ZFS for production in an ISP?

Is TextDrive nuts? I can't believe I just read this about TextDrive Support:

"And yes, for Solaris we will be using ZFS."

*boggle*

I mean, ZFS is cool and all; but it just got seeded to OpenSolaris developers in build 27a.

Any ISP that says they are going to be using ZFS today needs to be whacked upside the head. For testing? Sure. Long-term? Maybe after they go through a full Solaris 11 release cycle with it. I've heard conflicting reports as to whether zfs will be only in Nevada or if they will ship it as an 'add-on' to 10; my bet is that it only makes it into Nevada, but I could be dead wrong. Who knows.

ZFS today? For customer data? Come on.

[/software/solaris] permanent link

Patching Solaris 10 with zones

Sun now requires all Solaris 10 updates to go through the Update Manager. However, note the following blurb:

The Sun Update Connection cannot currently be used on systems
configured with Zones. This is a result of changes to the
low-level Solaris patch utility, patchadd, that rendered it
incompatible with the Sun Update Connection client software.
Updates for these systems must be obtained from SunSolve and
applied using patchadd.

Oops. Zones are by far the killer feature in Solaris 10.

Sun's argument is that because patchadd changed the result code that it is impossible to script it? How exactly did this change get through the ARC? Were they sleeping that day?

In the Update Connection support forum, they've been promising that a fix will come real soon now - I'm beginning to think they just don't care to fix it and produced this lame excuse.

Luckily, it's not terribly hard to script the patching process with zones installed.

Caveat emptor.

[/software/solaris] permanent link

Recall this patch, Sun!

Sun Solaris/10 Intel kernel patches 118844-18 through 118844-23 blow up Solaris 10/x86. Your machine will become a brick and enter an infinite reboot loop. Lovely.

I just spent the better part of an hour dealing with Sun's idiocy. If you have a bad patch, revoke it. Don't keep allowing poor admins to come along later after you know it's broken and fall into the same trap.

Not happy at all.

Here's the workaround so that you can boot:

[...at the Solaris boot prompt; enable kmdb, debugging, and single-user
    so that you can remove the patch and reboot...]
boot -kds
[...wait for it to boot...]
physmax:w
:c
[...you'll see 'stop on write of'...]
physmax/X
[...you'll see something like the following line:

   physmax: bff7f

 this is a hex number; add one; so if you see bff7f,
 your next line will need to be bff80...]
physxmax/W bff80
:c
[...system will boot and go into single user mode...
    now, go toss those patches...]
patchrm 118844-19 120662-03 118345-12 118376-04 \
  118565-03 118886-01 119076-10 118813-01 \
  118881-02 120082-07 119851-02
shutdown -i6 -y -g0 "sun should test their patches"

[/software/solaris] permanent link

Why this guy hates Apache

On one of the Unix admin lists I'm on, at the end of a long day, this poor guy just about had it:

And this brings me to the next thing, noticing things. Who the hell set that
up? How could you not notice that it takes apache about 30 seconds to serve
the first page, because that's how long it takes to spawn a single child?
Did you ever look at the log files? How about this for noticing: Before I
fixed the configuration, everytime you restart apache it throws this HUGS
ASS ERROR MESSAGE into your face that says "Proceeding with undefined
results." at the end. What were you thinking when you ignored it? What about
the fact that some VirtualHosts were defined twice? Apache doesn't even
complain about that, it just silently chooses one, apparently at random.
Did anyone ever test the configuration? Notice how every directory with a
.htaccess file came back with Internal Server Error?

Worth the chuckle. I've had that feeling before.

The analytical side of me does wonder what specific version they are using, but I gather the hardware is flaky.

Wonder if Rich is accepting submissions to add to his list of things to hate for the lightning talks at ApacheCon next month.

[/software] permanent link

Updated Atom Feed to 1.0

I use Blosxom on this blog and have an Atom feed.

Unfortunately, it seems that everyone else in the universe stopped using Blosxom and moved to something that requires MySQL or PHP which are non-starters for me.

So, I read this Moving from Atom 0.3 to 1.0 post and combined it with Feed Validator to update Blosxom's atom plugin.

My version of atomfeed Blosxom plugin for Atom 1.0 is here. Feel free to have at it.

The only issue is that the Feed Validator complains that the updated tag isn't early enough in the document. Besides being kind of bogus (XML can't define an ordering on the elements), there's just no way to do it easily in Blosxom without going through the stories twice. Close enough.

[/software] permanent link