Add support for swift-drive-audit cron job

This change introduces the new class to configure cron job to run
the swift-drive-audit CLI tool periodically. The CLI tool checks
the log files to detect any device failure and umount faulty disks.

Change-Id: I8aded9b59939e5a2eab04424277d9082cbda0a3a
This commit is contained in:
Takashi Kajinami 2022-12-09 15:22:19 +09:00
parent 928a0e467b
commit fd38b9cbfa
3 changed files with 286 additions and 0 deletions

View File

@ -0,0 +1,189 @@
# == Class swift::storage::drive_audit
#
# Set up swift-drive-audit cron job
#
# == Parameters
#
# [*user*]
# (Optional) User with access to swift files.
# Defaults to $::swift::params::user.
#
# [*minute*]
# (Optional) Defaults to '1'.
#
# [*hour*]
# (Optional) Defaults to '0'.
#
# [*monthday*]
# (Optional) Defaults to '*'.
#
# [*month*]
# (Optional) Defaults to '*'.
#
# [*weekday*]
# (Optional) Defaults to '*'.
#
# [*maxdelay*]
# (Optional) In Seconds. Should be a positive integer.
# Induces a random delay before running the cronjob to avoid running
# all cron jobs at the same time on all hosts this job is configured.
# Defaults to 0.
#
# [*log_facility*]
# (Optional) Syslog log facility.
# Defaults to 'LOG_LOCAL2'.
#
# [*log_level*]
# (Optional) Logging level.
# Defaults to 'INFO'.
#
# [*log_address*]
# (Optional) Location where syslog sends the logs to.
# Defaults to '/dev/log'.
#
# [*log_name*]
# (Optional) Label used when logging.
# Defaults to 'drive-audit'.
#
# [*log_udp_host*]
# (Optional) If not set, the UDP receiver for syslog is disabled.
# Defaults to undef.
#
# [*log_udp_port*]
# (Optional) Port value for UDP receiver, if enabled.
# Defaults to undef.
#
# [*device_dir*]
# (Optional) Directory devices are mounted under
# Defaults to $::os_service_default.
#
# [*minutes*]
# (Optional) Number of minutes to look back in the log file.
# Defaults to $::os_service_default.
#
# [*error_limit*]
# (Optional) Number of errors to find before a device is unmounted
# Defaults to $::os_service_default.
#
# [*recon_cache_path*]
# (Optional) The path for recon cache
# Defaults to $::os_service_default.
#
# [*log_file_pattern*]
# (Optional) Location of the log file with globbing pattern to check against
# device errors.
# Defaults to $::os_service_default.
#
# [*log_file_encoding*]
# (Optional) The encoding used to interpret the log files.
# Defaults to $::os_service_default.
#
# [*log_to_console*]
# (Optional) Make drive-audit log to console in addition to syslog
# Defaults to $::os_service_default.
#
# [*unmount_failed_device*]
# (Optional) Unmount the device with errors detected.
# Defaults to $::os_service_default.
#
# [*regex_pattern*]
# (Optional) Regular expression patterns to be used to locate device blocks
# with errors in the log file.
# Defaults to $::os_service_default.
#
# [*purge_config*]
# (Optional) Whether to set only the specified config options in the drive
# audit config.
# Defaults to false.
#
class swift::storage::drive_audit(
# cron options
$user = $::swift::params::user,
$minute = 1,
$hour = 0,
$monthday = '*',
$month = '*',
$weekday = '*',
$maxdelay = 0,
# drive-audit.conf options
$log_facility = 'LOG_LOCAL2',
$log_level = 'INFO',
$log_address = '/dev/log',
$log_name = 'drive-audit',
$log_udp_host = undef,
$log_udp_port = undef,
$device_dir = '/srv/node',
$minutes = $::os_service_default,
$error_limit = $::os_service_default,
$recon_cache_path = $::os_service_default,
$log_file_pattern = $::os_service_default,
$log_file_encoding = $::os_service_default,
$log_to_console = $::os_service_default,
$unmount_failed_device = $::os_service_default,
$regex_pattern = {},
$purge_config = false,
) inherits swift::params {
include swift::deps
validate_legacy(Hash, 'validate_hash', $regex_pattern)
resources { 'swift_drive_audit_config':
purge => $purge_config,
}
swift_drive_audit_config {
'drive-audit/log_name' : value => $log_name;
'drive-audit/log_facility': value => $log_facility;
'drive-audit/log_level' : value => $log_level;
'drive-audit/log_address' : value => $log_address;
}
if $log_udp_host {
swift_drive_audit_config {
'drive-audit/log_udp_host': value => $log_udp_host;
'drive-audit/log_udp_port': value => pick($log_udp_port, $::os_service_default);
}
} else {
swift_drive_audit_config {
'drive-audit/log_udp_host': value => $::os_service_default;
'drive-audit/log_udp_port': value => $::os_service_default;
}
}
swift_drive_audit_config {
'drive-audit/user' : value => $user;
'drive-audit/device_dir' : value => $device_dir;
'drive-audit/minutes' : value => $minutes;
'drive-audit/error_limit' : value => $error_limit;
'drive-audit/recon_cache_path' : value => $recon_cache_path;
'drive-audit/log_file_pattern' : value => $log_file_pattern;
'drive-audit/log_file_encoding' : value => $log_file_encoding;
'drive-audit/log_to_console' : value => $log_to_console;
'drive-audit/unmount_failed_device': value => $unmount_failed_device;
}
$regex_pattern.each | $number, $regex | {
swift_drive_audit_config {
"drive-audit/regex_pattern_${number}": value => $regex;
}
}
if $maxdelay == 0 {
$sleep = ''
} else {
$sleep = "sleep `expr \${RANDOM} \\% ${maxdelay}`; "
}
cron { 'swift-drive-audit':
command => "${sleep}swift-drive-audit /etc/swift/drive-audit.conf",
environment => 'PATH=/bin:/usr/bin:/usr/sbin SHELL=/bin/sh',
user => $user,
minute => $minute,
hour => $hour,
monthday => $monthday,
month => $month,
weekday => $weekday,
require => Anchor['swift::config::end'],
}
}

View File

@ -0,0 +1,4 @@
---
features:
- |
The new ``swift::storage::drive_audit`` class has been added.

View File

@ -0,0 +1,93 @@
require 'spec_helper'
describe 'swift::storage::drive_audit' do
shared_examples 'swift::storage::drive_audit' do
context 'with defaults' do
it 'should configure default values' do
should contain_swift_drive_audit_config('drive-audit/log_name').with_value('drive-audit')
should contain_swift_drive_audit_config('drive-audit/log_facility').with_value('LOG_LOCAL2')
should contain_swift_drive_audit_config('drive-audit/log_level').with_value('INFO')
should contain_swift_drive_audit_config('drive-audit/log_address').with_value('/dev/log')
should contain_swift_drive_audit_config('drive-audit/log_udp_host').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/log_udp_port').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/user').with_value('swift')
should contain_swift_drive_audit_config('drive-audit/device_dir').with_value('/srv/node')
should contain_swift_drive_audit_config('drive-audit/minutes').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/error_limit').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/recon_cache_path').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/log_file_pattern').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/log_file_encoding').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/log_to_console').with_value('<SERVICE DEFAULT>')
should contain_swift_drive_audit_config('drive-audit/unmount_failed_device').with_value('<SERVICE DEFAULT>')
should contain_cron('swift-drive-audit').with(
:command => 'swift-drive-audit /etc/swift/drive-audit.conf',
:environment => 'PATH=/bin:/usr/bin:/usr/sbin SHELL=/bin/sh',
:user => 'swift',
:minute => 1,
:hour => 0,
:monthday => '*',
:month => '*',
:weekday => '*',
)
end
end
context 'with parameters' do
let :params do
{
:maxdelay => 30,
:user => 'alt_swift',
:device_dir => '/opt/swift',
:minutes => 60,
:error_limit => 1,
:recon_cache_path => '/var/cache/swift',
:log_file_pattern => '/var/log/kern.*[!.][!g][!z]',
:log_file_encoding => 'auto',
:log_to_console => true,
:unmount_failed_device => true,
:regex_pattern => {'1' => 'pattern1', '2' => 'pattern2'},
}
end
it 'should configure the given values' do
should contain_swift_drive_audit_config('drive-audit/user').with_value('alt_swift')
should contain_swift_drive_audit_config('drive-audit/device_dir').with_value('/opt/swift')
should contain_swift_drive_audit_config('drive-audit/minutes').with_value(60)
should contain_swift_drive_audit_config('drive-audit/error_limit').with_value(1)
should contain_swift_drive_audit_config('drive-audit/recon_cache_path').with_value('/var/cache/swift')
should contain_swift_drive_audit_config('drive-audit/log_file_pattern').with_value('/var/log/kern.*[!.][!g][!z]')
should contain_swift_drive_audit_config('drive-audit/log_file_encoding').with_value('auto')
should contain_swift_drive_audit_config('drive-audit/log_to_console').with_value(true)
should contain_swift_drive_audit_config('drive-audit/unmount_failed_device').with_value(true)
should contain_swift_drive_audit_config('drive-audit/regex_pattern_1').with_value('pattern1')
should contain_swift_drive_audit_config('drive-audit/regex_pattern_2').with_value('pattern2')
should contain_cron('swift-drive-audit').with(
:command => 'sleep `expr ${RANDOM} \\% 30`; swift-drive-audit /etc/swift/drive-audit.conf',
:environment => 'PATH=/bin:/usr/bin:/usr/sbin SHELL=/bin/sh',
:user => 'alt_swift',
:minute => 1,
:hour => 0,
:monthday => '*',
:month => '*',
:weekday => '*',
)
end
end
end
on_supported_os({
:supported_os => OSDefaults.get_supported_os
}).each do |os,facts|
context "on #{os}" do
let (:facts) do
facts.merge(OSDefaults.get_facts())
end
it_behaves_like 'swift::storage::drive_audit'
end
end
end