A snapshot pruning script, similar in behavior to Apple's TimeMachine. Keep hourly snapshots for the last day, daily for the last week, and weekly thereafter. Also prune empty snapshots (0 bytes used) that aren't the most recent for any dataset. Now tracked at https://github.com/fracai/zfs-rollup and https://bitbucket.org/fracai/zfs-rollup

rollup.py

#!/usr/bin/env python

# rollup.py - Arno Hautala <arno@alum.wpi.edu>
#   This work is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.
#   (CC BY-SA-3.0) http://creativecommons.org/licenses/by-sa/3.0/
#
# For the latest version, visit:
#   https://github.com/fracai/zfs-rollup
#   https://bitbucket.org/fracai/zfs-rollup

# A snapshot pruning script, similar in behavior to Apple's TimeMachine
# Keep hourly snapshots for the last day, daily for the last week, and weekly thereafter.
# Also prune empty snapshots (0 bytes used) that aren't the most recent for any dataset

# TODO: 
#   arguments for configuring snapshots to keep (# daily, # weekly, etc)
#   define arbitrary intervals (daily = 86400, quarter-hour = 900, etc)
#   configuration file support
#   different intervals for different dataset roots (? can also just run multiple rollups)
#   rollup based on local time, not UTC
#     requires pytz, or manually determining and converting time offsets

import datetime
import calendar
import subprocess
import argparse
import sys
from collections import defaultdict

parser = argparse.ArgumentParser(description='Prune excess snapshots, keeping hourly for the last day, daily for the last week, and weekly thereafter.')
parser.add_argument('datasets', nargs='+', help='the root dataset(s) from which to prune snapshots')
parser.add_argument('--test', '-t', action="store_true", default=False, help='only display the snapshots that would be deleted, without actually deleting them')
parser.add_argument('--verbose', '-v', action="store_true", default=False, help='display verbose information about which snapshots are kept, pruned, and why')
parser.add_argument('--empty', '-z', action="store_true", default=False, help='prune empty snashots (will still retain the most recent even if empty)')
parser.add_argument('--recursive', '-r', action="store_true", default=False, help='recursively pruning snapshots from nested datasets')

args = parser.parse_args()

if args.test:
    args.verbose = True

now = datetime.datetime.utcnow()

one_hour = datetime.timedelta(hours = 1)
one_day  = datetime.timedelta(days  = 1)
one_week = datetime.timedelta(weeks = 1)

snapshots = defaultdict(lambda : defaultdict(lambda : defaultdict(int)))

for dataset in args.datasets:
    zfs_arguments = "-Hrpo" if args.recursive else "-Hpo"
    zfs_snapshots = subprocess.check_output(["zfs", "get", zfs_arguments, "name,property,value", "creation,used", dataset])

    for snapshot in zfs_snapshots.splitlines():
        name,property,value = snapshot.split('\t',3)

        # enforce that this is a snapshot (presence of '@')
        if "@" not in name:
            continue
        
        dataset,snapshot = name.split('@',2)
        snapshots[dataset][snapshot][property] = value

for dataset in sorted(snapshots.keys()):
    print dataset
    
    sorted_snapshots = sorted(snapshots[dataset].keys())
    most_recent = sorted_snapshots[-1]
    
    hours = []
    days = {}
    weeks = {}
    
    for snapshot in sorted_snapshots:
        # enforce that this is an automated snapshot (presence of 'auto')
        if "auto" not in snapshot:
            print "\tignoring:\t", "@"+snapshot
            continue

        prune = True

        epoch = snapshots[dataset][snapshot]['creation']
        
        snaptime = datetime.datetime.utcfromtimestamp(float(epoch))
        snapdate = snaptime.date()
        snapweek = str(snapdate.isocalendar()[0])+"-"+str(snapdate.isocalendar()[1])

        if (not args.empty or snapshots[dataset][snapshot]['used'] != '0') or snapshot == most_recent:
            if snaptime >= now - one_day:
                hours.append(epoch)
                prune = False
            
            if snaptime >= now - one_week and snapdate not in days:
                days[snapdate] = epoch
                prune = False
            
            if snapweek not in weeks:
                weeks[snapweek] = epoch
                prune = False
        
        if prune or args.verbose:
            print "\t","pruning\t" if prune else " \t", "@"+snapshot, 
            if args.verbose:
                print 'h' if epoch in hours else '-', 
                print 'd' if snapdate in days and days[snapdate] == epoch else '-', 
                print 'w' if snapweek in weeks and weeks[snapweek] == epoch else '-',
                print snapshots[dataset][snapshot]['used']
            else:
                print
        
        if prune:
            if (not args.test):
                # destroy the snapshot
                subprocess.call(["zfs", "destroy", dataset+"@"+snapshot])

Now tracked at https://github.com/fracai/zfs-rollup and https://bitbucket.org/fracai/zfs-rollup