Description:
There's been a couple of errors packaging up MySQL Community Server for
Solaris 10 ... critical errors, which prevent the system databases etc from
being initialized.
My platform is a Solaris 10, nov06 sparc install, reasonably patched, with
pretty much everything installed. (MySQL 4 SUNW packages were removed)
I am installing the mysql-5.50.51a-solaris10-sparc-64bit.pkg.gz which I
downloaded the other day. This is a typical package stream, with a
post-install script. The pkginfo indicates
PKGINST: mysql
NAME: MySQL Community Server (GPL)
CATEGORY: application
ARCH: sun4u
VERSION: 5.0.51a
BASEDIR: /opt/mysql
VENDOR: MySQL AB
PSTAMP: MySQL AB Build Engineers
INSTDATE: May 06 2008 14:28
EMAIL: build@mysql.com
STATUS: completely installed
FILES: 2319 installed pathnames
84 directories
178 executables
383815 blocks used (approx)
How to repeat:
Okay, so what's wrong? Two things:
1. The post-install script uses the "/bin/sh" shell, and the bang operator
to negate a test (if ! test -x ...) This generates the syntax error "!: not
found" a couple of times. To correct this, there are couple of solutions --
use /bin/bash as the script interpreter, or, you can probably use
/usr/xpg4/bin/sh as the script interpreter. XPG4 supports the bang syntax
but I did not test the whole post install script.
*My Workaround: I temporarily replaced the /bin/sh symlink with a link to
/bin/bash*
2. When the post-install script launches mysqld, it gives up root
permissions and becomes mysql. This is reasonable behaviour, however,
mysqld needs to be able to create files/dirs in /var/tmp/installXXXXXX.
This is not possible, because the post-install script creates that tmpdir
with 0700 perms and root ownership. I _did_ try modifying my shell's umask
and group membership before launching pkgadd.
Specifically, EACCES on open("/var/tmp//installczaqrE/#sql3cf3_1_0.frm",
O_WRONLY_O_CREAT|O_EXCL, 0100660) on a non-primordial thread (thread 3) of
mysqld causes it to close ./mysql/db.frm (which it had just opened), return
error 1004 to stderr, and the LWP exits. As soon as thread id 3 exits,
thread 1 reports it is aborting, kills thread 2, and exits with WEXITSTATUS
== 1. This is a fatal error, and causes the post-install script to abort,
leaving the package in a half-installed state.
*My Workaround: While the package was installing, I chmod 777'd
/var/tmp/install**
Suggested fix:
With my two work-arounds in place, I was able to successfully install MySQL
from the distribution package. However, these workarounds are pretty
unreasonable, and a proper fix should be a piece of cake.
Let me know if you want help testing. I also snagged a truss of a busted
install if you're curious.
Description: There's been a couple of errors packaging up MySQL Community Server for Solaris 10 ... critical errors, which prevent the system databases etc from being initialized. My platform is a Solaris 10, nov06 sparc install, reasonably patched, with pretty much everything installed. (MySQL 4 SUNW packages were removed) I am installing the mysql-5.50.51a-solaris10-sparc-64bit.pkg.gz which I downloaded the other day. This is a typical package stream, with a post-install script. The pkginfo indicates PKGINST: mysql NAME: MySQL Community Server (GPL) CATEGORY: application ARCH: sun4u VERSION: 5.0.51a BASEDIR: /opt/mysql VENDOR: MySQL AB PSTAMP: MySQL AB Build Engineers INSTDATE: May 06 2008 14:28 EMAIL: build@mysql.com STATUS: completely installed FILES: 2319 installed pathnames 84 directories 178 executables 383815 blocks used (approx) How to repeat: Okay, so what's wrong? Two things: 1. The post-install script uses the "/bin/sh" shell, and the bang operator to negate a test (if ! test -x ...) This generates the syntax error "!: not found" a couple of times. To correct this, there are couple of solutions -- use /bin/bash as the script interpreter, or, you can probably use /usr/xpg4/bin/sh as the script interpreter. XPG4 supports the bang syntax but I did not test the whole post install script. *My Workaround: I temporarily replaced the /bin/sh symlink with a link to /bin/bash* 2. When the post-install script launches mysqld, it gives up root permissions and becomes mysql. This is reasonable behaviour, however, mysqld needs to be able to create files/dirs in /var/tmp/installXXXXXX. This is not possible, because the post-install script creates that tmpdir with 0700 perms and root ownership. I _did_ try modifying my shell's umask and group membership before launching pkgadd. Specifically, EACCES on open("/var/tmp//installczaqrE/#sql3cf3_1_0.frm", O_WRONLY_O_CREAT|O_EXCL, 0100660) on a non-primordial thread (thread 3) of mysqld causes it to close ./mysql/db.frm (which it had just opened), return error 1004 to stderr, and the LWP exits. As soon as thread id 3 exits, thread 1 reports it is aborting, kills thread 2, and exits with WEXITSTATUS == 1. This is a fatal error, and causes the post-install script to abort, leaving the package in a half-installed state. *My Workaround: While the package was installing, I chmod 777'd /var/tmp/install** Suggested fix: With my two work-arounds in place, I was able to successfully install MySQL from the distribution package. However, these workarounds are pretty unreasonable, and a proper fix should be a piece of cake. Let me know if you want help testing. I also snagged a truss of a busted install if you're curious.