ZFS just blew up - Ubuntu 14.04 update to 0.6.6-3 zfsonlinux built

Looks like it was a deliberate change to stop ZFS automatically mounting the pools at system start, so something else (systemd) can do it instead, but that isn't implemented yet. This is going to be a Linux-specific (or possibly Ubuntu-specific) issue.

The ZFSonLinux mailing list and bug tracker came to the rescue...

for some reason my system (and several other folk I summise) hat an /etc/hostid file that contained a null host id (4 zero bytes).

This, since 0.6.3-3 seems to have been taken as a sentinel value instead of a legitimate host ID - so ZFS thought the pool was from a foreign system and did not want to auto mount it.

Setting the host id, doing a zpool export followed by an import and updating (regenerating) the initramfs seems to have fixed it.

Your hunch may be right - someone else mentioned systemd.

Rather a bad breakage...

But ultimately, it did not lose any data (only temporarily).

So it's still one up on ReiserFS.

There was plenty wrong with sysvinit. It was very difficult to get things to start in the right order, and one third party init script could totally mess up the the start order. While it was supposed to take advantage of multiple cores, for parallel starting of things, it didn't do it well. The chkconfig and lsb headers were rarely done properly, and often ended up with a loop in dependencies.

Regards, Dave Hodgins

Looks like systemd is an attempt to reimplement SMF from Solaris. Makes you wonder why they didn't simply use SMF - Sun opensourced it for that purpose, and it's had over 10 years of use, so significant bugs are long since shaken out, and it works very well.

The old wheel didn't have enough corners?

Cheers, Daniel.

License perhaps? That's why ZFS isn't in the kernel.

En el artículo , Tim Watts escribió:

Yes, but CentOS (which is built on the DeadRat Enterprise sources) has traditionally been one or two steps behind bleeding edge, focusing on reliability and stability. We use it for the servers at work.

I was disappointed when CentOS caved in and started using systemd. it goes counter to the Unix philosophy of "do one thing and do it very well." systemd tries to be the Swiss Army knife of *nix init and it shows.

Rubbish! CentOS tries to be as close to RHEL as possible, not lagging behind it ...

En el artículo , Andy Burns escribió:

It's usually one or two minor versions behind RHEL, but what I actually meant was that Fedora is bleeding-edge, CentOS (which is aimed squarely at servers) isn't.

What do you mean by minor? if e.g. you mean CentOS would be at 6.3 when RHEL 6.5 is available, I'd still say "rubbish", if you mean there might be a few days delay getting some centOS security fixes out, sure that can happen, it's the cost of free ...

Certainly you wouldn't /normally/ want Fedora on a server, due to short support lifecycle, and sometimes just not being ready for prime time, it's handy for seeing what is coming down the pipe for future RHEL/CentOS versions though.

En el artículo , Tim Watts escribió:

Aaaaandd... right on cue...

En el artículo , Mike Tomlinson escribió:

Following up to myself, sorry.

The comments are well worth a read too.