vidhar: zfssnap...

3 files changed, 387 insertions, 60 deletions

diff --git a/hosts/vidhar/zfs.nix b/hosts/vidhar/zfs.nix
index f5a0383a..bee4e88a 100644
--- a/hosts/vidhar/zfs.nix
+++ b/hosts/vidhar/zfs.nix
@@ -1,32 +1,5 @@
 { pkgs, lib, config, ... }:
-let
-  snapshotNames = ["frequent" "hourly" "daily" "monthly" "yearly"];
-  snapshotCount = {
-    frequent = 24;
-    hourly = 24;
-    daily = 30;
-    monthly = 12;
-    yearly = 5;
-  };
-  snapshotTimerConfig = {
-    frequent = { OnCalendar = "*:0/5 UTC"; Persistent = true; };
-    hourly = { OnCalendar = "hourly UTC"; Persistent = true; };
-    daily = { OnCalendar = "daily UTC"; Persistent = true; };
-    monthly = { OnCalendar = "monthly UTC"; Persistent = true; };
-    yearly = { OnCalendar = "yearly UTC"; Persistent = true; };
-  };
-  snapshotDescr = {
-    frequent = "few minutes";
-    hourly = "hour";
-    daily = "day";
-    monthly = "month";
-    yearly = "year";
-  };
-
-  zfs = config.boot.zfs.package;
-
-  autosnapPackage = pkgs.zfstools.override { inherit zfs; };
-in {
+{
   config = {
     fileSystems = {
       "/boot" =
@@ -136,38 +109,7 @@ in {
       echo "=== ZPOOL IMPORT COMPLETE ==="
     '';
 
-    systemd.services =
-      let mkSnapService = snapName: {
-            name = "zfs-snapshot-${snapName}";
-            value = {
-              description = "ZFS auto-snapshot every ${snapshotDescr.${snapName}}";
-              after = [ "zfs-import.target" ];
-              serviceConfig = {
-                Type = "oneshot";
-                ExecStart = "${autosnapPackage}/bin/zfs-auto-snapshot -k -p -u ${snapName} ${toString snapshotCount.${snapName}}";
-              };
-              restartIfChanged = false;
-
-              preStart = ''
-                ${zfs}/bin/zfs set com.sun:auto-snapshot=true hdd-raid6/safe
-                ${zfs}/bin/zfs set com.sun:auto-snapshot=false hdd-raid6/safe/home/mherold/eos/base
-                ${zfs}/bin/zfs set com.sun:auto-snapshot=true ssd-raid1/safe
-                ${zfs}/bin/zfs set com.sun:auto-snapshot=true boot
-              '';
-            };
-          };
-      in builtins.listToAttrs (map mkSnapService snapshotNames);
-
-    systemd.timers =
-      let mkSnapTimer = snapName: {
-            name = "zfs-snapshot-${snapName}";
-            value = {
-              wantedBy = [ "timers.target" ];
-              timerConfig = snapshotTimerConfig.${snapName};
-            };
-          };
-      in builtins.listToAttrs (map mkSnapTimer snapshotNames);
-
+    services.zfssnap.enable = true;
     services.zfs.trim.enable = false;
     services.zfs.autoScrub = {
       enable = true;
diff --git a/modules/zfssnap/default.nix b/modules/zfssnap/default.nix
new file mode 100644
index 00000000..a0590c9f
--- /dev/null
+++ b/modules/zfssnap/default.nix
@@ -0,0 +1,95 @@
+{ config, pkgs, lib, ... }:
+
+with lib;
+  
+let
+  zfssnap = pkgs.stdenv.mkDerivation rec {
+    name = "zfssnap";
+    src = ./zfssnap.py;
+
+    phases = [ "buildPhase" "checkPhase" "installPhase" ];
+
+    buildInputs = with pkgs; [makeWrapper];
+
+    python = pkgs.python39.withPackages (ps: with ps; [pyxdg pytimeparse dateutil systemd]);
+
+    buildPhase = ''
+      substitute $src zfssnap \
+        --subst-var-by python ${escapeShellArg python}
+    '';
+
+    doCheck = true;
+    checkPhase = ''
+      ${python}/bin/python -m py_compile zfssnap
+    '';
+
+    installPhase = ''
+      install -m 0755 -D -t $out/bin \
+        zfssnap
+
+      wrapProgram $out/bin/zfssnap \
+        --prefix PATH : ${makeBinPath [config.boot.zfs.package]}
+    '';
+  };
+
+  cfg = config.services.zfssnap;
+in {
+  options = {
+    services.zfssnap = {
+      enable = mkEnableOption "zfssnap service";
+      
+      config = mkOption {
+        type = with types; attrsOf (attrsOf str);
+        default = {
+          keep = {
+            within = "5m";
+            "5m" = "24";
+            hourly = "24";
+            daily = "31";
+            monthly = "24";
+            yearly = "-1";
+          };
+        };
+      };
+
+      snapInterval = mkOption {
+        type = types.str;
+        default = "*-*-* *:00/5:00 Europe/Berlin";
+      };
+    };
+  };
+
+  config = mkIf cfg.enable {
+    systemd.services."zfssnap" = {
+      description = "Create automatic ZFS snapshots";
+      after = [ "zfs-import.target" ];
+      wants = [ "zfssnap-prune.service" ];
+      before = [ "zfssnap-prune.service" ];
+      serviceConfig = {
+        Type = "oneshot";
+        ExecStart = "${zfssnap}/bin/zfssnap --no-stderr -vv";
+      };
+    };
+    systemd.services."zfssnap-prune" = {
+      description = "Prune automatic ZFS snapshots";
+      after = [ "zfs-import.target" "zfssnap.service" ];
+      serviceConfig = {
+        Type = "oneshot";
+        ExecStart = let
+          mkSectionName = name: strings.escape [ "[" "]" ] (strings.toUpper name);
+          zfssnapConfig = generators.toINI { inherit mkSectionName; } cfg.config;
+        in "${zfssnap}/bin/zfssnap --no-stderr -vv prune --config=${zfssnapConfig}";
+      };
+    };
+
+    systemd.timers."zfssnap" = {
+      wantedBy = ["timers.target"];
+      timerConfig = {
+        OnCalendar = cfg.snapInterval;
+        Persistent = true;
+      };
+    };
+
+    environment.systemPackages = [zfssnap];
+  };
+}
diff --git a/modules/zfssnap/zfssnap.py b/modules/zfssnap/zfssnap.py
new file mode 100644
index 00000000..86690127
--- /dev/null
+++ b/modules/zfssnap/zfssnap.py
@@ -0,0 +1,290 @@
+#!@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, tzlocal
+import pytimeparse
+import argparse
+import re
+
+import sys
+
+import logging
+
+import shlex
+
+from collections import defaultdict, OrderedDict
+
+import configparser
+from xdg import BaseDirectory
+
+from functools import cache
+
+from math import floor
+
+from systemd import journal
+
+
+@cache
+def _now():
+    return datetime.now(timezone.utc)
+
+def _snap_name(item, time=_now()):
+    suffix = re.sub(r'\+00:00$', r'Z', time.isoformat())
+    return f'{item}@auto_{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.reader(text_stdout, delimiter='\t', quoting=csv.QUOTE_NONE)
+        for row in reader:
+            name = row[0]
+            setting = bool(strtobool(row[1]))
+            items[name] = setting
+
+    return items
+        
+def prune(config, dry_run):
+    
+    items = defaultdict(list)
+
+    args = ['zfs', 'get', '-H', '-p', '-o', 'name,value', '-t', 'snapshot', 'creation']
+    _log_cmd(*args)
+    with subprocess.Popen(args, stdout=subprocess.PIPE) as proc:
+        text_stdout = io.TextIOWrapper(proc.stdout)
+        reader = csv.reader(text_stdout, delimiter='\t', quoting=csv.QUOTE_NONE)
+        for row in reader:
+            name = row[0]
+            timestamp = int(row[1])
+            creation = datetime.fromtimestamp(timestamp, timezone.utc)
+            base_name, _, _ = name.rpartition('@')
+            expected_name = _snap_name(base_name, time=creation)
+            if expected_name != name:
+                # logger.debug(f'Skipping ‘{name}’ since it does not conform to naming scheme')
+                continue
+            items[base_name].append({'name': name, 'creation': creation})
+
+    keep = set()
+    kept_count = defaultdict(lambda: defaultdict(lambda: 0))
+    def keep_because(base, snap, rule, period=None):
+        nonlocal kept_count
+        if snap not in keep:
+            kept_count[rule][base] += 1
+            logger.info(f'Keeping ‘{snap}’ because of rule ‘{rule}’ (#{kept_count[rule][base]} for ‘{base}’, period={period})')
+            keep.add(snap)
+
+    within = config.gettimedelta('KEEP', 'within')
+    within_cutoff = _now() - within
+
+    for base, snap in [(base, snap) for base, snaps in items.items() for snap in snaps]:
+        if snap['creation'] >= within_cutoff:
+            keep_because(base, snap['name'], 'within')
+
+    prune_timezone = config.gettimezone('KEEP', 'timezone', fallback=tzlocal)
+
+    PRUNING_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)),
+        ("hourly", lambda t: t.strftime('%Y-%m-%d %H')),
+        ("daily", lambda t: t.strftime('%Y-%m-%d')),
+        ("weekly", lambda t: t.strftime('%G-%V')),
+        ("monthly", lambda t: t.strftime('%Y-%m')),
+        ("yearly", lambda t: t.strftime('%Y')),
+    ])
+
+    for rule, pattern in PRUNING_PATTERNS.items():
+        desired_count = config.getint('KEEP', rule, fallback=0)
+
+        for base, snaps in items.items():
+            last_period = None
+            to_keep = desired_count
+
+            if to_keep == 0:
+                continue
+            
+            for snap in sorted(snaps, key=lambda snap: snap['creation'], reverse=True):
+                if to_keep == 0:
+                    break
+
+                period = pattern(snap['creation'])
+                if period != last_period:
+                    last_period = period
+                    keep_because(base, snap['name'], rule, period=period)
+                    to_keep -= 1
+
+            if to_keep > 0:
+                logger.debug(f'Missing {to_keep} to fulfill {rule}={desired_count} for ‘{base}’')
+
+    all_snaps = {snap['name'] for _, snaps in items.items() for snap in snaps}
+    to_delete = all_snaps - keep
+    if to_delete:
+        logger.info(f'Will prune: %s', ', '.join(map(lambda snap: f'‘{snap}’', to_delete)))
+    else:
+        logger.info('Nothing to prune')
+
+    for snap in to_delete:
+        args = ['zfs', 'destroy']
+        if dry_run:
+            args += ['-n']
+        args += [snap]
+        _log_cmd(*args)
+        subprocess.run(args, check=True)
+    
+def rename(snapshots):
+    args = ['zfs', 'get', '-H', '-p', '-o', 'name,value', 'creation', *snapshots]
+    _log_cmd(*args)
+    with subprocess.Popen(args, stdout=subprocess.PIPE) as proc:
+        text_stdout = io.TextIOWrapper(proc.stdout)
+        reader = csv.reader(text_stdout, delimiter='\t', quoting=csv.QUOTE_NONE)
+        for row in reader:
+            name = row[0]
+            timestamp = int(row[1])
+            creation = datetime.fromtimestamp(timestamp, timezone.utc)
+            base_name, _, _ = name.rpartition('@')
+            new_name = _snap_name(base_name, time=creation)
+            if new_name == name:
+                logger.debug(f'Not renaming ‘{name}’ since name is already correct')
+                continue
+            logger.info(f'Renaming ‘{name}’ to ‘{new_name}’')
+
+            args = ['zfs', 'rename', name, new_name]
+            _log_cmd(*args)
+            subprocess.run(args, check=True)
+
+def autosnap():
+    items = _get_items()
+
+    recursive, single = set(), set()
+
+    for item_name, is_included in items.items():
+        if not is_included:
+            continue
+
+        children = {sub_name for sub_name in items if sub_name.startswith(f'{item_name}/')}
+        is_recursive = all([items[sub_name] for sub_name in children])
+        if is_recursive and children:
+            recursive.add(item_name)
+        else:
+            single.add(item_name)
+
+    for item_name in recursive | single:
+        is_covered = any([item_name.startswith(f'{super_name}/') for super_name in recursive])
+        if is_covered:
+            try:
+                recursive.remove(item_name)
+            except KeyError:
+                pass
+            try:
+                single.remove(item_name)
+            except KeyError:
+                pass
+
+    def do_snapshot(*snap_items, recursive=False):
+        nonlocal items
+        snap_names = {_snap_name(item) for item in snap_items}
+        all_snap_names = None
+        if recursive:
+            all_snap_names = set()
+            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']
+        if recursive:
+            args += ['-r']
+        args += snap_names
+
+        _log_cmd(*args)
+        subprocess.run(args, check=True)
+        rename(snapshots=all_snap_names)
+
+    do_snapshot(*single)
+    do_snapshot(*recursive, recursive=True)
+
+def main():
+    global logger
+    logger = logging.getLogger(__name__)
+    systemd_handler = journal.JournalHandler()
+
+    # 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)
+    parser.add_argument('--no-stderr', dest='stderr', action='store_false', default=True)
+    subparsers = parser.add_subparsers()
+    parser.set_defaults(cmd=autosnap)
+    rename_parser = subparsers.add_parser('rename')
+    rename_parser.add_argument('snapshots', nargs='+')
+    rename_parser.set_defaults(cmd=rename)
+    prune_parser = subparsers.add_parser('prune')
+    prune_parser.add_argument('--config', '-c', dest='config_files', type=argparse.FileType('r'), nargs='*', default=list())
+    prune_parser.add_argument('--dry-run', '-n', action='store_true', default=False)
+    prune_parser.set_defaults(cmd=prune)
+    args = parser.parse_args()
+
+    if args.stderr:
+        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)
+
+    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'}:
+        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)
+
+sys.exit(main())