of eyes on the page when you're pitching to management types.

And just to get out ahead of Information Week's next mis-statement, "The final choice of name caused confusion, giving the impression that the language was a spin-off of the Java programming language." So, no, Information Week JavaScript has nothing to do with Java. Let me know when I can expect my Researcher check in the mail.

SQL Server object_name still takes int value for object_id

2010-07-15T19:40:23Z

If you're using the, not so new anymore, DMVs and pulling any bigints like resource_associated_entity_id you'll get an arithmetic overflow if you try to pass this value to object_name, because the object_id param to the function is still defined as int even in SQL Server 2008.

For example if you're looking at sys.dm_tran_locks to see what locks can't be granted during a block you need to only pass the resource_associated_entity_id if the resource_type is OBJECT otherwise pass NULL to get NULL back.

Granted looking up the object_name of a PAGE or KEY wouldn't give back anything useful, but it would be nice if it didn't blow up the query.

Tricky SQL XML support for binary values

2010-07-02T13:34:11Z

In setting up "Event" based block notifications for SQL 2005/2008 I had to get the binary SQL handle out of the XML provided by the event. This of course seems rather simple, except you can't just supply varbinary(64) as the type to @xml.value because that would too easy for an MS product. Trying this gives you back NULL instead of your binary value.

As this page tells you, you need to use an XQuery conversion, xs:hexBinary, to do instead. But wait, you're not off the hook yet, because xs:hexBinary doesn't understand 0x delimited binary, and doesn't tell you that in any way shape or form. Instead you get back your old buddy NULL. And that's when you notice buried in the XQuery that it's actually chopping off the 0x if the value has it (even though the test value in the code doesn't).

So the whole chunk you end up needing looks like this:

Process.value( 'xs:hexBinary( substring((frame/@sqlhandle)[1],3))', 'varbinary(64)' )

Empty User Names despite authenticating successfully

2010-05-06T14:51:09Z

Had an issue today where System.Web.HttpContext.Current.User.Identity.Name was returning String.Empty even though I was prompted to authenticate against the server when hitting the web page. After a few Googles I realized it was because the web.config was set to <authentication mode="None"/>. Changing it to "Windows" fixed the problem. I've never understood what that web.config line did in the past, since Auth is done at the IIS level and ASP.NET shouldn't really care. I guess now I know...

Telco analyst calls out the lying ISPs about bandwidth hogs

2009-12-05T21:39:24Z

Benoit Felten, an Telco analyst Yankee Group, calls out ISPs for claiming to suffer the effects of users who hog their precious bandwidth. He points out that no ISP has ever justified the existence of this class of users, nor have they ever released data about the usage of these hogs nor any other subset of their user population. Yet Time Warner Cable, amongst others, uses the "existence" of these users to justify arbitrary, and I would claim exceedingly low, bandwidth caps after which they gouge users with additional fees. It's nice to see someone close to the industry finally saying what a lot of us on the outside have been thinking and trying to shed light on for a long time.

He also lays down a gauntlet in front of the ISPs challenging them to provide him with usage data that he could analyse to understand their assertions a small number of aggressive users have such a large impact on the experience of the many. So he's not just throwing stones, he's asking the ISPs to provide real, raw, hard data to back up their assertions via independent analysis.

App Engine and Bloog not getting along

2009-12-04T13:48:42Z

So I noticed that the home page wasn't loading, and apparently a few others noticed too that something in the new version that Google has started pushing to it's servers breaks Bloog. I pushed my fix to github, which disables the Tag list on the hope page for the time being till the issue can be really fixed.

Update: Long term fix committed in two commits to github: 1, 2

I basically took the AppEngine code for the method that was causing the error, and pulled it into the Bloog code, changing as needed to get it to run there. I can now load my homepage, edit posts (as this proves), etc.