Puppet Class: profile::mediawiki::mcrouter_wancache

Defined in:
modules/profile/manifests/mediawiki/mcrouter_wancache.pp

Overview

Configures a mcrouter instance for multi-datacenter caching

Properties

memcached_tls_port

TLS port to connect to other memcached servers for cross dc key replication. Check profile::memcached::port

Parameters:

  • port (Stdlib::Port) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::port'))
  • memcached_notls_port (Stdlib::Port) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::memcached_notls_port'))
  • memcached_tls_port (Stdlib::Port) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::memcached_tls_port'))
  • num_proxies (Integer) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::num_proxies'))
  • timeouts_until_tko (Integer) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::timeouts_until_tko'))
  • gutter_ttl (Integer) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::gutter_ttl'))
  • use_onhost_memcached (Boolean) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::use_onhost_memcached'))
  • use_onhost_memcached_socket (Boolean) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::use_onhost_memcached_socket'))
  • prometheus_exporter (Boolean) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::prometheus_exporter'))
  • servers_by_datacenter_category (Hash) (defaults to: lookup('profile::mediawiki::mcrouter_wancache::shards'))


7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
# File 'modules/profile/manifests/mediawiki/mcrouter_wancache.pp', line 7

class profile::mediawiki::mcrouter_wancache(
    Stdlib::Port $port                           = lookup('profile::mediawiki::mcrouter_wancache::port'),
    Stdlib::Port $memcached_notls_port           = lookup('profile::mediawiki::mcrouter_wancache::memcached_notls_port'),
    Stdlib::Port $memcached_tls_port             = lookup('profile::mediawiki::mcrouter_wancache::memcached_tls_port'),
    Integer      $num_proxies                    = lookup('profile::mediawiki::mcrouter_wancache::num_proxies'),
    Integer      $timeouts_until_tko             = lookup('profile::mediawiki::mcrouter_wancache::timeouts_until_tko'),
    Integer      $gutter_ttl                     = lookup('profile::mediawiki::mcrouter_wancache::gutter_ttl'),
    Boolean      $use_onhost_memcached           = lookup('profile::mediawiki::mcrouter_wancache::use_onhost_memcached'),
    Boolean      $use_onhost_memcached_socket    = lookup('profile::mediawiki::mcrouter_wancache::use_onhost_memcached_socket'),
    Boolean      $prometheus_exporter            = lookup('profile::mediawiki::mcrouter_wancache::prometheus_exporter'),
    Hash         $servers_by_datacenter_category = lookup('profile::mediawiki::mcrouter_wancache::shards')
) {

    # install onhost memcached if this server is going to use a Warmup Route
    # MediaWiki servers are running an onhost memcached instance which
    # they query before reaching out to the memcached cluster
    # size should be 1/4 of total memory
    $onhost_port          = 11210

    if $use_onhost_memcached {
        class { '::memcached':
            size               => floor($facts['memorysize_mb'] * 0.25),
            port               => $onhost_port,
            enable_unix_socket => $use_onhost_memcached_socket,
            growth_factor      => 1.25,
            min_slab_size      => 48,
        }
        # NOTE: This should have been a systemd override puppet define but our puppetization
        # of systemd and memcached doesn't really allow for that due to the
        # following problems:
        # * >1 systemd override per service easily isn't supported
        # * memcached class is already using the above said override based on a
        # boolean parameter. It is already differently used on e.g. mc* vs mw* hosts.
        #
        # So, fallback to doing the crappy thing and create the directory
        # on this specific installation and populate the override via a file
        # resources

        # TODO: This should be fixed first at the systemd puppet module level by
        # allowing >1 arbitrary overrides, then memcached class level by
        # untangling the 2 usages above, then this one should become a 2nd
        # systemd override for memcached
        # TODO: uses systemd::override
        file { '/etc/systemd/system/memcached.service.d/':
            ensure => directory,
            owner  => 'root',
            group  => 'root',
            mode   => '0555',
        }
        file { '/etc/systemd/system/memcached.service.d/cpuaccounting-override.conf':
            ensure  => present,
            content => "[Service]\nCPUAccounting=yes\n",
            owner   => 'root',
            group   => 'root',
            mode    => '0444',
            notify  => Exec['systemd daemon-reload for memcached.service (memcached)']
        }
        include ::profile::prometheus::memcached_exporter
    }

    $servers_by_datacenter = $servers_by_datacenter_category['wancache']
    $gutters_by_datacenter = $servers_by_datacenter_category['gutter']
    $wikifunctions_servers = $servers_by_datacenter_category['wikifunctions'][$::site]

    if $use_onhost_memcached {
        if $use_onhost_memcached_socket {
            $onhost_pool =  { 'onhost' => {
                                'servers' => [
                                  'unix:/var/run/memcached/memcached.sock:ascii:plain'
                                  ]
                                }
                            }
        } else {
            $onhost_pool = { 'onhost' => {
                                'servers' => [
                                  "127.0.0.1:${onhost_port}:ascii:plain"
                                  ]
                                }
                            }
        }
    } else {
        $onhost_pool = {}
    }

    # Gutter pools:
    $gutter_pools = $gutters_by_datacenter.map |$dc, $servers| {
        if $dc == $::site {
            profile::mcrouter_pools("${dc}-gutter", $servers, 'plain', $memcached_notls_port)
        } else {
            profile::mcrouter_pools("${dc}-gutter", $servers, 'ssl', $memcached_tls_port)
        }
    }.reduce()|$memo, $value| { $memo + $value }

    # wikifunction pool. We just need the one for the local datacenter.
    # No need for tls here, either, as the traffic is all dc-local.
    $wikifunction_pool = profile::mcrouter_pools("wf-${::site}", $wikifunctions_servers, 'plain', $memcached_notls_port)
    # Server pools
    $pools = $servers_by_datacenter.map |$dc, $servers| {
        if $dc == $::site {
            profile::mcrouter_pools($dc, $servers, 'plain', $memcached_notls_port)
        } else {
            profile::mcrouter_pools($dc, $servers, 'ssl', $memcached_tls_port)
        }
    }.reduce($onhost_pool + $gutter_pools + $wikifunction_pool) |$memo, $value| { $memo + $value }

    $routes = union(
        # Local cache for each region
        $servers_by_datacenter.map |$dc, $_| {
            $failover_route = $dc ? {
                $::site => true,
                default => false
            };
                {
                    'aliases' => [ "/${dc}/mw/" ],
                    'route' => profile::mcrouter_route($dc, $gutter_ttl, $failover_route)
                }
        },
        # WAN cache: issues reads and add/cas/touch locally and issues set/delete everywhere.
        # MediaWiki will set a prefix of /*/mw-wan when broadcasting, explicitly matching
        # all the mw-wan routes. Broadcasting is thus completely controlled by MediaWiki,
        # but is only allowed for set/delete operations.
        $servers_by_datacenter.map |$dc, $_| {
            $failover_route = true;
            {
                'aliases' => [ "/${dc}/mw-wan/" ],
                'route'   => {
                    'type'               => 'OperationSelectorRoute',
                    'default_policy'     => profile::mcrouter_route($dc, $gutter_ttl, $failover_route),
                    # AllAsyncRoute is used by mcrouter when replicating data to the non-active DC:
                    # https://github.com/facebook/mcrouter/wiki/List-of-Route-Handles#allasyncroute
                    # More info in T225642
                    'operation_policies' => {
                        'set'    => {
                            'type'     => $dc ? {
                                $::site => 'AllSyncRoute',
                                default => 'AllAsyncRoute'
                            },
                            'children' => [ profile::mcrouter_route($dc, $gutter_ttl, $failover_route) ]
                        },
                        'delete' => {
                            'type'     => $dc ? {
                                $::site => 'AllSyncRoute',
                                default => 'AllAsyncRoute'
                            },
                            'children' => [ profile::mcrouter_route($dc, $gutter_ttl, $failover_route) ]
                        },
                    }
                }
            }
        },
        # On-host memcache tier: Keep a short-lived local cache to reduce network load for very hot
        # keys, at the cost of a few seconds' staleness.
        $servers_by_datacenter.map |$dc, $_| {
            $failover_route = $dc ? {
                $::site => true,
                default => false
            };
            {
                'aliases' => [ "/${dc}/mw-with-onhost-tier/" ],
                'route'   => $use_onhost_memcached ? {
                    true  => {
                        'type'               => 'OperationSelectorRoute',
                        'operation_policies' => {
                            # For reads, use WarmUpRoute to try on-host memcache first. If it's not
                            # there, WarmUpRoute tries the ordinary regional pool next, and writes the
                            # result back to the on-host cache, with a short expiration time. The
                            # exptime is ten seconds in order to match our tolerance for DB replication
                            # delay; that level of staleness is acceptable. Based on
                            # https://github.com/facebook/mcrouter/wiki/Two-level-caching#local-instance-with-small-ttl
                            'get' => {
                                'type'    => 'WarmUpRoute',
                                'cold'    => 'PoolRoute|onhost',
                                'warm'    => profile::mcrouter_route($dc, $gutter_ttl, $failover_route),
                                'exptime' => 10,
                            }
                        },
                        # For everything except reads, bypass the on-host tier completely. That means
                        # if a get, set, and get are sent within a ten-second period, they're
                        # guaranteed *not* to have read-your-writes consistency. (If sets updated the
                        # on-host cache, read-your-writes consistency would depend on whether the
                        # requests happened to hit the same host or not, so e.g. mwdebug hosts would
                        # behave differently from the rest of prod, which would be confusing.)
                        'default_policy'     => profile::mcrouter_route($dc, $gutter_ttl, $failover_route)
                    },
                    # If use_onhost_memcached is turned off, always bypass the onhost tier.
                    false => profile::mcrouter_route($dc, $gutter_ttl, $failover_route)
                }
            }
        },
        # wikifunctions pool, fully dc-local, no failover.
        [{
            'aliases' => [ '/local/wf/' ],
            'route'   => "PoolRoute|wf-${::site}"
        }]
    )

    class { 'mcrouter':
        pools                  => $pools,
        routes                 => $routes,
        region                 => $::site,
        cluster                => 'mw',
        num_proxies            => $num_proxies,
        timeouts_until_tko     => $timeouts_until_tko,
        probe_delay_initial_ms => 60000,
        port                   => $port,
    }
    file { '/etc/systemd/system/mcrouter.service.d/cpuaccounting-override.conf':
        content => "[Service]\nCPUAccounting=yes\n",
        owner   => 'root',
        group   => 'root',
        mode    => '0444',
        notify  => Exec['systemd daemon-reload for mcrouter.service (mcrouter)']
    }

    ferm::rule { 'skip_mcrouter_wancache_conntrack_out':
        desc  => 'Skip outgoing connection tracking for mcrouter',
        table => 'raw',
        chain => 'OUTPUT',
        rule  => "proto tcp sport (${port} ${memcached_tls_port}) NOTRACK;",
    }

    ferm::rule { 'skip_mcrouter_wancache_conntrack_in':
        desc  => 'Skip incoming connection tracking for mcrouter',
        table => 'raw',
        chain => 'PREROUTING',
        rule  => "proto tcp dport ${port} NOTRACK;",
    }
    if $prometheus_exporter {
        class {'profile::prometheus::mcrouter_exporter':
            mcrouter_port => $mcrouter::port
        }
    }
}