Sometimes (quite often) a page on the Drupal.org site continually loads (i.e. Chrome shows it's still loading as it shows its page spinner for that tab).
It doesn't always happen, however it takes up about 70% CPU on a reasonably high end machine when it does. Once the offending tabs are killed Chrome goes down to < 4% CPU.
Example forum post that did it (although it won't consistently do it)
https://www.drupal.org/node/2445521
Configuration
Chrome Version 56.0.2924.87 (64-bit).
OSX 10.11.6.
Let me know if you want me to debug this further. Always happy to help if I can :)
Note the problem doesn't seem to occur on other websites.
Comment | File | Size | Author |
---|---|---|---|
#9 | Shot 06032017-09-28.png | 30.3 KB | apaderno |
Comments
Comment #2
CatsFromStonehenge CreditAttribution: CatsFromStonehenge commentedComment #3
cilefen CreditAttribution: cilefen commentedThe forums are Drupal 7. I think we usually send these to the webmasters queue. Please attach the output of the browser's console log if you can catch it happening.
Comment #4
apadernoI am having the same problem with this very issue queue. When Google Chrome keeps spinning, the Dev Tools show it's a missing cache on Varnish. For example, one of the failed requests had the following header.
Comment #5
apadernoI thought it could be Google Chrome to create problems in some way, but I tested also with Firefox Developer Edition, and I am getting the same issue.
Comment #6
apadernoComment #7
drummWhat’s happening on the browser’s network log/timeline when this happens? What specific request(s) are hanging?
Comment #8
apadernoAfter changing DNS settings for my computer, in the past days I didn't notice any slow-down. Using Firefox I just noticed the page request keeps too much, but no error code was given from Firefox Network tool.
As for when it happens, or it's an admin view, or it's a page with a form. I noticed the slow-down when I was deleting spam messages; in that case, the confirmation form was not appearing, or the page shown after submitting the confirmation form was not appearing (i.e. the browser loading spin kept spinning).
Now that I am using Chrome again, I will try with the spam view and see if I still have problems.
Comment #9
apadernoI just had a problem accessing https://www.drupal.org/admin. This is the network output.
These are the response headers. (I removed the part about the cookies.)
Notice that I was still seeing a white screen after 5 minutes. I just clicked on the button to return here to write this comment; otherwise, I was still waiting for the page to show.
Comment #10
apadernoIn the case it's not readable, the error shown from the console for https://www.drupal.org/admin is net::ERR_CONTENT_LENGTH_MISMATCH.
Comment #11
apadernoI am trying with a different Chromium-based browser (Brave), and so far, I am not having the same issue.
I am going to use this browser for the next days, so I will be able to verify it's Google Chrome that is causing me any trouble with drupal.org.
Comment #12
apadernoWith the Brave browser, the worst I got was 16 seconds for the front page, and 8 seconds for an issue node. With this browser, I have never seen a page not loading as with Google Chrome.
The only difference I saw is that this browser blocks both https://www.google-analytics.com/analytics.js and https://js-agent.newrelic.com/nr-1026.min.js. Could that make a difference, in someway?
Notice that the problems I had with drupal.org with Chrome and Firefox were the same problems I had with api.drupal.org.
Comment #13
apadernoTo make a comparison, these are the response headers for https://www.drupal.org/admin the Brave browser reports.
Comment #14
drummThe Content-Length mismatch definitely looks like the problem. The URL for that is a www.drupal.org page, not any 3rd-party assets.
Drupal itself doesn’t serve a Content-Length header, so it must be added by our CDN or web server. I’m checking with Fastly support to see if they can help, although it is possible it is a problem on our end.
Comment #15
drummI think this header is being set by nginx. The only recent error logs I can find are occasional
upstream timed out (110: Connection timed out) while reading response header from upstream
. The nginx proxying has a timeout of 90s.Comment #16
apadernoI am changing the title, since this is not just about forums. Then, I can get the same behavior for Drupal API too.
Comment #17
apadernoWith Google Chrome 60.0.3112.50 beta and Google Chrome 61.0.3152.0 canary, I get first a white page and, refreshing, the correct page. The content length shown is different between the two cases, but it's never 0.
When I get a white page, the content shown from the Developer Tools are truncated, for example as follows:
In both the cases, the HTTP code returned for the main resource (e.g. https://www.drupal.org/u/ramyareddy) is 200.
Comment #18
apadernoI am not having these problems anymore. I am setting this as fixed.