657431 - pastebin often has random name in it

Status: RESOLVED FIXED

(mozilla.org Graveyard :: Server Operations, task)

Description

[John Ford [:jhford] CET/CEST Berlin Time]

On a fresh pageload of pastebin.mozilla.org, I see someone who I have never heard of's name in the 'Your Name' box.  This happens about 80% of the time that I use pastebin and it seems to be a different name almost every time.

Comment 1

[Wil Clouser [:clouserw]]

I see this pretty often too, although I usually recognize the name as someone from mozilla.

Comment 2

[Justin Dow [:jabba]]

Looks related to https://bugzilla.mozilla.org/show_bug.cgi?id=651675 . Not sure what the fix was.

Comment 3

[Jake Maul [:jakem]]

No idea what's broken here. I cannot find anything cached for 'pastebin.mozilla.org' on any of the Zeus clusters, but caching does sound like the most likely culprit.

I am getting a "Via: NS-CACHE-6.0:   4" header when I hit the page. Jabba, any idea what that is? It's not our usual Zeus header. But then, I can't figure out for sure which Zeus cluster it's going in through.


Unfortunately the pastebin we're running is no longer in active development- we have the newest version, and it's 4 years old. If this is some type of bug in the code (which would be my next guess after caching- like perhaps it associates you to a name based on your IP, which would explain why I - remote - never see it), then we'll either have to fix it ourselves or switch to a pastebin system that's still in active production.

Comment 4

[Dave Miller [:justdave]]

> I am getting a "Via: NS-CACHE-6.0:   4" header

That comes from a Netscaler, not Zeus.

Comment 5

[Jake Maul [:jakem]]

This may be fixed now. I discovered that the servers were not sending any cache-related headers with their responses, and that the NetScaler was caching responses for a short while. It may have been serving up cached versions of the page, based on whomever was lucky enough to "seed" the cache at just the right time. By default it appears to cache for a little while (an hour, I think), even if no headers were sent telling it to.

I'm not sure if the NetScaler is taking the source IP into consideration when it decides whether or not to server from cache... I believe it does, at least for pastebin.mozilla.org (or possibly for any site that doesn't send cache headers, like pastebin was). This would cause the problem to affect people sharing a NAT connection much more heavily, and would explain why I haven't been able to trigger it (I'm remote).

In any case, I have added Expires and Cache-Control headers to Apache's config for pastebin.mozilla.org- The NetScaler should no longer cache anything that is text/html, but should still cache the CSS and Javascript files.


Let me know if the problem appears any better now.

Comment 6

[Jake Maul [:jakem]]

I believe this is taken care of by the caching change above. If not, please reopen.