Question

Starting sometime in November we ceased to be able to save edits on several pages in our 4 year old (and large) TWiki install. One can edit fine, but any attempt to preview or save your work leads to a long hang, followed by (in FireFox) a dialog that says "The document contains no data". The error in Safari (which takes a lot longer to turn up) is 'Safari can’t open the page “http://csci.mrs.umn.edu/UMMCSciWiki/bin/save/TWiki/TWikiPreferences” because the network connection was reset, which sometimes occurs when the server is busy. You might be able to open the page later.'

The problem appears to be spreading, or perhaps we're just noticing it on more pages as we try to edit things people hadn't touched in a while. Once a page can't be edited, it seems to remain uneditable. Unfortunately it's reached a crisis point because users aren't sure they can trust the wiki and are moving away from it. We've been digging on our own for several weeks, but come up with nothing, so any help folks can provide will be greatly appreciated.

Looking through the logs doesn't turn anything up that we can make sense of. Looking at the .txt and .txt,v files doesn't suggest anything.

It's possible that it's related to the size or complexity of the page (or attachment that you're adding - that sometimes hangs as well), although I've not been able to work out any coherent pattern :-(.

When we first noticed the problem we were running the BeijingRelease. We upgraded to Cairo about 2 weeks ago, and it had no effect positive or negative. testenv is happy. We've been through the Apache error logs, and can't find anything that seems helpful.

Some pages that seem possibly related include DocumentContainsNoData and SlowResponsePreviewAndSaveMore4k, but neither suggested anything that would help fix our problem.

As I mentioned above, this is becoming a critical threat to the survival of TWiki here, and any help would be greatly appreciated.

-- NicMcPhee - 19 Jan 2005

Additional Information

when the pages time out, it is b/c they don't get correctly authenticated. tries perhaps 2x and then dies. (I think)

the pages that can be edited produce a log like this:

12.215.34.167 - MikeAnderson [21/Jan/2005:15:54:00 -0600] "POST /UMMCSciWiki/bin/save/Sandbox/TestTopic4 HTTP/1.1" 302 5 "http ://csci.mrs.umn.edu/UMMCSciWiki/bin/edit/Sandbox/TestTopic4?t=1 106344380" "Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.5) G ecko/20041111 Firefox/1.0"

whereas, the unsuccessful attempts produce logs like this:

12.215.34.167 - - [21/Jan/2005:16:02:15 -0600] "POST /UMMCSciWi ki/bin/save/TWiki/TWikiPlugins HTTP/1.1" 401 - "http://csci.mrs .umn.edu/UMMCSciWiki/bin/edit/TWiki/TWikiPlugins?t=1106343430" "Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.5) Gecko/200411 11 Firefox/1.0"

the difference in these two files are the result codes. 302 vs. 401 , and the fact that in the one that works, there is MikeAnderson or any other vaild wiki user who can edit the page. the ones that fail have a "-" in this place.

I hope that this added information may help lead to a solution- Thanks!

-- MichaelAnderson - 21 Jan 2005

Environment

-- NicMcPhee - 19 Jan 2005

Answer

HangsSavingLargePages and ApacheSnipsOopsmore might be of help. Also feel free to drop by TWikiIRC for more help. It'd be a real shame not to see this long standing twiki deployment running smoothly.

-- MattWilkie - 20 Jan 2005

Indeed. As Matt says, your best bet is to drop by IRC (early in your morning is best, as you'll catch the Europe-based developers that way).

Some things to check before you do.

1. Are you using your own skin? Are you sure the form in the edit page has method="POST"?
2. Have you looked at what happens to the transaction (using a tool such as lynx)?

Chances are this is a timeout issue.

-- CrawfordCurrie - 20 Jan 2005

Seems to me that I saw similar problem once when there was an issue of running out of disk space.

-- LynnwoodBrown - 20 Jan 2005

Thanks for all the quick feedback. We may indeed be dropping by the IRC if none of these suggestions shed light on the problem.

Matt: The pointer to HangsSavingLargePages looks very helpful. What's described there certainly seems like the kind of thing we're seeing. Hopefully something there will help us out.

Crawford: We do have our own skin on two of the webs, but everything else was running the standard skins. As far as I'm aware, nothing was changed on the skin front that would account for this new behavior. The edit page does have method="POST". Do I need to do something special with lynx to see more about the transaction? I tried editing a troublesome page with lynx, and it hung for a very long time with the message "HTTP request sent; waiting for response.", then gave me a message about an unexpected network error and aborting (it flashed by, so I didn't capture it exactly ).

Lynnwood: I'm pretty sure disk space isn't the issue. The partition the wiki install lives on has over 500Mb free.

Again, thanks to all. We'll keep you posted.

-- NicMcPhee - 20 Jan 2005

try http://livehttpheaders.mozdev.org/ for recording all the transactions in Firefox/Mozilla.

-- MattWilkie - 22 Jan 2005

The problem seems to be a problem with authentication when saving the page. Changine the .htaccess file so that require valid-user has 'allow from all' lets us save the pages that could not be saved before.

-- MichaelAnderson - 24 Jan 2005