zfs开源已经13年了，最近开始使用。zfs现在进入freebsd基本配置，使用方便，几条命令起来了，先准备好磁盘ad1
echo 'zfs_enable="YES"' >> /etc/rc.conf /etc/rc.d/zfs start echo 'daily_status_zfs_enable="YES"'>> /etc/periodic.conf zpool create z01 ad1
zfs管理简洁得很！把数据库复制到新建的z01上面，启动数据库。每天做一个快照。查看结果：
df -h | egrep 'usr$|z01$' ; zfs list -t snapshot ; zpool status z01 ; top -no size /dev/ada0s1f 434G 247G 152G 62% /usr z01 433G 54G 379G 12% /z01 NAME USED AVAIL REFER MOUNTPOINT z01/pgsql/data@t20180507 206M - 9.65G - z01/pgsql/data@201805080820 71.0M - 9.56G - z01/pgsql/data@20180509 70.3M - 9.56G - pool: z01 state: ONLINE scan: none requested config: NAME STATE READ WRITE CKSUM z01 ONLINE 0 0 0 ada1 ONLINE 0 0 0 errors: No known data errors last pid: 41091; load averages: 1.55,0.77,0.66 up 0+07:23:44 15:29:10 137 processes: 1 running,132 sleeping,4 zombie Mem: 231M Active,468M Inact,1867M Wired,415M Buf,5364M Free ARC: 964M Total,486M MFU,464M MRU,414K Anon,1995K Header,12M Other 925M Compressed,966M Uncompressed,1.04:1 Ratio Swap: 4096M Total,4096M Free PID USERNAME THR PRI NICE SIZE RES STATE C TIME Wcpu COMMAND 1260 pgsql 1 4 0 1172M 103M select 4 0:32 0.00% postgres 1219 pgsql 1 4 0 1170M 102M select 6 0:33 0.00% postgres 1258 pgsql 1 4 0 1170M 102M select 3 0:24 0.00% postgres 1261 pgsql 1 4 0 1168M 98M select 5 0:11 0.00% postgres 1259 pgsql 1 4 0 1168M 98M select 7 0:11 0.00% postgres 1227 pgsql 1 4 0 1168M 91852K select 2 0:24 0.00% postgres 1245 pgsql 1 4 0 1168M 91500K select 7 0:26 0.00% postgres 1262 pgsql 1 4 0 1168M 75920K select 6 0:07 0.00% postgres 1266 pgsql 1 4 0 1166M 74656K select 2 0:06 0.00% postgres 1236 pgsql 1 4 0 1120M 125M select 4 21:45 0.00% postgres
平时没什么负载的时候zfs使用了900M字节作为管理和cache。
基于时间点的恢复其实是很耗时间的，如果你的基础备份每7天一次， 那么读取XLOG并执行到指定的时间点，再取出当时的数据，可能要半小时；如果你每天备份一次数据文件，而又没有使用zfs，那么备份的空间又占用很多。用zfs每天备份数据文件夹占用很少空间，从上面快照列表可以看到。
我们做一次基于时间点的恢复，克隆20180509凌晨的数据文件快照，数据恢复到2018-05-09 10:00这个时间点：
zfs clone z01/pgsql/data@20180509 z01/pgsql/datatest cd /z01/pgsql/ cd datatest cp ../postgresql.conf.recov_time ./postgresql.conf cp ../recovery.conf.recov_time ./recovery.conf cat ./recovery.conf >restore_command='cp >/z01/pgsql/arch/%f %p || cp >/usr/local/pgsql/arch/%f %p' >recovery_target_timeline='1' >recovery_target_time = '2018-05-09 10:00:00+08' rm -rf postmaster.pid rm -rf pg_wal mkdir pg_wal cp -R /usr/local/pg_wal/* pg_wal/ chown -R pgsql:pgsql ../datatest chmod -R og-rwx ../datatest su pgsql -c "pg_ctl -D ../datatest start"
克隆数据文件夹是秒回的，执行XLOG跨越的时间段不会超过一天，既快又节省空间。后台向前滚动的事务日志：
cd /z01/pgsql/datatest/log tail -f `ls -t | head -n 1` 2018-05-09 02:24:29 CST root@postgres [local] LOG: duration: 29474.253 ms statement: select pg_start_backup('') 2018-05-09 15:34:17 CST @ LOG: database system was interrupted; last known up at 2018-05-09 02:24:29 CST 2018-05-09 15:34:17 CST @ LOG: starting point-in-time recovery to 2018-05-09 10:00:00+08 2018-05-09 15:34:17 CST @ LOG: restored log file "000000010000001900000011" from archive 2018-05-09 15:34:18 CST @ LOG: redo starts at 19/11000028 2018-05-09 15:34:18 CST @ LOG: consistent recovery state reached at 19/1105D218 2018-05-09 15:34:18 CST @ LOG: database system is ready to accept read only connections 2018-05-09 15:34:18 CST @ LOG: restored log file "000000010000001900000012" from archive 2018-05-09 15:34:19 CST @ LOG: restored log file "000000010000001900000013" from archive 2018-05-09 15:34:20 CST @ LOG: restored log file "000000010000001900000014" from archive 2018-05-09 15:34:20 CST @ LOG: restored log file "000000010000001900000015" from archive 2018-05-09 15:34:21 CST @ LOG: restored log file "000000010000001900000016" from archive 2018-05-09 15:34:22 CST @ LOG: restored log file "000000010000001900000017" from archive 2018-05-09 15:34:22 CST @ LOG: restored log file "000000010000001900000018" from archive 2018-05-09 15:34:23 CST @ LOG: restored log file "000000010000001900000019" from archive 2018-05-09 15:34:24 CST @ LOG: recovery stopping before commit of transaction 51331481,time 2018-05-09 10:00:00.043179+08
su pgsql -c "pg_ctl -D ../datatest -mfast stop" zfs destroy z01/pgsql/datatest
