Comments on: 10+ Ways to Crash or Overload MySQL

By: MySQL for Hosting Providers - how do they manage ? | MySQL Performance Blog

MySQL for Hosting Providers - how do they manage ? | MySQL Performance Blog — Fri, 28 Nov 2008 22:24:32 +0000

[...] to really restrict how much resources single user can consume. I have written over a year ago about 10+ ways to crash or overload MySQL and since that people have come to me and suggested more ways to do the [...]

By: 5 kedvenc oldalam, amin hasznos volt kockulni |

5 kedvenc oldalam, amin hasznos volt kockulni | — Thu, 23 Oct 2008 09:45:42 +0000

[...] – http://www.mysqlperformanceblog.com/2007/11/13/10-ways-to-crash-or-overload-mysql/ [...]

By: gigiduru

gigiduru — Tue, 20 Nov 2007 00:43:02 +0000

Given this statement: “Really – there are many ways to crash or otherwise made unavailable server with any MySQL version if you have access to it with normal privileges”, shouldn’t the title be something like “10+ Ways to keep alive MySQL”?

Not to mention, this blog entry would be even shorter than it actually is right now.

By: 10 moyens de crasher MySQL

10 moyens de crasher MySQL — Sun, 18 Nov 2007 16:18:18 +0000

[...] 10+ Ways to Crash or Overload MySQL (0 visite) [...]

By: Pythian Group Blog » Blog Archive » Log Buffer #71: a Carnival of the Vanities for DBAs

Fri, 16 Nov 2007 17:53:16 +0000

[...] the MySQL Performance Blog, Peter Zaitsev has another list: 10+ Ways to Crash or Overload MySQL. Writes Peter, “[There] are many ways to crash or otherwise made unavailable server with any [...]

By: peter

peter — Fri, 16 Nov 2007 17:07:48 +0000

Gleb,

What does BENCHMARK give you comparing (for example) to self join of same table without indexes few times or other cpu hog query ?

By: Gleb Pakharenko

Gleb Pakharenko — Fri, 16 Nov 2007 12:57:37 +0000

There is a famous BENCHMARK() to DOS web-sites

By: peter

peter — Wed, 14 Nov 2007 18:45:17 +0000

OK Jeremy, I agree with you – whatever way you look at it it is either bug or nonsense

By: Jeremy Cole

Jeremy Cole — Wed, 14 Nov 2007 18:38:58 +0000

Actually, to add one further bullet point:

* localhost doesn’t count either, as it’s not really looked up and thus doesn’t have an entry to hostname_cache in order to keep count of errors.

By: Jeremy Cole

Jeremy Cole — Wed, 14 Nov 2007 18:15:54 +0000

Hi Peter,

Yes, it’s strange. Aborted_connects has no relation to max_connect_errors. The only place it’s ever checked is in sql/sql_parse.cc:

848 if (!(specialflag & SPECIAL_NO_RESOLVE))
849 {
…
861 if (connect_errors > max_connect_errors)
862 return(ER_HOST_IS_BLOCKED);
863 }

Where connect_errors comes from:

851 thd->main_security_ctx.host=
852 ip_to_hostname(&thd->remote.sin_addr, &connect_errors);

There are a couple of things to consider with this:
* Since inc_host_errors is never called on password failure, bad password attempts don’t count.
* If you use skip_name_resolve, none of that code is called at all.
* If a host doesn’t have a valid PTR (ip to host mapping), nothing is ever counted.

As you probably know, I’ve been trying to get MySQL to overhaul how it handles hosts for a long time now. My host cache patches are a good start to this:

http://jcole.us/patches/mysql/5.0/host_cache/

Regards,

Jeremy