Closed
Bug 465529
Opened 16 years ago
Closed 16 years ago
Netscaler flush for AMO
Categories
(Infrastructure & Operations Graveyard :: WebOps: Other, task)
Tracking
(Not tracked)
VERIFIED
FIXED
People
(Reporter: wenzel, Assigned: oremj)
Details
Yesterday's push hasn't completely made it into the netscaler: Fashion your firefox's CSS file is out of date. Please flush that for us, will you? If you can selectively flush, get AMO/css/*, otherwise all of AMO. Thanks. It's being presented in NYC at the moment, so we should get this out ASAP.
Comment 1•16 years ago
|
||
Flushing but I'm surprised it didn't age itself out of the cache. Max cache age is 3600 seconds unless you specified some crazy cache-age. Not even sure how to test, assuming it's fixed.
Status: NEW → RESOLVED
Closed: 16 years ago
Resolution: --- → FIXED
Reporter | ||
Comment 2•16 years ago
|
||
Thanks a lot. To test: https://addons.mozilla.org/en-US/firefox/fashionyourfirefox/ should look like https://preview.addons.mozilla.org/en-US/firefox/fashionyourfirefox/ Which it doesn't :( The good news is, the NS was not at fault. Bad news: We don't know who is serving the old (pre-push) CSS file.
Comment 3•16 years ago
|
||
What makes it seem like the NS is that it looks right when you're logged in to AMO and have the AMOv3 cookie which bypasses the NS cache for a logged-in user.
Comment 4•16 years ago
|
||
Is there a specific file I can test against with curl?
Reporter | ||
Comment 5•16 years ago
|
||
Sorry, I need to reopen: https://preview.addons.mozilla.org/css/style.min.css?19906 is served with the following headers: https://addons.mozilla.org/css/style.min.css?19906 GET /css/style.min.css?19906 HTTP/1.1 Host: addons.mozilla.org User-Agent: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:1.9.0.4) Gecko/2008102920 Firefox/3.0.4 Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8 Accept-Language: en-us,en;q=0.7,de;q=0.3 Accept-Encoding: gzip,deflate Accept-Charset: UTF-8,* Keep-Alive: 300 Connection: keep-alive Cookie: __utma=150903082.174226280.1213422549.1226005181.1226005650.18; __utmz=150903082.1224270483.16.8.utmccn=(organic)|utmcsr=google|utmctr=get+firefox+3|utmcmd=organic; __utma=164683759.1806091917.1213674996.1226986837.1227019833.328; __utmz=164683759.1226720602.320.77.utmccn=(referral)|utmcsr=bugzilla.mozilla.org|utmcct=/show_bug.cgi|utmcmd=referral; dloadday=128.237.229.48.1214256165114786; __utmc=164683759; AMOappName=firefox; __utmb=164683759 Pragma: no-cache Cache-Control: no-cache HTTP/1.x 200 OK Date: Tue, 18 Nov 2008 03:36:26 GMT Expires: Fri, 16 Nov 2018 03:37:20 GMT Cache-Control: max-age=315360000 Content-Length: 11572 Connection: Keep-Alive Via: NS-CACHE-6.0: 4 Etag: "c7da" Server: Apache/2.2.3 (Red Hat) Last-Modified: Tue, 04 Nov 2008 21:30:08 GMT Accept-Ranges: bytes ntCoent-Length: 51162 Content-Type: text/css Content-Encoding: gzip The last-modified header is not supposed to be November 4th. Yes it has a crazy-long max-age, but we need to get that file off the cache somehow :-/
Status: RESOLVED → REOPENED
Resolution: FIXED → ---
Reporter | ||
Comment 6•16 years ago
|
||
Oh, well, my comment is a little late. Specific test would be the Last-Modified header. If it's not November 4th, it worked.
Reporter | ||
Comment 7•16 years ago
|
||
I am not sure if you did anything additional, but the Last-Modified header is showing November 18th now, and it is now magically working the way it is supposed to. Thanks! For the future, we should find out though what caused this, so we can avoid it going forward.
Status: REOPENED → RESOLVED
Closed: 16 years ago → 16 years ago
Resolution: --- → FIXED
Comment 8•16 years ago
|
||
I think the ? parameter is causing unique entries to be cached. root@nslb02# nscachemgr -a | grep addons.mozilla.org | grep style.min.css 0x000000018166571c5bd2 DEFAULT GET //addons.mozilla.org:443/css/style.min.css?19906 0x00000003ea8c572ca27b DEFAULT GET //versioncheck.addons.mozilla.org:443/css/style.min.css?19906 0x00000007d8965736d540 DEFAULT GET //addons.mozilla.org:443/css/style.min.css?17704 0x000000086a755709a4cd DEFAULT GET //addons.mozilla.org:443/css/style.min.css?18878 0x0000000ca86f573b48f9 DEFAULT GET //addons.mozilla.org:443/css/style.min.css?19426 0x0000000e1f43572759df DEFAULT GET //addons.mozilla.org:443/css/style.min.css?19143 0x0000000e5856554df35c DEFAULT GET //addons.mozilla.org:443/css/style.min.css?19465 oremj, any idea how to make those non-unique?
Assignee: server-ops → oremj
Severity: blocker → normal
Status: RESOLVED → REOPENED
Resolution: FIXED → ---
Assignee | ||
Comment 9•16 years ago
|
||
Shouldn't they by unique? Is addons still on the 15k?
Comment 10•16 years ago
|
||
It is on the 15k and I don't know.
Assignee | ||
Comment 11•16 years ago
|
||
Sorry, I am jumping in to this bug late. What issues currently need to be a addressed to fix this bug?
Reporter | ||
Comment 12•16 years ago
|
||
Can you un-mark this as a private bug, please? Then we can discuss how to avoid this in the future.
Updated•16 years ago
|
Group: infra
Reporter | ||
Comment 13•16 years ago
|
||
(In reply to comment #11) > Sorry, I am jumping in to this bug late. What issues currently need to be a > addressed to fix this bug? What I think *seems* to have happened is: - during push, the data is rsynced. The php files made it over yet, the CSS files didn't. - a user performs a GET request for /css/style.min.css?19906 (due to the updated php files) - the app node fetches the *OLD* CSS file from the file system, it is cached for eternity. - rest of the update is performed ... - => all users get an old CSS file disguised as the new revision. To fix it, we need to make sure this does not happen again. - One way would be to routinely perform a netscaler flush (for AMO anyway) after a push has happened completely. That'd also get rid of the old cached revisions that are not needed anymore. - or we make a push atomic: That is, the rsync goes to a different directory, and only when all is in place on all app nodes, a switch is flipped (symlink switched, for example) to make all new files available at once. Oremj: What do you think is the best (and least painful for IT) way to approach this?
Comment 14•16 years ago
|
||
(In reply to comment #13) > To fix it, we need to make sure this does not happen again. > - One way would be to routinely perform a netscaler flush (for AMO anyway) > after a push has happened completely. That'd also get rid of the old cached > revisions that are not needed anymore. Wil made the point that this won't work b/c the users cache will still hang on to the files because the expire time in 10 years. Looks like atomic updates are the best solution?
Reporter | ||
Comment 15•16 years ago
|
||
Also, we need to push a new update soon, just to increment that CSS number, because all people who visited the site this morning are now stuck with the outdated file forever.
Assignee | ||
Comment 16•16 years ago
|
||
Since git is used to push out to the cluster the updates are pretty atomic. My guesses at what happened are: * The auto commit script ran half way through the svn up, so only a group of the files were synced out at first and then the rest 5 or so min later. * The old css file was served for around one second, but still was cached by the netscaler/clients.
Assignee | ||
Updated•16 years ago
|
Status: REOPENED → RESOLVED
Closed: 16 years ago → 16 years ago
Resolution: --- → FIXED
Verified FIXED; Fashion Your Firefox and AMO are both fine, now.
Status: RESOLVED → VERIFIED
Updated•11 years ago
|
Component: Server Operations: Web Operations → WebOps: Other
Product: mozilla.org → Infrastructure & Operations
Updated•5 years ago
|
Product: Infrastructure & Operations → Infrastructure & Operations Graveyard
You need to log in
before you can comment on or make changes to this bug.
Description
•