From 8759a5a63e0c54e56955b3bcbfaf28dcdc7ab116 Mon Sep 17 00:00:00 2001
From: Cornelius Mika <cornelius.mika@gmail.com>
Date: Mon, 4 Dec 2017 11:17:46 +0100
Subject: [PATCH] systemd: add option to automatically start services

---
 modules/misc/news.nix       |  12 +++
 modules/systemd-activate.rb | 189 ++++++++++++++++++++++++++++++++++++
 modules/systemd.nix         |  32 +++++-
 3 files changed, 231 insertions(+), 2 deletions(-)
 create mode 100644 modules/systemd-activate.rb

diff --git a/modules/misc/news.nix b/modules/misc/news.nix
index 4dfedd39b..c8726da2b 100644
--- a/modules/misc/news.nix
+++ b/modules/misc/news.nix
@@ -493,6 +493,18 @@ in
           A new module is available: 'services.parcellite'
         '';
       }
+
+      {
+        time = "2017-12-11T17:23:12+00:00";
+        condition = config.home.activation ? reloadSystemD;
+        message = ''
+          The Boolean option 'systemd.user.startServices' is now
+          available. When enabled the current naive systemd unit
+          reload logic is replaced by a more sophisticated one that
+          attempts to automatically start, stop, and restart units as
+          necessary.
+        '';
+      }
     ];
   };
 }
diff --git a/modules/systemd-activate.rb b/modules/systemd-activate.rb
new file mode 100644
index 000000000..2708280a1
--- /dev/null
+++ b/modules/systemd-activate.rb
@@ -0,0 +1,189 @@
+require 'set'
+require 'open3'
+
+@dry_run = ENV['DRY_RUN']
+@verbose = ENV['VERBOSE']
+
+UnitsDir = 'home-files/.config/systemd/user'
+
+# 1. Stop all services from the old generation that are not present in the new generation.
+# 2. Ensure all services from the new generation that are wanted by active targets are running:
+#    - Start services that are not already running.
+#    - Restart services whose unit config files have changed between generations.
+# 3. If any services were (re)started, wait 'start_timeout_ms' and report services
+#    that failed to start. This helps debugging quickly failing services.
+#
+# Whenever service failures are detected, show the output of
+# 'systemd --home status' for the affected services.
+#
+def setup_services(old_gen_path, new_gen_path, start_timeout_ms_string)
+  start_timeout_ms = start_timeout_ms_string.to_i
+
+  old_units_path = File.join(old_gen_path, UnitsDir) unless old_gen_path.empty?
+  new_units_path = File.join(new_gen_path, UnitsDir)
+
+  old_services = get_services(old_units_path)
+  new_services = get_services(new_units_path)
+
+  exit if old_services.empty? && new_services.empty?
+
+  services_to_run = get_services_to_run(new_units_path)
+  maybe_changed_services = services_to_run & old_services
+
+  # Only stop active services, otherwise we might get a 'service not loaded' error
+  # for inactive services that were removed in the current generation.
+  to_stop = get_active_units(old_services - new_services)
+  to_restart = get_changed_services(old_units_path, new_units_path, maybe_changed_services)
+  to_start = get_inactive_units(services_to_run - to_restart)
+
+  raise "daemon-reload failed" unless run_cmd('systemctl --user daemon-reload')
+
+  # Exclude services that aren't allowed to be manually started or stopped
+  no_manual_start, no_manual_stop = get_restricted_units(to_stop + to_restart + to_start)
+  to_stop -= no_manual_stop
+  to_restart -= no_manual_stop + no_manual_start
+  to_start -= no_manual_start
+
+  if to_stop.empty? && to_start.empty? && to_restart.empty?
+    print_service_msg("All services are already running", services_to_run)
+  else
+    puts "Setting up services" if @verbose
+    systemctl('stop', to_stop)
+    systemctl('start', to_start)
+    systemctl('restart', to_restart)
+    started_services = to_start + to_restart
+    if start_timeout_ms > 0 && !started_services.empty? && !@dry_run
+      failed = wait_and_get_failed_services(started_services, start_timeout_ms)
+      if failed.empty?
+        print_service_msg("All services are running", services_to_run)
+      else
+        puts
+        puts "Error. These services failed to start:", failed
+        show_failed_services_status(failed)
+        exit 1
+      end
+    end
+  end
+end
+
+def get_services(dir)
+  services = get_service_files(dir) if dir && Dir.exists?(dir)
+  Set.new(services)
+end
+
+def get_service_files(dir)
+  Dir.chdir(dir) { Dir['*.service'] }
+end
+
+def get_changed_services(dir_a, dir_b, services)
+  services.select do |service|
+    a = File.join(dir_a, service)
+    b = File.join(dir_b, service)
+    (File.size(a) != File.size(b)) || (File.read(a) != File.read(b))
+  end
+end
+
+TargetDirRegexp = /^(.*\.target)\.wants$/
+
+# @return all services wanted by active targets
+def get_services_to_run(units_dir)
+  return Set.new unless Dir.exists?(units_dir)
+  targets = Dir.entries(units_dir).map { |entry| entry[TargetDirRegexp, 1] }.compact
+  active_targets = get_active_units(targets)
+  services_to_run = active_targets.map do |target|
+    get_service_files(File.join(units_dir, "#{target}.wants"))
+  end.flatten
+  Set.new(services_to_run)
+end
+
+# @return true on success
+def run_cmd(cmd)
+  print_cmd cmd
+  @dry_run || system(cmd)
+end
+
+def systemctl(cmd, services)
+  return if services.empty?
+
+  verb = (cmd == 'stop') ? 'Stopping' : "#{cmd.capitalize}ing"
+  puts "#{verb}: #{services.join(' ')}"
+
+  cmd = ['systemctl', '--user', cmd, *services]
+  if @dry_run
+    puts cmd
+    return
+  end
+
+  output, status = Open3.capture2e(*cmd)
+  print output
+  # Show status for failed services
+  unless status.success?
+    # Due to a bug in systemd, the '--user' argument is not always provided
+    output.scan(/systemctl (?:--user )?(status .*?)['"]/).flatten.each do |status_cmd|
+      puts
+      run_cmd("systemctl --user #{status_cmd}")
+    end
+    exit 1
+  end
+end
+
+def print_cmd(cmd)
+  puts cmd if @verbose || @dry_run
+end
+
+def get_active_units(units)
+  get_units_by_activity(units, true)
+end
+
+def get_inactive_units(units)
+  get_units_by_activity(units, false)
+end
+
+def get_units_by_activity(units, active)
+  return [] if units.empty?
+  units = units.to_a
+  is_active = `systemctl --user is-active #{units.join(' ')}`.split
+  units.select.with_index do |_, i|
+    (is_active[i] == 'active') == active
+  end
+end
+
+def get_restricted_units(units)
+  infos = `systemctl --user show -p RefuseManualStart -p RefuseManualStop #{units.to_a.join(' ')}`
+          .split("\n\n")
+  no_manual_start = []
+  no_manual_stop = []
+  infos.zip(units).each do |info, unit|
+    no_start, no_stop = info.split("\n")
+    no_manual_start << unit if no_start.end_with?('yes')
+    no_manual_stop << unit if no_stop.end_with?('yes')
+  end
+  [no_manual_start, no_manual_stop]
+end
+
+def wait_and_get_failed_services(services, start_timeout_ms)
+  puts "Waiting #{start_timeout_ms} ms for services to fail"
+  # Force the previous message to always be visible before sleeping
+  STDOUT.flush
+  sleep(start_timeout_ms / 1000.0)
+  get_inactive_units(services)
+end
+
+def show_failed_services_status(services)
+  puts
+  services.each do |service|
+    run_cmd("systemctl --user status #{service}")
+    puts
+  end
+end
+
+def print_service_msg(msg, services)
+  return if services.empty?
+  if @verbose
+    puts "#{msg}:", services.to_a
+  else
+    puts msg
+  end
+end
+
+setup_services(*ARGV)
diff --git a/modules/systemd.nix b/modules/systemd.nix
index 95026d985..9e0c04d25 100644
--- a/modules/systemd.nix
+++ b/modules/systemd.nix
@@ -42,6 +42,8 @@ let
   buildServices = style: serviceCfgs:
     concatLists (mapAttrsToList (buildService style) serviceCfgs);
 
+  servicesStartTimeoutMs = builtins.toString cfg.servicesStartTimeoutMs;
+
 in
 
 {
@@ -83,6 +85,25 @@ in
         type = types.attrs;
         description = "Definition of systemd per-user timers";
       };
+
+      startServices = mkOption {
+        default = false;
+        type = types.bool;
+        description = ''
+          Start all services that are wanted by active targets.
+          Additionally, stop obsolete services from the previous
+          generation.
+        '';
+      };
+
+      servicesStartTimeoutMs = mkOption {
+        default = 0;
+        type = types.int;
+        description = ''
+          How long to wait for started services to fail until their
+          start is considered successful.
+        '';
+      };
     };
   };
 
@@ -118,8 +139,15 @@ in
           (buildServices "timer" cfg.timers)
         );
 
-      home.activation.reloadSystemD = dagEntryAfter ["linkGeneration"]
-        (import ./systemd-activate.nix cfg.systemctlPath);
+      home.activation.reloadSystemD = dagEntryAfter ["linkGeneration"] (
+        if cfg.startServices then
+          ''
+            PATH=${dirOf cfg.systemctlPath} \
+              ${pkgs.ruby}/bin/ruby ${./systemd-activate.rb} \
+                "''${oldGenPath=}" "$newGenPath" "${servicesStartTimeoutMs}"
+          ''
+        else import ./systemd-activate.nix cfg.systemctlPath
+      );
     })
   ];
 }