Random footnotes of Engineering

After my moments of stupidity documented below I decided I'd fix the problem. I've coded up a script that will use zfs send and recv to replicate my home directory. At the moment it's manual, but I'll look into hooking into an automatic system:

#!/bin/bash


USER=fifors

HOST=stormspike.sfbay.sun.com

FS="rpool/export/home"

RS="rpool/zion-home-backup"

MAX_SNAP=15

NOW=`date +%Y%m%d%H%M%S`

INIT=0


set -- `getopt i $*`

if [ $? != 0 ]

then

    echo ${USAGE}

    exit 2

fi


for i in $*

do

    case $i in

	-i) echo Initialising Snapshot backup;

	    INIT=1;

	    shift;

	    continue;;

    esac

done


ALIVE=`ping ${HOST} 2>&1 | grep -c alive`


if [ ${ALIVE} ]

then

	SNAPS=`zfs list -t snapshot |grep "${FS}@" | wc -l | tr -d " "`

	echo Determine incremental

	if [ "${INIT}" == "0" ]

	then

	    ISNAP=`zfs list -t snapshot | grep rpool/export/home@ |tail -1 |cut -d " " -f 1`

	fi


	echo "Creating zfs snapshot ${FS}@${NOW} ..."

	zfs snapshot ${FS}@${NOW} > /dev/null 2> /tmp/error.$$

	if [ $? -ne 0 ]

	then

		echo "Failed ($?)!"

		cat /tmp/error.$$

		rm /tmp/error.$$ 

		exit 1

	else

		echo "Successfully created snapshot: ${FS}@${NOW} !"

	fi

	rm /tmp/error.$$

	if [ ${SNAPS} -ge ${MAX_SNAP} ]

	then

		LAST=`zfs list -t snapshot | grep "${FS}@" | head -1 | awk '{ print $1}'`

		echo "Destroying last snapshot: ${LAST}..."

		zfs destroy ${LAST} > /dev/null 2> /tmp/error.$$

		if [ $? -ne 0 ]

		then

			echo "Failed ($?)!"

			cat /tmp/error.$$

			rm /tmp/error.$$

			exit 1

		else

			echo "Successfully destroyed snapshot: ${LAST} !"

		fi

		rm /tmp/error.$$

	fi


	if [ "${INIT}" == "1" ]

	then

	    echo "Initialising the zfs replication of ${FS}..."

	    zfs send ${FS}@${NOW} | ssh ${USER}@${HOST} pfexec /sbin/zfs recv ${RS}@${NOW}

	else

	    echo "Initiate the zfs send of ${FS}@${NOW} to ${RS}@${NOW}..."

	    zfs send -i ${ISNAP} ${FS}@${NOW} | ssh ${USER}@${HOST} pfexec /sbin/zfs recv -F ${RS}@${NOW}

	fi


	if [ $? -ne 0 ]

	then

		echo "Failed ($?)!"

		exit 1

	else

		echo "Successfully synced snapshot: ${FS}@${NOW} and ${RS}@${NOW} !"

	fi

else

	echo "Cannot connect to ${HOST}"

fi

It's pretty simple really, just relies on the underlying aspects of ZFS to make it all ok.

Grr.. it looks like html destroys the format... I'll fix that later.

Turns out a judicious import from a LiveCD does the right thing. I just needed to pass the correct device string (s0, not p2).

It may turn out to prove that I am stupid in many other ways, but at least in this case it was recoverable ;-).

I always tend to prefer stuff on the bleeding edge warts and all. So I was in Baltimore for PASIG during the week and I suggested to a few folks they should consider using Celeste and a basis for a data store, seems fine right? It worked out of the box for me on OpenSolaris, they had a few problems so I rebooted into Ubuntu to install and try it there. Again no problems.

So whilst in Ubuntu I decided to use zfs-fuse to pull in my mail filters for Thunderbird. Again should be simple enough. I'd had problems before with this causing problems to the underlying zpool, but I thought this would be fine since I just wanted read access. Big mistake.

Ok, so what happens? When I try to reboot the ZFS filesystem isn't recognised by GRUB. Ok that's fine, I've seen it before. Simple solution, boot with the LiveCD import the pool done, right? Nope, unfortunately no LiveCD and I'm also not patient (hence the title of this entry). So what do I do next? Try and fail to create a OpenSolaris Live USB keyring. Then I get the idea that I should update ZFS fuse and see if there is anything I can do.

hg update, scons, sudo scons install. Good to go. Then I notice the ZFS version on the pool is only 10 and ZFS fuse goes all the way to 14. Maybe that's the problem.. This is where the good fairy should have been crying "Don't do it, Don't do it". Unfortunately there was no one around to pry my fingers away from the keyboard and I upgraded the filesystem.

Now ZFS doesn't see the pool at all... Now that I'm back in the office I'll try to do some surgery with zdb, but I'm not hopeful..