openbsd-sparc
[Top] [All Lists]

MySQL 4.0.17/8 on OpenBSD 3.4 SPARC64

To: sparc@openbsd.org
Subject: MySQL 4.0.17/8 on OpenBSD 3.4 SPARC64
From: tonyb@vegan.net
Date: Fri, 9 Apr 2004 19:17:50 -0400 (EDT)
Sender: owner-sparc@openbsd.org
Hi All,

I've got a curious problem with running MySQL 4.0.17 and 4.0.18 on OpenBSD
3.4 SPARC64 on my Sun Ultra 5.

MySQL compiles cleanly and starts up fine.  When I run sql-bench, the
included benchmarking utility, it completes a few tests successfully, then
dies on error during the "create" test, and any subsequent tests fail.

The errors are this:

Can't execute command 'create table onek (unique1 int(5) NOT NULL,unique2
int(4) NOT NULL,two int(4),four int(4),ten int(4),twenty int(4),hundred
int(4) NOT NULL,thousand int(4),twothousand int(4),fivethous
int(4),tenthous int(4),odd int(4),even int(4),stringu1 char(16),stringu2
char(16),string4 char(16),UNIQUE (unique1),UNIQUE (unique2),INDEX hundred1
(hundred))'
Error: Can't create/write to file './test/onek.frm' (Errcode: 9)

(The MySQL user has the ability to write to that directory with no
problem)

or this:

Accessing tables
Can't find file: './test/bench_60.frm' (errno: 9) at ./test-create line
137

(The file is, in fact, there)

Initially, I thought it might be a permissions problem.  But the mysql
users and groups are setup correctly, and the files are there.  I even (in
a very un-OpenBSD-esque manner) chmod'd the MySQL data directory to 777,
yet still the problem persists.  MySQL couldn't find files that were
there in some cases and couldn't write to the directory in others.

The file system has plenty of space and inodes available, so that isn't
the issue either.

Could this be an issue where there are too many files open?  Or perhaps
some odd MySQL bug with OpenBSD (these test for the same version of MySQL
run fine under Solaris, Linux, and FreeBSD on my Ultra 5).

I tried changing some sysctl variables, but the problem still persisted.

Specifically:

  sysctl -w kern.maxfiles=11000
  sysctl -w kern.maxproc=6000

I'm not familiar enough with OpenBSD sysctls to know if that would or
wouldn't be effective.  I boosted them off the prep list for this:
http://bulk.fefe.de/scalability/

Since the tests start OK, and several complete fine, that is what led me
to believe it might be a too-many-files open issue.

Google only provided on instance of someone having a similar problem:

http://vbmysql.com/forums/index.php?t=tree&th=16988&rid=0

I believe his problem is identical to mine.

Any thoughts?

Cheers,

Tony

<Prev in Thread] Current Thread [Next in Thread>