git.scottworley.com Git - nix-profile-gc/blob - modules/profile-gc.nix

   1 { lib, config, pkgs, ... }:
   2 let
   3   inherit (lib) escapeShellArg;
   4   cfg = config.nix.profile-gc;
   5   parse-duration = duration: pkgs.runCommand "duration" { buildInputs = with pkgs; [ systemd ]; } ''
   6     set -euo pipefail
   7     parsed=$(systemd-analyze timespan ${escapeShellArg duration} | awk '$1 == "μs:" { print $2 }')
   8     echo "$parsed" > "$out"
   9   '';
  10 in {
  11   options = {
  12     nix.profile-gc = {
  13       enable = lib.mkEnableOption "Automatic profile garbage collection";
  14       dryRun = lib.mkOption {
  15         description = "Say what would have been deleted rather than actually deleting profiles";
  16         type = lib.types.bool;
  17         default = true;
  18       };
  19       keepLast = lib.mkOption {
  20         description = ''
  21           Number of recent profiles to keep.
  22           This control is similar to nix-env --delete-generation's +5 syntax.
  23         '';
  24         type = lib.types.ints.unsigned;
  25         default = 5;
  26       };
  27       keepLastActive = lib.mkOption {
  28         description = "Number of recent active profiles to keep";
  29         type = lib.types.ints.unsigned;
  30         default = 5;
  31       };
  32       keepLastActiveSystem = lib.mkOption {
  33         description = "Number of recent active system profiles to keep";
  34         type = lib.types.ints.unsigned;
  35         default = 5;
  36       };
  37       keepLastActiveBoot = lib.mkOption {
  38         description = "Number of recent active boot profiles to keep";
  39         type = lib.types.ints.unsigned;
  40         default = 3;
  41       };
  42       activeThreshold = lib.mkOption {
  43         description = ''
  44           A system profile that is active (or is either /run/current-system or /run/booted-system)
  45           for at least this long (of powered-on machine time) is considered 'active' for
  46           the purpose of evaluating the keepLastActive number of profiles.  This mechanism is
  47           intended to preserve profiles that are in some sense stable, that have served us well,
  48           so they don't immediately become gc-elligible when a system hasn't been updated in
  49           awhile (so keepLatest won't protect them) generates a bunch of broken profiles (so
  50           keepLast won't protect them) while trying to get up to date.
  51
  52           This is approximate and has a useful granularity of an hour
  53           (config.systemd.timers.profile-gc-log-active.timerConfig.OnUnitActiveSec).
  54           Do not set less than this.
  55         '';
  56         # We admonish the user "Do not set less than this." and check it at runtime rather
  57         # than verifying this with an assertion now because parsing these durations at
  58         # configuration-time requires import-from-derivation, which we want to avoid.  :(
  59         type = lib.types.str;
  60         default = "5 days";
  61       };
  62       keepLatest = lib.mkOption {
  63         description = ''
  64           Keep all profiles younger than this duration (systemd.time format).
  65           This control is similar to nix-collect-garbage's --delete-older-than.
  66         '';
  67         type = lib.types.str;
  68         default = "6 months";
  69       };
  70       keepFuture = lib.mkOption {
  71         description = "Keep profiles 'ahead' of the current profile (happens after rollback)";
  72         type = lib.types.bool;
  73         default = true;
  74       };
  75       logdir = lib.mkOption {
  76         description = "Where to keep liveness logs";
  77         type = lib.types.str;
  78         default = "/var/log/profile-gc";
  79       };
  80     };
  81   };
  82   config = lib.mkIf cfg.enable {
  83     assertions = [
  84       {
  85         assertion = cfg.enable -> config.nix.gc.automatic;
  86         message = ''nix.profile-gc.enable requires nix.gc.automatic'';
  87       }
  88     ];
  89     systemd.services.nix-gc.serviceConfig.ExecStartPre = pkgs.writeShellScript "nix-profile-gc" ''
  90       set -euo pipefail
  91
  92       if [[ ! -e ${cfg.logdir}/active-system
  93          || ! -e ${cfg.logdir}/active-boot
  94          || ! -e ${cfg.logdir}/active-profiles ]]
  95       then
  96         echo "Liveness logs not found.  Not doing any profile garbage collection." >&2
  97         exit 0
  98       fi
  99
 100       alive_threshold="$(< ${parse-duration cfg.activeThreshold})"
 101       alive_loginterval="$(< ${parse-duration config.systemd.timers.profile-gc-log-active.timerConfig.OnUnitActiveSec})"
 102       if (( alive_threshold < alive_loginterval ));then
 103         echo "Liveness threshold is too low.  Not doing any profile garbage collection." >&2
 104         exit 0
 105       fi
 106
 107       topn() {
 108         ${pkgs.coreutils}/bin/tac "$1" |
 109           ${pkgs.gawk}/bin/awk \
 110             --assign key="$2" \
 111             --assign n="$3" \
 112             --assign threshold="$alive_threshold" \
 113             --assign loginterval="$alive_loginterval" \
 114             '
 115               !key || $1 == key {
 116                 val = key ? $2 : $1
 117                 if (++count[val] == int(threshold/loginterval)) {
 118                   print val
 119                   if (++found == n) {
 120                     exit 0
 121                   }
 122                 }
 123               }
 124             '
 125       }
 126
 127       verbose_topn() {
 128         topn "$@" | tee >(
 129           echo "Keeping the last $3 $2 entries from $1:" >&2
 130           ${pkgs.gawk}/bin/gawk '{ print "  " $0 }' >&2 )
 131       }
 132
 133       declare -A active_targets
 134       while read target;do
 135         active_targets[$target]=1
 136       done < <(
 137         verbose_topn ${cfg.logdir}/active-system "" ${escapeShellArg cfg.keepLastActiveSystem}
 138         verbose_topn ${cfg.logdir}/active-boot   "" ${escapeShellArg cfg.keepLastActiveBoot  }
 139       )
 140
 141       now=$(${pkgs.coreutils}/bin/date +%s)
 142       age_threshold="$(< ${parse-duration cfg.keepLatest})"
 143       while read profile;do
 144         echo "Contemplating profiles for $profile:" >&2
 145         unset active
 146         declare -A active
 147         while read p;do
 148           active[$p]=1
 149         done < <(verbose_topn ${cfg.logdir}/active-profiles "$profile" ${escapeShellArg cfg.keepLastActive})
 150         current=$(${pkgs.coreutils}/bin/readlink "$profile")
 151         currentgen=''${current%-link}
 152         currentgen=''${currentgen##*-}
 153         for p in "$profile"-*-link;do
 154           pgen=''${p%-link}
 155           pgen=''${pgen##*-}
 156           if [[ "$p" != "$profile-$pgen-link" ]];then
 157             echo "(Disregarding unrelated profile $p)" >&2
 158             continue
 159           fi
 160           if [[ "$p" == "$current" ]];then
 161             echo "Keeeping current profile $p" >&2
 162             continue
 163           fi
 164           if [[ "''${active_targets[$(${pkgs.coreutils}/bin/readlink "$p")]:-}" ]];then
 165             echo "Keeeping active system/boot profile $p" >&2
 166             continue
 167           fi
 168           if [[ "''${active[$p]:-}" ]];then
 169             echo "Keeeping active profile $p" >&2
 170             continue
 171           fi
 172           if (( (now - "$(${pkgs.findutils}/bin/find "$p" -printf %Ts)") < age_threshold/1000000 ));then
 173             echo "Keeeping young profile $p" >&2
 174             continue
 175           fi
 176           ${lib.optionalString cfg.keepFuture ''
 177             if (( pgen > currentgen ));then
 178               echo "Keeeping future profile $p" >&2
 179               continue
 180             fi
 181           ''}
 182           ${if cfg.dryRun then ''
 183             echo "Would remove profile $p" >&2
 184           '' else ''
 185             echo "Removing profile $p" >&2
 186             rm "$p"
 187           ''}
 188         done
 189       done < <(${pkgs.findutils}/bin/find ''${NIX_STATE_DIR:-/nix/var/nix}/profiles/ -type l -not -name '*[0-9]-link')
 190     '';
 191     systemd.timers.profile-gc-log-active = {
 192       wantedBy = [ "timers.target" ];
 193       timerConfig.OnUnitActiveSec = "1 hour";
 194     };
 195     systemd.services.profile-gc-log-active = {
 196       description =
 197         "Log the active profiles for gc collection policy evaluation";
 198       serviceConfig.Type = "oneshot";
 199       script = ''
 200         ${pkgs.coreutils}/bin/mkdir -p ${cfg.logdir}
 201         ${pkgs.coreutils}/bin/readlink /run/current-system >> ${cfg.logdir}/active-system
 202         ${pkgs.coreutils}/bin/readlink /run/booted-system  >> ${cfg.logdir}/active-boot
 203         ${pkgs.findutils}/bin/find ''${NIX_STATE_DIR:-/nix/var/nix}/profiles/ \
 204           -type l -not -name '*[0-9]-link' \
 205           -exec ${pkgs.stdenv.shell} -c '
 206             for f;do
 207               echo -n "$f "
 208               ${pkgs.coreutils}/bin/readlink "$f"
 209             done' - {} + \
 210           >> ${cfg.logdir}/active-profiles
 211       '';
 212     };
 213   };
 214 }