What should I do to resovle the flush-hosts problem? If it's hardware we need, then it's hardware we shall have. Is the problem solvable by ram? By CPU?
I'm ready to throw some money at this thing again. Suggestions would be much appreciated.
Of course, it'd also be groovy if everyone who is working on a Phase IV idea might be willing to stop porting, or to port with a STRONG eye towards helping with the performance problem. Clever tweaks to the site to avoid DB queries would probably be *most* helpful, but which queries to avoid?
--Jimbo
Jimmy Wales wrote:
What should I do to resovle the flush-hosts problem? If it's hardware we need, then it's hardware we shall have. Is the problem solvable by ram? By CPU?
I'm ready to throw some money at this thing again. Suggestions would be much appreciated.
Of course, it'd also be groovy if everyone who is working on a Phase IV idea might be willing to stop porting, or to port with a STRONG eye towards helping with the performance problem. Clever tweaks to the site to avoid DB queries would probably be *most* helpful, but which queries to avoid?
--Jimbo
See http://www.mysql.com/doc/en/Blocked_host.html for information about our problem.
I think we should increase max_connect_errors, but I would have liked to have someone else's opinion on that. I don't know much about SQL. All the same, I'm convinced it's a good idea. Here's how you would do it: connect to the DB with a SUPER account, and type:
SET GLOBAL max_connect_errors=1000000;
See http://www.mysql.com/doc/en/SET_OPTION.html to read about what this does.
That sets the variable until next time pliny is restarted. To make it permanent, edit my.cnf to include something like the following:
[mysql] set-variable = max_connect_errors=1000000
I don't know what the underlying problem is. It may be a network infrastructure problem: say, dodgy cables, a malfunctioning network card, something like that. But I have to repeat: I REALLY DON'T KNOW WHAT I'M TALKING ABOUT. There's no harm in setting max_connect_errors to a high value, unless we really are being attacked by crackers, which I doubt. So set the variable and cross your fingers that the problem goes away.
-- Tim Starling
Jimmy Wales wrote:
What should I do to resovle the flush-hosts problem? If it's hardware we need, then it's hardware we shall have. Is the problem solvable by ram? By CPU?
I've no idea what exactly triggers the connection errors, but increasing the max_connect_errors (whice I've just done) should keep it from blocking future access.
I'm ready to throw some money at this thing again. Suggestions would be much appreciated.
I'd highly recommend a second database server, which would feed off the main one using MySQL's replication feature.
* It could run searches and nonessential but nice special pages; or perhaps simply round-robin load sharing
* If the main server crashes again, the wiki can fall back to the second server instead of being inaccessible.
-- brion vibber (brion @ pobox.com)
wikitech-l@lists.wikimedia.org