| 1 |
<?php |
| 2 |
|
| 3 |
if (!defined('ABSPATH')) { |
| 4 |
exit; |
| 5 |
} |
| 6 |
|
| 7 |
/** |
| 8 |
* Shapes the three schema-versioned checkpoint journal record kinds. |
| 9 |
* |
| 10 |
* This class performs no journal or directory I/O. The logger owns lifecycle, |
| 11 |
* phase timing, ordering, and failure handling; this factory owns the compact |
| 12 |
* on-disk representation those operations persist. |
| 13 |
*/ |
| 14 |
final class ABJ_404_Solution_CheckpointRecordFactory { |
| 15 |
|
| 16 |
const SCHEMA_VERSION = 8; |
| 17 |
const ENVELOPE_FULL = 'full'; |
| 18 |
const ENVELOPE_FREQUENT = 'frequent'; |
| 19 |
const ENVELOPE_INTENT = 'intent'; |
| 20 |
|
| 21 |
/** |
| 22 |
* getrusage() keys worth carrying on every full checkpoint, mapped to the |
| 23 |
* compact names written to the journal. |
| 24 |
* |
| 25 |
* The full 17-key array was the single largest thing in the journal: 305 |
| 26 |
* of the 545 bytes an average record occupied, repeated on all 27 records |
| 27 |
* of every request, most of it fields that are structurally zero on Linux |
| 28 |
* (ixrss/idrss/isrss/nswap) or irrelevant to a stall (msgsnd/msgrcv/ |
| 29 |
* nsignals). What survives is what a stall is actually diagnosed with: |
| 30 |
* the user/system CPU split (CPU burn vs blocked), resident memory, |
| 31 |
* voluntary vs involuntary context switches (blocked-on-IO vs preempted, |
| 32 |
* the signature of host-level throttling), page faults, and block IO. |
| 33 |
*/ |
| 34 |
const RUSAGE_FIELDS = array( |
| 35 |
'maxrss' => 'ru_maxrss', |
| 36 |
'minflt' => 'ru_minflt', |
| 37 |
'majflt' => 'ru_majflt', |
| 38 |
'nvcsw' => 'ru_nvcsw', |
| 39 |
'nivcsw' => 'ru_nivcsw', |
| 40 |
'inblock' => 'ru_inblock', |
| 41 |
'oublock' => 'ru_oublock', |
| 42 |
); |
| 43 |
|
| 44 |
/** |
| 45 |
* The most recent intent's CPU sample, so the next intent can report a |
| 46 |
* delta instead of an absolute reading. Single most-recent slot rather |
| 47 |
* than a per-request map (mirrors |
| 48 |
* ABJ_404_Solution_AjaxCheckpointLogger::$previousWriteTelemetry): a |
| 49 |
* request id mismatch means the sample belongs to a different request |
| 50 |
* and the delta is reported as unavailable rather than leaked across the |
| 51 |
* boundary, and nothing has to be evicted from an ever-growing map over |
| 52 |
* an FPM worker's lifetime. |
| 53 |
* |
| 54 |
* @var array{request_id: string, utime_us: int, stime_us: int}|null |
| 55 |
*/ |
| 56 |
private static $previousIntentRusage = null; |
| 57 |
|
| 58 |
/** |
| 59 |
* @param array<string, mixed> $context |
| 60 |
* @return array<string, mixed> |
| 61 |
*/ |
| 62 |
public static function full(array $context): array { |
| 63 |
$record = array( |
| 64 |
'schema_version' => self::SCHEMA_VERSION, |
| 65 |
'envelope' => self::ENVELOPE_FULL, |
| 66 |
'ts' => $context['ts'] ?? null, |
| 67 |
'hrtime_ns' => $context['hrtime_ns'] ?? null, |
| 68 |
'rusage' => self::resourceUsage(), |
| 69 |
'host_pressure' => $context['host_pressure'] ?? array( |
| 70 |
'status' => 'unavailable', |
| 71 |
'reason' => 'sampler_result_unavailable', |
| 72 |
), |
| 73 |
); |
| 74 |
// Host-WIDE pressure above; THIS SITE's own concurrency next. A |
| 75 |
// per-account worker cap (LiteSpeed/CloudLinux LVE) throttles a site |
| 76 |
// whose box looks idle, so the two answer different questions and a |
| 77 |
// record carrying only the first cannot tell them apart. The census |
| 78 |
// owns the shape of its own contribution; see |
| 79 |
// ABJ_404_Solution_SameSiteCensusReading::checkpointFields(). |
| 80 |
$record += class_exists('ABJ_404_Solution_SameSiteCensusReading') |
| 81 |
? ABJ_404_Solution_SameSiteCensusReading::checkpointFields() |
| 82 |
: array('same_site_requests' => -1); |
| 83 |
$record['previous_checkpoint_write'] = $context['previous_checkpoint_write']; |
| 84 |
$record['request_id'] = $context['request_id']; |
| 85 |
$record['event'] = $context['event']; |
| 86 |
$record['checkpoint_id'] = $context['checkpoint_id']; |
| 87 |
$record['pid'] = $context['pid']; |
| 88 |
return $record; |
| 89 |
} |
| 90 |
|
| 91 |
/** |
| 92 |
* @param array<string, mixed> $context |
| 93 |
* @return array<string, mixed> |
| 94 |
*/ |
| 95 |
public static function intent(array $context): array { |
| 96 |
$requestId = is_string($context['request_id'] ?? null) ? $context['request_id'] : ''; |
| 97 |
$delta = self::intentCpuDelta($requestId); |
| 98 |
return array( |
| 99 |
'schema_version' => self::SCHEMA_VERSION, |
| 100 |
'envelope' => self::ENVELOPE_INTENT, |
| 101 |
'hrtime_ns' => $context['hrtime_ns'] ?? null, |
| 102 |
'request_id' => $context['request_id'], |
| 103 |
'event' => 'checkpoint_intent', |
| 104 |
'intended_event' => $context['event'], |
| 105 |
'checkpoint_id' => $context['checkpoint_id'], |
| 106 |
'pid' => $context['pid'], |
| 107 |
'utime_delta_us' => $delta['utime_delta_us'], |
| 108 |
'stime_delta_us' => $delta['stime_delta_us'], |
| 109 |
); |
| 110 |
} |
| 111 |
|
| 112 |
/** |
| 113 |
* @param array<string, mixed> $context |
| 114 |
* @return array<string, mixed> |
| 115 |
*/ |
| 116 |
public static function frequent(array $context): array { |
| 117 |
return array( |
| 118 |
'schema_version' => self::SCHEMA_VERSION, |
| 119 |
'envelope' => self::ENVELOPE_FREQUENT, |
| 120 |
'ts' => $context['ts'] ?? null, |
| 121 |
'hrtime_ns' => $context['hrtime_ns'] ?? null, |
| 122 |
'request_id' => $context['request_id'], |
| 123 |
'event' => $context['event'], |
| 124 |
'checkpoint_id' => $context['checkpoint_id'], |
| 125 |
'pid' => $context['pid'], |
| 126 |
); |
| 127 |
} |
| 128 |
|
| 129 |
/** |
| 130 |
* Shape the completed call telemetry embedded in the next full record. |
| 131 |
* |
| 132 |
* @param array<string, mixed> $context |
| 133 |
* @return array<string, mixed> |
| 134 |
*/ |
| 135 |
public static function completedWriteTelemetry(array $context): array { |
| 136 |
$write = is_array($context['write'] ?? null) ? $context['write'] : array(); |
| 137 |
$intent = is_array($context['intent'] ?? null) ? $context['intent'] : array(); |
| 138 |
$phases = is_array($context['phases_us'] ?? null) ? $context['phases_us'] : array(); |
| 139 |
$telemetry = array( |
| 140 |
'status' => is_string($write['status'] ?? null) ? $write['status'] : 'failed', |
| 141 |
'request_id' => $context['request_id'], |
| 142 |
'event' => $context['event'], |
| 143 |
'checkpoint_id' => $context['checkpoint_id'], |
| 144 |
'elapsed_us' => is_numeric($phases['append'] ?? null) ? max(0, (int)$phases['append']) : 0, |
| 145 |
'total_us' => is_numeric($context['total_us'] ?? null) ? max(0, (int)$context['total_us']) : 0, |
| 146 |
'phases_us' => $phases, |
| 147 |
); |
| 148 |
return array_merge($telemetry, self::failureDetails($write, $intent)); |
| 149 |
} |
| 150 |
|
| 151 |
/** |
| 152 |
* @param array<string, mixed> $write |
| 153 |
* @param array<string, mixed> $intent |
| 154 |
* @return array<string, string> |
| 155 |
*/ |
| 156 |
private static function failureDetails(array $write, array $intent): array { |
| 157 |
$details = array(); |
| 158 |
$intentStatus = is_string($intent['status'] ?? null) ? $intent['status'] : 'failed'; |
| 159 |
if ($intentStatus !== 'complete') { |
| 160 |
$details['intent_status'] = $intentStatus; |
| 161 |
} |
| 162 |
if (is_string($write['reason'] ?? null) && $write['reason'] !== '') { |
| 163 |
$details['reason'] = $write['reason']; |
| 164 |
} |
| 165 |
if (is_string($intent['reason'] ?? null) && $intent['reason'] !== '') { |
| 166 |
$details['intent_reason'] = $intent['reason']; |
| 167 |
} |
| 168 |
return $details; |
| 169 |
} |
| 170 |
|
| 171 |
/** |
| 172 |
* The diagnostic subset of getrusage(), or null where it is unavailable. |
| 173 |
* |
| 174 |
* Absolute counters rather than deltas against a previous record: the |
| 175 |
* excerpt that carries these is allowed to drop records it cannot afford, |
| 176 |
* and a delta chain with a hole in it is unreadable, while an absolute |
| 177 |
* sample stays interpretable on its own. CPU times are folded into single |
| 178 |
* microsecond fields so the tv_sec/tv_usec pairs do not have to be |
| 179 |
* recombined by hand at read time. |
| 180 |
* |
| 181 |
* @return array<string, int>|null |
| 182 |
*/ |
| 183 |
private static function resourceUsage(): ?array { |
| 184 |
$rusage = ABJ_404_Solution_PhpRuntimeCapabilityAdapter::resourceUsage(); |
| 185 |
if (!is_array($rusage)) { |
| 186 |
return null; |
| 187 |
} |
| 188 |
$usage = array( |
| 189 |
'utime_us' => self::microseconds($rusage, 'ru_utime'), |
| 190 |
'stime_us' => self::microseconds($rusage, 'ru_stime'), |
| 191 |
); |
| 192 |
foreach (self::RUSAGE_FIELDS as $name => $key) { |
| 193 |
if (isset($rusage[$key]) && is_numeric($rusage[$key])) { |
| 194 |
$usage[$name] = (int)$rusage[$key]; |
| 195 |
} |
| 196 |
} |
| 197 |
return $usage; |
| 198 |
} |
| 199 |
|
| 200 |
/** |
| 201 |
* One getrusage() tv_sec/tv_usec pair as microseconds. |
| 202 |
* |
| 203 |
* @param array<string, mixed> $rusage |
| 204 |
*/ |
| 205 |
private static function microseconds(array $rusage, string $prefix): int { |
| 206 |
$seconds = isset($rusage[$prefix . '.tv_sec']) && is_numeric($rusage[$prefix . '.tv_sec']) |
| 207 |
? (int)$rusage[$prefix . '.tv_sec'] : 0; |
| 208 |
$micros = isset($rusage[$prefix . '.tv_usec']) && is_numeric($rusage[$prefix . '.tv_usec']) |
| 209 |
? (int)$rusage[$prefix . '.tv_usec'] : 0; |
| 210 |
return ($seconds * 1000000) + $micros; |
| 211 |
} |
| 212 |
|
| 213 |
/** |
| 214 |
* User/system CPU microseconds burned since the previous intent record |
| 215 |
* IN THIS REQUEST, or null on either field when there is no in-request |
| 216 |
* predecessor to diff against (report 193: 833 intents and only 2 full |
| 217 |
* records with rusage meant a 165-second gap could only be classified as |
| 218 |
* spin-vs-blocked because those two full records happened to survive; |
| 219 |
* every intent carrying its own CPU delta means the very next intent |
| 220 |
* after any gap self-classifies it, with no full record required). |
| 221 |
* |
| 222 |
* Deltas rather than the absolute counters full() carries: an intent is |
| 223 |
* written before EVERY checkpoint of any kind (record() and |
| 224 |
* recordFrequent() both call this first), so the intent-to-intent delta |
| 225 |
* already covers every gap in the stream at negligible incremental cost |
| 226 |
* over the getrusage() call intent() already has to make. The absolute |
| 227 |
* reading remains available every 27th-or-so record via full()'s own |
| 228 |
* rusage field, so a reader can still recover a running total. |
| 229 |
* |
| 230 |
* @return array{utime_delta_us: int|null, stime_delta_us: int|null} |
| 231 |
*/ |
| 232 |
private static function intentCpuDelta(string $requestId): array { |
| 233 |
$rusage = ABJ_404_Solution_PhpRuntimeCapabilityAdapter::resourceUsage(); |
| 234 |
if (!is_array($rusage)) { |
| 235 |
return array('utime_delta_us' => null, 'stime_delta_us' => null); |
| 236 |
} |
| 237 |
$utimeUs = self::microseconds($rusage, 'ru_utime'); |
| 238 |
$stimeUs = self::microseconds($rusage, 'ru_stime'); |
| 239 |
$previous = self::$previousIntentRusage; |
| 240 |
self::$previousIntentRusage = array( |
| 241 |
'request_id' => $requestId, |
| 242 |
'utime_us' => $utimeUs, |
| 243 |
'stime_us' => $stimeUs, |
| 244 |
); |
| 245 |
if ($requestId === '' || $previous === null || $previous['request_id'] !== $requestId) { |
| 246 |
return array('utime_delta_us' => null, 'stime_delta_us' => null); |
| 247 |
} |
| 248 |
return array( |
| 249 |
'utime_delta_us' => max(0, $utimeUs - $previous['utime_us']), |
| 250 |
'stime_delta_us' => max(0, $stimeUs - $previous['stime_us']), |
| 251 |
); |
| 252 |
} |
| 253 |
|
| 254 |
/** Test-only: clear the in-process CPU-delta baseline between test cases. */ |
| 255 |
public static function resetForTests(): void { |
| 256 |
self::$previousIntentRusage = null; |
| 257 |
} |
| 258 |
} |
| 259 |
|
