Why is LPS causing aspnet_wp to run away?

[yoDon]

I have LPS-2.1.2 demo running on port 8080 of an XP Pro SP2 machine with IIS 5.1 running on port 80 (it's a default windows install of the LPS system, the only config change I made was to increase the connection timeout to 60 seconds).

I can have Jakarta Tomcat 4.1.12 open without any problems, but as soon as I bring up a laszlo page in my browser (such as http://localhost:8080/lps-2.1.2/lz-utils/welcome.html), the IIS aspnet_wp.exe process starts going crazy.

Every ten seconds, aspnet_wp.exe starts consuming as much CPU as it can get (driving system idle to 0% and CPU usage to 100%). The processor stays pegged for a few hundred milliseconds and then aspnet_wp.exe returns to idle until the next tick of the ten second clock.

Once I've opened the LPS welcome page, aspnet_wp will continue to burn CPU every ten seconds even after I close the browser that welcome.html was running in.

The strobing continues until I click the close widget on the Tomcat console window or use the "Stop the Laszlo Presentation Server" start menu item to shut down Tomcat.

-Don

[bloch]

Wow, that sounds pretty un-fun.

If you disable IIS and ASP.NET completely and then simply start tomcat via the LPS, do you have any other probs?

Fwiw, we haven't yet qa-ed against SP2. And, we typically don't qa with IIS running.

In general, at startup, the LPS will cause Tomcat to chew up a bunch of memory and cpu cycles. But that should only go to the java process. And things should settle down once the welcome page is up.

The only guess I can make is that SP2 comes with something that makes aspnet_wp.exe (I don't even know what that really is) do something bad every 10 seconds when the system is generally under hi cpu and memory utilization.

How much memory does you machine have?

-Eric

[yoDon]

> we haven't yet qa-ed against SP2

One developer to another, I can't recommend strongly enough that you switch your qa machines over to SP2. For most applications, the switch to SP2 is a non-event, but for an unlucky few products the switch to SP2 rocks your world in a very unhappy way. MS knew they had to make changes to the security model that would break some existing apps (not many, but enough). That's why they had it in public beta for six months, to give all developers plenty of time to test and react. The beta's over, you've lost your lead time, and SP2 is on its way into the wild over the updater. You need to have SP2 on your test machines today. Period. My experience could be completely unrelated to SP2, or it could be the tip of the iceberg. There's no way to know at this point. It takes about 30 seconds of tester time and 20 minutes of latency while the pack installs before you are good to go. There's simply no excuse other than denial for not already having SP2 on your test machines.

> How much memory does your machine have?

1024 Meg, but the cpu utilization recovery is very fast and periodic so I don't think it's driven by any kind of swapping behavior, if that's what you were thinking. I think aspnet_wp is doing something CPU intensive for a few hundred ms. and then it's done for 10 seconds.

[yoDon]

> If you disable IIS and ASP.NET completely and then simply start tomcat via the LPS, do you have any other probs?

Tomcat and LPS seem to be working fine, if that's what you mean. I can't work on my app (because aspx WebService projects under VisualStudio need to handshake with IIS at development time) and I can't test my app (again, because it consumes aspx WebServices hosted under IIS), but I can run the Laszlo welcome page without any problems

-Don

[bloch]

Thanks, yoDon.

1) Do other laszlo apps work (amazon demo for example) ?

2) Here's what I was thinking:
Tomcat uses up enough memory that I wouldn't run too many other "big" things at the same time. It's *posssible* you simply need more memory in your box. I'd check out the memory usage of IIS and tomcat and verify that you're not going into swap land. (I believe we ship tomcat with -Xmx512). It's also completely possible that something else is going on.

FYI, we believe that pre-2.2 Laszlo apps can be served to and will work on SP2 systems unmodified; but the development environment *combined* with IIS and ASP.NET services is another story. In general, we don't QA the dev environment on Windows with IIS turned on.

Do you think you should file a bug with microsoft? As you report, it's aspnet that's messing with your system, unless there's some other thing that's not working in tomcat.

I'd do some more googling/searching of msdn for issues with aspnet_wp.exe.

In the meantime, you could try setting the priority of aspnet process down low.

What does that process do, btw?


-Eric

[yoDon]

Depending on your point of view, this post is either good news or bad news. I've lost the ability to repro this bug. Yesterday, it was 100% reproducible. If I started LPS and viewed any laszlo site, aspnet_wp started going nuts and continued to do so until I shut down the tomcat server. Today, I decided to do run a moderately extensive test matrix, rebooting with IIS on and off, starting tomcat before IIS and tomcat after IIS, launching laszlo apps before and after touching any IIS-served pages, etc. Somewhere in that process I did something that made the IIS runaway behavior stop. Good news is it's gone, bad news is I don't know why it happened, why it stopped, or whether it will return.

Hopefully it was some very low-probability event that caused the state and neither I nor anyone else will ever encounter it again.

> what does [aspnet_wp.exe] do?

I believe the "wp" means "worker process". Basically, if IIS is serving up any .net code (eg. a visitor requests an aspx page or consumes a web service that IIS is hosting) then the code runs inside the aspnet_wp.exe process. If I want to debug my web service code, I attach the debugger to the aspnet_wp.exe process so that I can set breakpoints in my code.

> Do you think you should file a bug with microsoft?

During the SP2 beta cycle, MS was very responsive to ISV issues (I had friends both on the SP2 team and at ISVs who worked with MS during the beta, so I know quite a bit about that process... it was nothing like what the slashdot posts would lead you to believe). Now that the beta is over, flexibility on ISV issues is close to zero because the service pack is on its way to tens (hundreds?) of millions of boxes.