path: root/modules/zfssnap/zfssnap.py

#!@python@/bin/python

import csv
import subprocess
import io
from distutils.util import strtobool
from datetime import datetime, timezone, timedelta
from dateutil.tz import gettz, tzutc
import pytimeparse
import argparse
import re

import sys

import logging

import shlex

from collections import defaultdict, OrderedDict, deque, namedtuple

import configparser
from xdg import BaseDirectory

from functools import cache

from math import floor

import asyncio

from dataclasses import dataclass


TIME_PATTERNS = OrderedDict([
    ("secondly", lambda t: t.strftime('%Y-%m-%d %H:%M:%S')),
    ("minutely", lambda t: t.strftime('%Y-%m-%d %H:%M')),
    ("5m", lambda t: (t.strftime('%Y-%m-%d %H'), floor(t.minute / 5) * 5)),
    ("15m", lambda t: (t.strftime('%Y-%m-%d %H'), floor(t.minute / 15) * 15)),
    ("hourly", lambda t: t.strftime('%Y-%m-%d %H')),
    ("4h", lambda t: (t.strftime('%Y-%m-%d'), floor(t.hour / 4) * 4)),
    ("12h", lambda t: (t.strftime('%Y-%m-%d'), floor(t.hour / 12) * 12)),
    ("daily", lambda t: t.strftime('%Y-%m-%d')),
    ("halfweekly", lambda t: (t.strftime('%G-%V'), floor(int(t.strftime('%u')) / 4) * 4)),
    ("weekly", lambda t: t.strftime('%G-%V')),
    ("monthly", lambda t: t.strftime('%Y-%m')),
    ("yearly", lambda t: t.strftime('%Y')),
])

@dataclass(eq=True, order=True, frozen=True)
class Snap:
    name: str
    creation: datetime

@dataclass(eq=True, order=True, frozen=True)
class KeptBecause:
    rule: str
    ix: int
    base: str
    period: str


@cache
def _now():
    return datetime.now(timezone.utc)

def _snap_name(item, time=_now()):
    suffix = re.sub(r'\+00:00$', r'Z', time.isoformat(timespec='seconds'))
    return f'{item}@{suffix}'

def _log_cmd(*args):
    fmt_args = ' '.join(map(shlex.quote, args))
    logger.debug(f'Running command: {fmt_args}')

def _get_items():
    items = {}

    args = ['zfs', 'get', '-H', '-p', '-o', 'name,value', '-t', 'filesystem,volume', '-s', 'local,default,inherited,temporary,received', 'li.yggdrasil:auto-snapshot']
    _log_cmd(*args)
    with subprocess.Popen(args, stdout=subprocess.PIPE) as proc:
        text_stdout = io.TextIOWrapper(proc.stdout)
        reader = csv.DictReader(text_stdout, fieldnames=['name', 'setting'], delimiter='\t', quoting=csv.QUOTE_NONE)
        Row = namedtuple('Row', reader.fieldnames)
        for row in [Row(**data) for data in reader]:
            items[row.name] = bool(strtobool(row.setting))

    return items

def _get_snaps(only_auto=True):
    snapshots = defaultdict(list)
    args = ['zfs', 'list', '-H', '-p', '-t', 'snapshot', '-o', 'name,li.yggdrasil:is-auto-snapshot,creation']
    _log_cmd(*args)
    with subprocess.Popen(args, stdout=subprocess.PIPE) as proc:
        text_stdout = io.TextIOWrapper(proc.stdout)
        reader = csv.DictReader(text_stdout, fieldnames=['name', 'is_auto_snapshot', 'timestamp'], delimiter='\t', quoting=csv.QUOTE_NONE)
        Row = namedtuple('Row', reader.fieldnames)
        for row in [Row(**data) for data in reader]:
            if only_auto and not bool(strtobool(row.is_auto_snapshot)):
                continue

            base_name, _, _ = row.name.rpartition('@')
            creation = datetime.fromtimestamp(int(row.timestamp), timezone.utc)
            snapshots[base_name].append(Snap(name=row.name, creation=creation))

    return snapshots

def prune(config, dry_run, keep_newest, do_exec):
    do_exec = do_exec and 'EXEC' in config
    prune_timezone = config.gettimezone('KEEP', 'timezone', fallback=tzutc())
    logger.debug(f'prune timezone: {prune_timezone}')

    items = _get_snaps()

    exec_candidates = set()
    if do_exec:
        exec_timezone = config.gettimezone('EXEC', 'timezone', fallback=prune_timezone)
        logger.debug(f'exec timezone: {exec_timezone}')

        for rule, pattern in TIME_PATTERNS.items():
            desired_count = config.getint('EXEC', rule, fallback=0)

            for base, snaps in items.items():
                periods = OrderedDict()

                for snap in sorted(snaps, key=lambda snap: snap.creation):
                    period = pattern(snap.creation.astimezone(exec_timezone))
                    if period not in periods:
                        periods[period] = deque()
                    periods[period].append(snap)

                to_exec = desired_count
                ordered_periods = periods.items()
                for period, period_snaps in ordered_periods:
                    if to_exec == 0:
                        break

                    for snap in period_snaps:
                        exec_candidates.add(snap)
                        logger.debug(f'{snap.name} is exec candidate')
                        to_exec -= 1
                        break

                if to_exec > 0:
                    logger.debug(f'Missing {to_exec} to fulfill exec {rule}={desired_count} for ‘{base}’')

        check_cmd = config.get('EXEC', 'check', fallback=None)
        if check_cmd:
            already_execed = set()
            for snap in exec_candidates:
                args = []
                args += shlex.split(check_cmd)
                args += [snap.name]
                _log_cmd(*args)
                check_res = subprocess.run(args)
                if check_res.returncode == 0:
                    already_execed.add(snap)
                    logger.debug(f'{snap.name} already execed')
            exec_candidates -= already_execed

        exec_cmd = config.get('EXEC', 'cmd', fallback=None)
        exec_count = config.getint('EXEC', 'count', fallback=1)
        if exec_cmd:
            execed = set()
            for snap in sorted(exec_candidates, key=lambda snap: snap.creation):
                if len(execed) >= exec_count:
                    logger.debug(f'exc_count of {exec_count} reached')
                    break

                args = []
                args += shlex.split(exec_cmd)
                args += [snap.name]
                _log_cmd(*args)
                subprocess.run(args).check_returncode()
                execed.add(snap)

            exec_candidates -= execed

    kept_count = defaultdict(lambda: defaultdict(lambda: 0))
    kept_because = OrderedDict()
    def keep_because(base, snap, rule, period=None):
        nonlocal kept_count, kept_because
        kept_count[rule][base] += 1
        if snap not in kept_because:
            kept_because[snap] = deque()
        kept_because[snap].append(KeptBecause(rule=rule, ix=kept_count[rule][base], base=base, period=period))

    for candidate in exec_candidates:
        base_name, _, _ = candidate.name.rpartition('@')
        keep_because(base_name, candidate.name, 'exec-candidate')

    within = config.gettimedelta('KEEP', 'within')
    if within > timedelta(seconds=0):
        for base, snaps in items.items():
            time_ref = max(snaps, key=lambda snap: snap.creation, default=None)
            if not time_ref:
                logger.warn(f'Nothing to keep for ‘{base}’')
                continue

            logger.info(f'Using ‘{time_ref.name}’ as time reference for ‘{base}’')
            within_cutoff = time_ref.creation - within

            for snap in snaps:
                if snap.creation >= within_cutoff:
                    keep_because(base, snap.name, 'within')
    else:
        logger.warn('Skipping rule ‘within’ since retention period is zero')

    for rule, pattern in TIME_PATTERNS.items():
        desired_count = config.getint('KEEP', rule, fallback=0)

        for base, snaps in items.items():
            periods = OrderedDict()

            for snap in sorted(snaps, key=lambda snap: snap.creation, reverse=keep_newest):
                period = pattern(snap.creation.astimezone(prune_timezone))
                if period not in periods:
                    periods[period] = deque()
                periods[period].append(snap)

            to_keep = desired_count
            ordered_periods = periods.items() if keep_newest else reversed(periods.items())
            for period, period_snaps in ordered_periods:
                if to_keep == 0:
                    break

                for snap in period_snaps:
                    keep_because(base, snap.name, rule, period=period)
                    to_keep -= 1
                    break

            if to_keep > 0:
                logger.debug(f'Missing {to_keep} to fulfill prune {rule}={desired_count} for ‘{base}’')

    for snap, reasons in kept_because.items():
        reasons_str = ', '.join(map(str, reasons))
        logger.info(f'Keeping ‘{snap}’ because: {reasons_str}')
    all_snaps = {snap.name for _, snaps in items.items() for snap in snaps}
    to_destroy = all_snaps - {*kept_because}
    if not to_destroy:
        logger.info('Nothing to prune')

    for snap in sorted(to_destroy):
        args = ['zfs', 'destroy']
        if dry_run:
            args += ['-n']
        args += [snap]
        _log_cmd(*args)
        subprocess.run(args, check=True)
        if dry_run:
            logger.info(f'Would have pruned ‘{snap}’')
        else:
            logger.info(f'Pruned ‘{snap}’')

def rename(snapshots, destroy=False, set_is_auto=False):
    args = ['zfs', 'get', '-H', '-p', '-o', 'name,value', 'creation', *snapshots]
    _log_cmd(*args)
    renamed_to = set()
    with subprocess.Popen(args, stdout=subprocess.PIPE) as proc:
        text_stdout = io.TextIOWrapper(proc.stdout)
        reader = csv.DictReader(text_stdout, fieldnames=['name', 'timestamp'], delimiter='\t', quoting=csv.QUOTE_NONE)
        Row = namedtuple('Row', reader.fieldnames)
        for row in [Row(**data) for data in reader]:
            creation = datetime.fromtimestamp(int(row.timestamp), timezone.utc)
            base_name, _, _ = row.name.rpartition('@')
            new_name = _snap_name(base_name, time=creation)
            if new_name == row.name:
                logger.debug(f'Not renaming ‘{row.name}’ since name is already correct')
                continue

            if new_name in renamed_to:
                if destroy:
                    logger.warning(f'Destroying ‘{row.name}’ since ‘{new_name}’ was already renamed to')
                    args = ['zfs', 'destroy', row.name]
                    _log_cmd(*args)
                    subprocess.run(args, check=True)
                else:
                    logger.info(f'Skipping ‘{row.name}’ since ‘{new_name}’ was already renamed to')

                continue

            logger.info(f'Renaming ‘{row.name}’ to ‘{new_name}’')
            args = ['zfs', 'rename', row.name, new_name]
            _log_cmd(*args)
            subprocess.run(args, check=True)
            renamed_to.add(new_name)

            if set_is_auto:
                logger.info(f'Setting is-auto-snapshot on ‘{new_name}’')
                args = ['zfs', 'set', 'li.yggdrasil:is-auto-snapshot=true', new_name]
                _log_cmd(*args)
                subprocess.run(args, check=True)

def autosnap():
    items = _get_items()

    all_snap_names = set()
    async def do_snapshot(*snap_items, recursive=False):
        nonlocal items, all_snap_names
        snap_names = {_snap_name(item) for item in snap_items if items[item]}
        if recursive:
            for snap_item in snap_items:
                all_snap_names |= {_snap_name(item) for item in items if item.startswith(snap_item)}
        else:
            all_snap_names |= snap_names

        args = ['zfs', 'snapshot', '-o', 'li.yggdrasil:is-auto-snapshot=true']
        if recursive:
            args += ['-r']
        args += snap_names

        _log_cmd(*args)
        subprocess.run(args, check=True)

    pool_items = defaultdict(set)
    for item in items:
        pool, _, _ = item.partition('/')
        pool_items[pool].add(item)

    tasks = []
    for snap_items in pool_items.values():
        tasks.append(do_snapshot(*snap_items))
    if not tasks:
        logger.warning('No snapshots to create')
    else:
        async def run_tasks():
            await asyncio.gather(*tasks)
        asyncio.run(run_tasks())
    for snap in all_snap_names:
        logger.info(f'Created ‘{snap}’')
    if all_snap_names:
        rename(snapshots=all_snap_names)

def main():
    global logger
    logger = logging.getLogger(__name__)
    console_handler = logging.StreamHandler()
    console_handler.setFormatter( logging.Formatter('[%(levelname)s](%(name)s): %(message)s') )
    if sys.stderr.isatty():
        console_handler.setFormatter( logging.Formatter('%(asctime)s [%(levelname)s](%(name)s): %(message)s') )
    logger.addHandler(console_handler)

    # log uncaught exceptions
    def log_exceptions(type, value, tb):
        global logger

        logger.error(value)
        sys.__excepthook__(type, value, tb) # calls default excepthook

    sys.excepthook = log_exceptions

    parser = argparse.ArgumentParser(prog='zfssnap')
    parser.add_argument('--verbose', '-v', action='count', default=0)
    subparsers = parser.add_subparsers()
    parser.set_defaults(cmd=autosnap)
    rename_parser = subparsers.add_parser('rename')
    rename_parser.add_argument('snapshots', nargs='+')
    rename_parser.add_argument('--destroy', action='store_true', default=False)
    rename_parser.add_argument('--set-is-auto', action='store_true', default=False)
    rename_parser.set_defaults(cmd=rename)
    prune_parser = subparsers.add_parser('prune')
    prune_parser.add_argument('--config', '-c', dest='config_files', nargs='*', default=list())
    prune_parser.add_argument('--dry-run', '-n', action='store_true', default=False)
    prune_parser.add_argument('--keep-newest', action='store_true', default=False)
    prune_parser.add_argument('--no-exec', dest='do_exec', action='store_false', default=True)
    prune_parser.set_defaults(cmd=prune)
    args = parser.parse_args()

    if args.verbose <= 0:
        logger.setLevel(logging.WARNING)
    elif args.verbose <= 1:
        logger.setLevel(logging.INFO)
    else:
        logger.setLevel(logging.DEBUG)

    cmdArgs = {}
    for copy in {'snapshots', 'dry_run', 'destroy', 'keep_newest', 'set_is_auto', 'do_exec'}:
        if copy in vars(args):
            cmdArgs[copy] = vars(args)[copy]
    if 'config_files' in vars(args):
        def convert_timedelta(secs_str):
            secs=pytimeparse.parse(secs_str)
            if secs is None:
                raise ValueError(f'Could not parse timedelta expression ‘{secs_str}’')
            return timedelta(seconds=secs)
        config = configparser.ConfigParser(converters={
            'timedelta': convert_timedelta,
            'timezone': gettz
        })
        search_files = args.config_files if args.config_files else [*BaseDirectory.load_config_paths('zfssnap.ini')]
        read_files = config.read(search_files)

        def format_config_files(files):
            if not files:
                return 'no files'
            return ', '.join(map(lambda file: f'‘{file}’', files))

        if not read_files:
            raise Exception(f'Found no config files. Tried: {format_config_files(search_files)}')

        logger.debug(f'Read following config files: {format_config_files(read_files)}')

        cmdArgs['config'] = config

    args.cmd(**cmdArgs)

if __name__ == '__main__':
    sys.exit(main())