Bug 26621
| Summary: | REGRESSION: Pages reload as blank at wccnet.edu | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Sarah Rick <srick> |
| Component: | Page Loading | Assignee: | Nobody <webkit-unassigned> |
| Status: | RESOLVED INVALID | ||
| Severity: | Normal | CC: | ap, koivisto, markc.barrett, phil, ryan.lists.warpshare, sift.del |
| Priority: | P1 | Keywords: | InRadar, Regression |
| Version: | 528+ (Nightly build) | ||
| Hardware: | Mac | ||
| OS: | OS X 10.5 | ||
| URL: | http://www.wccnet.edu | ||
Sarah Rick
We believe that we have found a bug with Webkit based on using the Safari 4 browser on our website.
Our website uses extensive URL redirecting on its assests. On user navigation or reload, the page either doesn't load
certain elements (missing images, missing CSS styling, etc.) or doesn't load a page at all (the screen remains white). AJAX-loaded
content also seems affected.
This problem is intermittant and it has taken anywhere from under a minute to over ten minutes for our testing to
uncover a partially-loaded or non-loaded page. Our logs do not show any indication of there being too many redirects for
either the URL or for the assets being loaded on the page.
This bug is affecting both web pages as well as specific files: for example, on Safari 4.0.0, a page that doesn't load (blank page)
gives the error "too many HTTP redirects" in the Activity window. This error generally shows while the page is attempting to load
favicon.ico. In addition, when a page partially loads and has missing graphics, attempting to load a missing graphic in a new
tab or window results in a Safari error telling us that "Too many redirects occured while trying to open the image."
When the browser is restarted with an emptied cache and history, the page has no problems loading.
This bug affects various pages on the wccnet domain, but is most noticeable while navigating through the following sections:
http://www.wccnet.edu/lifelong-learning/
http://www.wccnet.edu/news-events/
http://www.wccnet.edu/calendar/
Especially noticable while navigating through the arrows around the "June 2009" label (may need to reload calendar to see it)
We suspect that the problem is with cached redirects. Have you identified this issue in Safari 4? The problem is not present
in Safari 3, or any other major browsers (such as Firefox 2 & 3 or IE 7 & 8).
After light testing in both Panic's Coda and NetNewsWire and seeing similar results, it's our guess that WebKit may be the
source of the issue. We intend to further research this via WebKit, but would also appreciate a response from Apple regarding
the issue being present in Safari. If, however, you discover that the problem is on our end, we would be happy to hear
your suggestions for fixing it. We did find other users reporting similar problems:
http://forum.cmsmadesimple.org/index.php?topic=34710.msg163850
We did our testing on Safari 4.0 (4530.17), 4.0.1 (5530.18), NetNewsWire 3.1.7 and Coda 1.6.4, all for Mac OSX 10.5.
In addition, we installed the nightly build of Webkit r44932 onto Safari 4.0.1, and are still experiencing these problems.
The instances of pages partially loading seemed to go down, however nearly every page load resulted in a blank page.
If you would like additional data to further research this problem, feel free to email us at websupport@wccnet.edu.
Thanks for looking into this,
Washtenaw Community College
Web Operations
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Alexey Proskuryakov
Confirmed with a local debug build of r44975 on Mac OS X 10.5.7. Re-titling to talk about the symptom seen with current nightlies only.
Steps to reproduce:
1) Open http://www.wccnet.edu/lifelong-learning/
2) Reload once or twice.
Results: a blank page.
This doesn't occur if WebCore cache is disabled via Debug menu. This also doesn't occur with Safari/WebKit 3.2.3 on the same system, so this is a regression.
Alexey Proskuryakov
<rdar://problem/7001796>
Alexey Proskuryakov
Looking at tcpdump trace, I see that HTTP requests for the main resource and the icon are sent at once over the same TCP connection, without waiting for response. As far as I know, HTTP pipelining is not supported by CFNetwork, so it looks like a bug.
So, this could be a CFNetwork bug - but something in WebKit now triggers it, making it a regression.
Antti Koivisto
The blank page seems to occur if the previous TCP connection(s) is still open when the new load starts. Reloading is not necessary, enter in address field fails as well.
I see the server terminating the connection immediately after GET with an RST packet. This reply packet is also maformed (according to Wireshark):
Acknowledgement number: Broken TCP. The acknowledge field is nonzero while the ACK flag is not set
Seems like a networking stack level problem.
Ryan Parman
Experiencing an issue with identical symptoms on a password-protected site. This happens on Safari 4.0.x on Windows and WebKit r48398 with Safari 4.0.34 chrome on Mac OS X 10.5.x & 10.6.1.
The latest Google Chrome dev-build (r4.0.207) on Mac OS X 10.6.1 and Ubuntu 8.04 LTS does NOT have this issue.
Mark Barrett
Has this came back? We seem to be getting almost exactly the same intermittent problem at https://www.moneyclaim.gov.uk with Safari users. It seems to be happening initially during redirects and refreshing often doesn't help.
I can easily reproduce this myself with just a few clicks using 5.0 (7533.16) version of Safari on Windows XP.
Ryan Parman
When our organization encountered this issue, it was because our VIP was using *selective* chunking. Changing this to *persistent* chunking and explicitly sending the `Connection: closed` HTTP header solved the issue.
Mark Barrett
Upgraded to Safari 5.0.1 and still an issue.
What's strange is that this seems linked to error messages seen in the error console: "Resource interpreted as image but transferred with MIME type text/css". Our site is transferring the correct mime types and this only appears when using conditional GETs.
When the blank page shows up, Web Inspector actually shows the response as a 304 when the access logs show the server sent a 302! Safari seems to be getting majorly confused and mixing up responses with different requests. I can tell this because of other HTTP headers in the response that I know that do not belong to that request and belong to another type of request.
I wish I could Wireshark this but can't when the site is using HTTPS.
Alexey Proskuryakov
We do not believe that this is a bug in open source WebKit code. <rdar://problem/7001796> tracks making Apple's lower level libraries more robust in this case.
Closing as INVALID per our policy as a non-WebKit issue. But thanks for reporting it, Apple engineers working on the networking libraries are investigating it.