Merge "Add .pipeline/ with dev image variant"
[lhc/web/wiklou.git] / includes / libs / rdbms / loadmonitor / LoadMonitor.php
1 <?php
2 /**
3 * This program is free software; you can redistribute it and/or modify
4 * it under the terms of the GNU General Public License as published by
5 * the Free Software Foundation; either version 2 of the License, or
6 * (at your option) any later version.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License along
14 * with this program; if not, write to the Free Software Foundation, Inc.,
15 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
16 * http://www.gnu.org/copyleft/gpl.html
17 *
18 * @file
19 * @ingroup Database
20 */
21
22 namespace Wikimedia\Rdbms;
23
24 use Psr\Log\LoggerInterface;
25 use Psr\Log\NullLogger;
26 use Wikimedia\ScopedCallback;
27 use BagOStuff;
28 use WANObjectCache;
29
30 /**
31 * Basic DB load monitor with no external dependencies
32 * Uses memcached to cache the replication lag for a short time
33 *
34 * @ingroup Database
35 */
36 class LoadMonitor implements ILoadMonitor {
37 /** @var ILoadBalancer */
38 protected $lb;
39 /** @var BagOStuff */
40 protected $srvCache;
41 /** @var WANObjectCache */
42 protected $wanCache;
43 /** @var LoggerInterface */
44 protected $replLogger;
45
46 /** @var float Moving average ratio (e.g. 0.1 for 10% weight to new weight) */
47 private $movingAveRatio;
48 /** @var int Amount of replication lag in seconds before warnings are logged */
49 private $lagWarnThreshold;
50
51 /** @var int cache key version */
52 const VERSION = 1;
53 /** @var int Default 'max lag' in seconds when unspecified */
54 const LAG_WARN_THRESHOLD = 10;
55
56 /**
57 * @param ILoadBalancer $lb
58 * @param BagOStuff $srvCache
59 * @param WANObjectCache $wCache
60 * @param array $options
61 * - movingAveRatio: moving average constant for server weight updates based on lag
62 * - lagWarnThreshold: how many seconds of lag trigger warnings
63 */
64 public function __construct(
65 ILoadBalancer $lb, BagOStuff $srvCache, WANObjectCache $wCache, array $options = []
66 ) {
67 $this->lb = $lb;
68 $this->srvCache = $srvCache;
69 $this->wanCache = $wCache;
70 $this->replLogger = new NullLogger();
71
72 $this->movingAveRatio = $options['movingAveRatio'] ?? 0.1;
73 $this->lagWarnThreshold = $options['lagWarnThreshold'] ?? self::LAG_WARN_THRESHOLD;
74 }
75
76 public function setLogger( LoggerInterface $logger ) {
77 $this->replLogger = $logger;
78 }
79
80 final public function scaleLoads( array &$weightByServer, $domain ) {
81 $serverIndexes = array_keys( $weightByServer );
82 $states = $this->getServerStates( $serverIndexes, $domain );
83 $newScalesByServer = $states['weightScales'];
84 foreach ( $weightByServer as $i => $weight ) {
85 if ( isset( $newScalesByServer[$i] ) ) {
86 $weightByServer[$i] = $weight * $newScalesByServer[$i];
87 } else { // server recently added to config?
88 $host = $this->lb->getServerName( $i );
89 $this->replLogger->error( __METHOD__ . ": host $host not in cache" );
90 }
91 }
92 }
93
94 final public function getLagTimes( array $serverIndexes, $domain ) {
95 return $this->getServerStates( $serverIndexes, $domain )['lagTimes'];
96 }
97
98 protected function getServerStates( array $serverIndexes, $domain ) {
99 $writerIndex = $this->lb->getWriterIndex();
100 if ( count( $serverIndexes ) == 1 && reset( $serverIndexes ) == $writerIndex ) {
101 # Single server only, just return zero without caching
102 return [
103 'lagTimes' => [ $writerIndex => 0 ],
104 'weightScales' => [ $writerIndex => 1.0 ]
105 ];
106 }
107
108 $key = $this->getCacheKey( $serverIndexes );
109 # Randomize TTLs to reduce stampedes (4.0 - 5.0 sec)
110 // @phan-suppress-next-line PhanTypeMismatchArgumentInternal
111 $ttl = mt_rand( 4e6, 5e6 ) / 1e6;
112 # Keep keys around longer as fallbacks
113 $staleTTL = 60;
114
115 # (a) Check the local APC cache
116 $value = $this->srvCache->get( $key );
117 if ( $value && $value['timestamp'] > ( microtime( true ) - $ttl ) ) {
118 $this->replLogger->debug( __METHOD__ . ": got lag times ($key) from local cache" );
119 return $value; // cache hit
120 }
121 $staleValue = $value ?: false;
122
123 # (b) Check the shared cache and backfill APC
124 $value = $this->wanCache->get( $key );
125 if ( $value && $value['timestamp'] > ( microtime( true ) - $ttl ) ) {
126 $this->srvCache->set( $key, $value, $staleTTL );
127 $this->replLogger->debug( __METHOD__ . ": got lag times ($key) from main cache" );
128
129 return $value; // cache hit
130 }
131 $staleValue = $value ?: $staleValue;
132
133 # (c) Cache key missing or expired; regenerate and backfill
134 if ( $this->srvCache->lock( $key, 0, 10 ) ) {
135 # Let only this process update the cache value on this server
136 $sCache = $this->srvCache;
137 /** @noinspection PhpUnusedLocalVariableInspection */
138 $unlocker = new ScopedCallback( function () use ( $sCache, $key ) {
139 $sCache->unlock( $key );
140 } );
141 } elseif ( $staleValue ) {
142 # Could not acquire lock but an old cache exists, so use it
143 return $staleValue;
144 }
145
146 $lagTimes = [];
147 $weightScales = [];
148 $movAveRatio = $this->movingAveRatio;
149 foreach ( $serverIndexes as $i ) {
150 if ( $i == $this->lb->getWriterIndex() ) {
151 $lagTimes[$i] = 0; // master always has no lag
152 $weightScales[$i] = 1.0; // nominal weight
153 continue;
154 }
155
156 # Handles with open transactions are avoided since they might be subject
157 # to REPEATABLE-READ snapshots, which could affect the lag estimate query.
158 $flags = ILoadBalancer::CONN_TRX_AUTOCOMMIT | ILoadBalancer::CONN_SILENCE_ERRORS;
159 $conn = $this->lb->getAnyOpenConnection( $i, $flags );
160 if ( $conn ) {
161 $close = false; // already open
162 } else {
163 // Get a connection to this server without triggering other server connections
164 $conn = $this->lb->getServerConnection( $i, ILoadBalancer::DOMAIN_ANY, $flags );
165 $close = true; // new connection
166 }
167
168 $lastWeight = $staleValue['weightScales'][$i] ?? 1.0;
169 $coefficient = $this->getWeightScale( $i, $conn ?: null );
170 $newWeight = $movAveRatio * $coefficient + ( 1 - $movAveRatio ) * $lastWeight;
171
172 // Scale from 10% to 100% of nominal weight
173 $weightScales[$i] = max( $newWeight, 0.10 );
174
175 $host = $this->lb->getServerName( $i );
176
177 if ( !$conn ) {
178 $lagTimes[$i] = false;
179 $this->replLogger->error(
180 __METHOD__ . ": host {db_server} is unreachable",
181 [ 'db_server' => $host ]
182 );
183 continue;
184 }
185
186 $lagTimes[$i] = $conn->getLag();
187 if ( $lagTimes[$i] === false ) {
188 $this->replLogger->error(
189 __METHOD__ . ": host {db_server} is not replicating?",
190 [ 'db_server' => $host ]
191 );
192 } elseif ( $lagTimes[$i] > $this->lagWarnThreshold ) {
193 $this->replLogger->warning(
194 "Server {host} has {lag} seconds of lag (>= {maxlag})",
195 [
196 'host' => $host,
197 'lag' => $lagTimes[$i],
198 'maxlag' => $this->lagWarnThreshold
199 ]
200 );
201 }
202
203 if ( $close ) {
204 # Close the connection to avoid sleeper connections piling up.
205 # Note that the caller will pick one of these DBs and reconnect,
206 # which is slightly inefficient, but this only matters for the lag
207 # time cache miss cache, which is far less common that cache hits.
208 $this->lb->closeConnection( $conn );
209 }
210 }
211
212 # Add a timestamp key so we know when it was cached
213 $value = [
214 'lagTimes' => $lagTimes,
215 'weightScales' => $weightScales,
216 'timestamp' => microtime( true )
217 ];
218 $this->wanCache->set( $key, $value, $staleTTL );
219 $this->srvCache->set( $key, $value, $staleTTL );
220 $this->replLogger->info( __METHOD__ . ": re-calculated lag times ($key)" );
221
222 return $value;
223 }
224
225 /**
226 * @param int $index Server index
227 * @param IDatabase|null $conn Connection handle or null on connection failure
228 * @return float
229 */
230 protected function getWeightScale( $index, IDatabase $conn = null ) {
231 return $conn ? 1.0 : 0.0;
232 }
233
234 private function getCacheKey( array $serverIndexes ) {
235 sort( $serverIndexes );
236 // Lag is per-server, not per-DB, so key on the master DB name
237 return $this->srvCache->makeGlobalKey(
238 'lag-times',
239 self::VERSION,
240 $this->lb->getServerName( $this->lb->getWriterIndex() ),
241 implode( '-', $serverIndexes )
242 );
243 }
244 }