1
0
Fork 0
mirror of https://github.com/nix-community/home-manager synced 2024-11-23 11:39:46 +01:00
home-manager/modules/systemd-activate.rb
Erik Arvstedt e2414c4a4f
systemd: improve systemd-activate.rb script
- Pass arguments verbatim to the `systemctl` subprocess, obviating the
  need for shell escaping.

- Use open3 for capturing subprocess output.

- Fix printing of commands during dry run.

- Simplify `X-RestartIfChanged` regular expression.

  1. Use \s to match whitespace, \b to match a word boundary.

  2. Rename variable to conform to Ruby's underscore naming
     conventions.

- Remove no-op set operation. Specifically, 'no_restart' and 'to_stop'
  are disjunct since

  1. After reloading the daemon with the new generation, units in
     'to_stop' (i.e. units from the old gen that are missing in the
     new gen) are not registered anymore in the systemd daemon.

  2. Hence, 'systemctl cat' returns no output for these units.

  3. Because this output is needed to detect 'no_restart' units,
     'no_restart' includes no units from 'to_stop'.

  So 'to_stop -= to_restart' is a no-op.

- Only notify about units that would otherwise be restarted. That is,
  exclude units that are started but not restarted.

- Previously, all inactive units, like short-running services, were
  handled as failed units.

  Now systemd activation doesn't fail for oneshot services like
  'setxkbmap' while 'servicesStartTimeoutMs' is set.

- Don't start unchanged oneshot services.

PR #1110
2020-03-24 00:00:44 +01:00

216 lines
6.6 KiB
Ruby

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 --user 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?
all_services = get_active_targets_units(new_units_path)
maybe_changed = all_services & old_services
changed_services = get_changed_services(old_units_path, new_units_path, maybe_changed)
unchanged_oneshots = get_oneshot_services(maybe_changed - changed_services)
# These services should be running when this script is finished
services_to_run = all_services - unchanged_oneshots
# 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 = changed_services
to_start = get_inactive_units(services_to_run - to_restart)
raise "daemon-reload failed" unless run_cmd('systemctl', '--user', 'daemon-reload')
# Exclude units that shouldn't be (re)started or stopped
no_manual_start, no_manual_stop, no_restart = get_restricted_units(to_stop + to_restart + to_start)
notify_skipped_units(to_restart & no_restart)
to_stop -= no_manual_stop
to_restart -= no_manual_stop + no_manual_start + no_restart
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_action('stop', to_stop)
systemctl_action('start', to_start)
systemctl_action('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,socket,timer}'] }
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 units wanted by active targets
def get_active_targets_units(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)
active_units = active_targets.map do |target|
get_service_files(File.join(units_dir, "#{target}.wants"))
end.flatten
Set.new(active_units)
end
# @return true on success
def run_cmd(*cmd)
print_cmd cmd
@dry_run || system(*cmd)
end
def systemctl_action(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.join(' ')
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 systemctl(*cmd)
output, _ = Open3.capture2('systemctl', '--user', *cmd)
output
end
def print_cmd(cmd)
puts [*cmd].join(' ') if @verbose || @dry_run
end
def get_active_units(units)
filter_units(units) { |state| state == 'active' }
end
def get_inactive_units(units)
filter_units(units) { |state| state != 'active' }
end
def get_failed_units(units)
filter_units(units) { |state| state == 'failed' }
end
def filter_units(units)
return [] if units.empty?
states = systemctl('is-active', *units).split
units.select.with_index { |_, i| yield states[i] }
end
def get_oneshot_services(units)
return [] if units.empty?
types = systemctl('show', '-p', 'Type', *units).split
units.select.with_index do |_, i|
types[i] == 'Type=oneshot'
end
end
def get_restricted_units(units)
infos = systemctl('show', '-p', 'RefuseManualStart', '-p', 'RefuseManualStop', *units)
.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
# Get units that should not be restarted even if a change has been detected.
no_restart_regexp = /^\s*X-RestartIfChanged\s*=\s*false\b/
no_restart = units.select { |unit| systemctl('cat', unit) =~ no_restart_regexp }
[no_manual_start, no_manual_stop, no_restart]
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_failed_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
def notify_skipped_units(no_restart)
puts "Not restarting: #{no_restart.join(' ')}" unless no_restart.empty?
end
setup_services(*ARGV)