That does not sound like what I have observed.

On 2011-08-24, at 11:42 AM, Tim Worman wrote:

> On Aug 22, 2011, at 12:53 PM, Chuck Hill wrote:
> 
>> On 2011-08-22, at 12:48 PM, Tim Worman wrote:
>>> On Aug 22, 2011, at 11:51 AM, Tim Worman wrote:
>>> On Aug 22, 2011, at 11:38 AM, Chuck Hill wrote:
>>>>> On 2011-08-21, at 12:43 PM, Tim Worman wrote:
>>>>>> On Aug 21, 2011, at 11:52 AM, Chuck Hill wrote:
>>>>>>> On 2011-08-20, at 4:02 PM, Tim Worman wrote:
>>>>>>> 
>>>>>>>> Back in January I started this discussion on this same topic:
>>>>>>>> 
>>>>>>>> http://lists.apple.com/archives/webobjects-dev/2011/Jan/msg00224.html
>>>>>>>> 
>>>>>>>>>> I have an app that, during the course of normal usage, is starting 
>>>>>>>>>> httpd processes on the server that instantly hit 100% CPU usage of 
>>>>>>>>>> one core. This can happen multiple times during times when the app 
>>>>>>>>>> is under heavier load. After some time I can have many httpd 
>>>>>>>>>> processes where TOP reports each using 100% of a core. When I try to 
>>>>>>>>>> log into the app and poke around to try and reproduce the issue, I 
>>>>>>>>>> am unable.
>>>>>>>> 
>>>>>>>> This is an update to my original post hoping to see if there are 
>>>>>>>> anymore thoughts on origin. More recently, I have been able to 
>>>>>>>> reproduce the issue in my own usage of the app - something I wasn't 
>>>>>>>> able to do before. It seems to be easier to generate the issue now 
>>>>>>>> that there are more ajax requests. The methods executed by these 
>>>>>>>> requests are not intensive or long responses and should return a 
>>>>>>>> result in seconds. Some symptoms:
>>>>>>>> 
>>>>>>>> - When the actions are executed, busy indicators properly spin while 
>>>>>>>> the browser awaits a response from the server. When the issue occurs, 
>>>>>>>> the response never comes.
>>>>>>>> - while continuing to await a response there is concurrently an httpd 
>>>>>>>> process that pegs he processor at 100%
>>>>>>>> - if I kill the process on the server, the browser immediately updates 
>>>>>>>> properly as if the request had run properly
>>>>>>>> 
>>>>>>>> It's almost as if apache is somehow receiving an ill-formed request 
>>>>>>>> and chokes on it. The problem is, there are no errors in the console 
>>>>>>>> or anything strange in any apache logs. Has anyone ever seen behavior 
>>>>>>>> like this or have any ideas as to how I could analyze it further? 
>>>>>>> 
>>>>>>> I've seen something like this.  It appeared that the woadaptor (i.e. 
>>>>>>> mod_webobjects) did not believe that it had received all of the 
>>>>>>> response from the application.  The app had nothing more to send and so 
>>>>>>> the woadaptor just hung there waiting for data that would never come.  
>>>>>>> I did not track down why this happened, but it did seem to be load 
>>>>>>> related.  My suspicion was that there is a concurrency bug in the 
>>>>>>> woadaptor.
>>>>>> 
>>>>>> I'm really at a loss about what to do about it. It's only gotten worse 
>>>>>> as I've included more ajax actions in my app - and, of course, I don't 
>>>>>> experience this behavior in development. I just deployed a major update 
>>>>>> to my app - pretty much unaware that a small problem was going to become 
>>>>>> a big problem with the new version.
>>>>>> 
>>>>>> In one example, a have a calendar where clicking on a day simply calls 
>>>>>> an AjaxUpdate marking that date as selected to the calendar. The result 
>>>>>> also has to update the entire page though because other things on the 
>>>>>> page need to change in those circumstances. This alone can cause the 
>>>>>> issue - but not always. And it happens even when I'm the only logged in 
>>>>>> user - so the load isn't high.
>>>>>> 
>>>>>> As one solution, I've considered rolling a custom apache instead of 
>>>>>> using Apple's. But since the server also runs shibboleth, the setup 
>>>>>> isn't exactly simple. But I'm really not sure how to ascertain if the 
>>>>>> problem is the woadaptor or how I can settle it.
>>>>> 
>>>>> 
>>>>> You could try the CGI adaptor and see if that makes any difference.  If 
>>>>> it IS a bug in mod_webobjects...  that could be hard to find and fix.
>>>> 
>>>> Yeah, I'd say so. I don't even know C. :-) I'm gonna pursue some other 
>>>> things first - like a possible hardware issue, or other software. I'm 
>>>> going to try a deployment on another server and see if that has any effect 
>>>> - also without shibboleth. Is there any chance that an issue with the id's 
>>>> of page elements could cause an issue like this - say if they're dynamic 
>>>> and the response doesn't find a matching id?
>>> 
>>> One other thing - if this was a bug in mod_webobjects, wouldn't a whole lot 
>>> of projects being experiencing this same issue? Obviously, there are wo 
>>> projects out there taking on a lot more load than mine. I have to think 
>>> there is something about this specific deployment OR something about my 
>>> specific requests that's causing my little hell here. Obviously, in wo, we 
>>> don't construct our own requests in a fine-grained way so that is why I'm 
>>> wondering about some of the elements on my pages.
>> 
>> 
>> If it is a bug in the adaptor it has to be something very specific, maybe a 
>> size or the value of some bytes at the start or end or the phase of the moon 
>> or...  What platform do you deploy on?
> 
> What's the possibility that this could be related?
> 
> http://www.h-online.com/open/news/item/Tool-causes-Apache-web-server-to-freeze-1330105.html
> 
>> Chuck

-- 
Chuck Hill             Senior Consultant / VP Development

Practical WebObjects - for developers who want to increase their overall 
knowledge of WebObjects or who are trying to solve specific problems.    
http://www.global-village.net/products/practical_webobjects







 _______________________________________________
Do not post admin requests to the list. They will be ignored.
Webobjects-dev mailing list      ([email protected])
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/webobjects-dev/archive%40mail-archive.com

This email sent to [email protected]

Reply via email to