mirror of
https://gitee.com/ulthon/ulthon_admin.git
synced 2026-08-31 05:05:33 +08:00
511 lines
21 KiB
PHP
511 lines
21 KiB
PHP
<?php
|
||
|
||
declare(strict_types=1);
|
||
|
||
namespace base\common\service;
|
||
|
||
use app\admin\model\DebugLogPosition;
|
||
use app\common\service\HostService;
|
||
use think\log\DebugLogToolkit;
|
||
|
||
/**
|
||
* debug 日志增量读取 service(Base 层)——JSONL 按行 + CSV fgetcsv 多行安全.
|
||
*
|
||
* 与 XhprofLogReaderServiceBase 的两处差异(有意为之,勿"顺手同步"回去):
|
||
* 1. JSONL read() 在 reader 内联 decode:每行经 DebugLogToolkit::decodeLine()
|
||
* 数组化后才 yield(坏行计 parseFails 不产出);MAX_LINE_BYTES 收紧为 8MB
|
||
* ——debug 行是短文本行,远小于 xhprof 采样行(~100KB),32MB 防御阈值不必要。
|
||
* 2. 新增 CSV 路径 readCsv():遗留 CSV(DebugMysql.saveByFile 产物 {ymd}.csv)
|
||
* 的 content 含换行是常态,一条逻辑记录跨多物理行——必须用 fgetcsv() 有状态
|
||
* 解析(fgetcsv 原生处理引号内嵌入换行),【禁止】按 \n 切行后 str_getcsv
|
||
* (会把跨行记录撕碎成两条坏行)。
|
||
*
|
||
* 其余协议与 xhprof 版一致:按 offset 增量读、inode/size 双条件轮转检测归零、
|
||
* max 上限、每次产出前推进安全偏移(消费方任意时刻 getCurrentPosition() 取到的
|
||
* 都是"最后一条已消费行末尾")、JSONL 尾部半行不产出(下轮写入 \n 后补齐)、
|
||
* 不自动 savePosition(进度落盘由 T6 导入任务在 insertBatch 同一事务内显式调用)、
|
||
* position 按 (node_id, file_path) 唯一定位 upsert。
|
||
*
|
||
* 平台限制:Windows 下 fstat()['ino'] 恒为 0,inode 轮转检测退化为仅 size<offset
|
||
* 兜底(生产 Linux 有真实 inode,无此问题)——继承自 xhprof 版协议的已知特性,
|
||
* 不是缺陷。
|
||
*
|
||
* 依赖倒置:model 走 app\admin\model\DebugLogPosition 入口类;节点身份走
|
||
* app\common\service\HostService。本表为全新表,无旧数据接管问题,不含 lazyAdopt。
|
||
*/
|
||
class DebugLogReaderServiceBase
|
||
{
|
||
/** 单次 fread 的 buffer 大小(64KB) */
|
||
protected const BUFFER_SIZE = 65536;
|
||
|
||
/** 单行硬上限(8MB):超过整行丢弃计 dropped。真实 debug 行是短文本,8MB 已是异常防御 */
|
||
protected const MAX_LINE_BYTES = 8388608;
|
||
|
||
/** 解析失败样本最多保留条数(超出丢弃,样本只做排查线索不做全量记录) */
|
||
protected const MAX_FAIL_SAMPLES = 3;
|
||
|
||
/** 单条失败样本截断长度(字符) */
|
||
protected const FAIL_SAMPLE_LENGTH = 200;
|
||
|
||
/** @var int 当前文件 inode(read/readCsv 期间有效;Windows 恒为 0) */
|
||
protected $currentInode = 0;
|
||
|
||
/** @var int 安全偏移:最后一条已消费(产出/跳过/失败均算消费)逻辑记录的末尾 */
|
||
protected $currentOffset = 0;
|
||
|
||
/** @var string|null 最后一条已产出行的 md5 */
|
||
protected $currentLineHash = null;
|
||
|
||
/** @var bool 本轮读取是否干净到达 EOF(JSONL:尾部无半行;CSV:fgetcsv 返 false) */
|
||
protected $reachedEof = false;
|
||
|
||
/** @var int 本轮 read 丢弃的超长行数(仅 JSONL 路径;CSV 无缓冲切分逻辑恒为 0) */
|
||
protected $droppedLines = 0;
|
||
|
||
/** @var int 本轮读取解析失败计数(JSONL 坏行 + CSV 字段数不符) */
|
||
protected $parseFails = 0;
|
||
|
||
/** @var string[] 解析失败样本(截断后的原始行,最多 MAX_FAIL_SAMPLES 条) */
|
||
protected $failSamples = [];
|
||
|
||
/**
|
||
* 增量读取 jsonl 文件,yield decodeLine 数组化后的行.
|
||
*
|
||
* 读取协议:
|
||
* 1. fopen + fstat 取当前 inode/size;读 position(无则视为新建 offset=0)
|
||
* 2. 轮转检测:inode 不一致 OR filesize < offset → 归零重读
|
||
* 3. fseek 到 offset,循环 fread 64KB,定位已读段内最后 \n,仅产出完整行
|
||
* 4. 每次 yield 前 currentOffset 推进到该行末尾(含 \n),保证事务内
|
||
* savePosition 与已消费行严格对齐(坏行同样推进——毒行不重读)
|
||
* 5. 完整行经 DebugLogToolkit::decodeLine() 数组化,decode 返 null 计
|
||
* parseFails 不产出(失败样本进 failSamples)
|
||
* 6. 尾部半行(无 \n)不产出,currentOffset 停在半行起点,下次重读
|
||
* 7. 半行缓冲超过 MAX_LINE_BYTES:整行丢弃计 dropped,扫描至下一个 \n 继续
|
||
* 8. yield maxLines 后退出(不自动 savePosition,由上层事务内落盘)
|
||
*
|
||
* @param string $filePath jsonl 文件绝对路径
|
||
* @param int $maxLines 单次产出的最大行数
|
||
*
|
||
* @return \Generator<array> 每次迭代返回一行(键值与 DebugLogToolkit::FIELDS 一致)
|
||
*
|
||
* @throws \RuntimeException 文件无法打开或 stat/fseek/fread 失败时抛出(交由上层 try-catch)
|
||
*/
|
||
public function read(string $filePath, int $maxLines = 100000): \Generator
|
||
{
|
||
// per-run 状态复位(同一个 reader 实例可串行处理多个文件)
|
||
$this->reachedEof = false;
|
||
$this->droppedLines = 0;
|
||
$this->currentLineHash = null;
|
||
$this->parseFails = 0;
|
||
$this->failSamples = [];
|
||
|
||
$fp = @fopen($filePath, 'rb');
|
||
if ($fp === false) {
|
||
throw new \RuntimeException("DebugLogReaderService: 无法打开文件 {$filePath}");
|
||
}
|
||
|
||
try {
|
||
clearstatcache(true, $filePath);
|
||
$stat = fstat($fp);
|
||
if ($stat === false) {
|
||
throw new \RuntimeException("DebugLogReaderService: fstat 失败 {$filePath}");
|
||
}
|
||
$currentInode = (int) $stat['ino'];
|
||
$currentSize = (int) $stat['size'];
|
||
$this->currentInode = $currentInode;
|
||
|
||
// 读 position(无则默认 offset=0)
|
||
$offset = 0;
|
||
$position = $this->getLastPosition($filePath);
|
||
if ($position !== null) {
|
||
$offset = (int) $position['offset'];
|
||
|
||
// 轮转检测:inode 变化(同日文件被重建)OR size < offset(防御性归零)
|
||
if ((int) $position['inode'] !== $currentInode || $currentSize < $offset) {
|
||
$offset = 0;
|
||
}
|
||
}
|
||
$this->currentOffset = $offset;
|
||
|
||
if ($offset > 0 && fseek($fp, $offset) !== 0) {
|
||
throw new \RuntimeException("DebugLogReaderService: fseek 失败 offset={$offset} file={$filePath}");
|
||
}
|
||
|
||
$yielded = 0;
|
||
$buffer = '';
|
||
// $bufferOffset 始终等于 buffer[0] 对应的文件字节偏移
|
||
$bufferOffset = $offset;
|
||
|
||
while (true) {
|
||
$chunk = fread($fp, self::BUFFER_SIZE);
|
||
if ($chunk === false) {
|
||
throw new \RuntimeException("DebugLogReaderService: fread 失败 file={$filePath}");
|
||
}
|
||
if ($chunk === '') {
|
||
// EOF:buffer 为空说明干净收尾;有半行则 currentOffset 已停在最后完整行末尾
|
||
$this->reachedEof = ($buffer === '');
|
||
return;
|
||
}
|
||
$buffer .= $chunk;
|
||
|
||
// 内层循环:处理 buffer 中所有完整行(含末尾 \n 的)
|
||
$pos = 0;
|
||
while (($nlPos = strpos($buffer, "\n", $pos)) !== false) {
|
||
$line = substr($buffer, $pos, $nlPos - $pos);
|
||
$pos = $nlPos + 1;
|
||
|
||
// 先推进安全偏移再产出/判失败:坏行也算已消费(毒行不重读),
|
||
// 消费方在事务内 savePosition 拿到的 offset 与已处理行严格对齐
|
||
$this->currentOffset = $bufferOffset + $pos;
|
||
|
||
$row = DebugLogToolkit::decodeLine($line);
|
||
if ($row === null) {
|
||
$this->parseFails++;
|
||
$this->recordFailSample($line);
|
||
|
||
continue;
|
||
}
|
||
|
||
$this->currentLineHash = md5($line);
|
||
|
||
yield $row;
|
||
$yielded++;
|
||
|
||
if ($yielded >= $maxLines) {
|
||
return;
|
||
}
|
||
}
|
||
|
||
// buffer 中已消费 $pos 字节,半行(buffer[$pos..])保留
|
||
$bufferOffset += $pos;
|
||
$buffer = substr($buffer, $pos);
|
||
|
||
// 超长行防御:整行丢弃(禁止强制切行——会把一条 JSON 切成两条坏行)
|
||
if (strlen($buffer) >= self::MAX_LINE_BYTES) {
|
||
$this->droppedLines++;
|
||
$buffer = '';
|
||
// 扫描至下一个换行符,将其后内容作为新 buffer 继续
|
||
while (true) {
|
||
$chunk = fread($fp, self::BUFFER_SIZE);
|
||
if ($chunk === false || $chunk === '') {
|
||
// 至 EOF 仍未等到换行符:整个尾部按已丢弃处理,偏移推到文件末尾
|
||
$bufferOffset = (int) ftell($fp);
|
||
$this->currentOffset = $bufferOffset;
|
||
$this->reachedEof = true;
|
||
return;
|
||
}
|
||
$nl = strpos($chunk, "\n");
|
||
if ($nl !== false) {
|
||
$bufferOffset = (int) ftell($fp) - (strlen($chunk) - $nl - 1);
|
||
$this->currentOffset = $bufferOffset;
|
||
$buffer = substr($chunk, $nl + 1);
|
||
break;
|
||
}
|
||
}
|
||
}
|
||
}
|
||
} finally {
|
||
if (is_resource($fp)) {
|
||
fclose($fp);
|
||
}
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 增量读取遗留 CSV 文件,yield 关联数组化的数据行.
|
||
*
|
||
* 读取协议(与 read() 的差异全因 CSV 的多物理行特性):
|
||
* 1. fopen 'rb' + fstat 取 inode/size + position 轮转检测归零——与 read() 一致
|
||
* 2. fseek 到 offset 后循环 fgetcsv():fgetcsv 是有状态解析,原生处理引号内
|
||
* 嵌入换行(一条逻辑记录跨多物理行),返回值即一条完整逻辑记录
|
||
* 【禁止】按 \n 切行后 str_getcsv——引号内换行会被撕碎
|
||
* 3. 每读出一条逻辑记录(无论产出/表头跳过/失败)都把 currentOffset 推进到
|
||
* ftell($fp)(记录末尾),保证事务内 savePosition 与已消费记录对齐
|
||
* 4. position==0(从文件头开始)时首行经 DebugLogToolkit::isCsvHeader() 判定,
|
||
* 是表头则跳过不产出(offset 仍推进);续读 offset 处不会是表头,不判定
|
||
* 5. 行字段数 !== 8 计 parseFails 不产出(失败样本进 failSamples)
|
||
* 6. 行字段按 DebugLogToolkit::FIELDS 顺序 map 成关联数组;create_time 为
|
||
* 数字字符串时转 int,create_time_title 保留字符串(CSV 一切皆字符串)
|
||
* 7. 产出 maxRows 条后退出(不自动 savePosition,由上层事务内落盘)
|
||
*
|
||
* 注:CSV 路径无"尾部半行"概念——遗留 CSV 是已封存的静态文件(非追加中),
|
||
* fgetcsv 读到 EOF 返 false 即干净收尾(reachedEof=true)。
|
||
*
|
||
* @param string $filePath csv 文件绝对路径
|
||
* @param int $maxRows 单次产出的最大数据行数(表头/失败行不计)
|
||
*
|
||
* @return \Generator<array> 每次迭代返回一行(键值与 DebugLogToolkit::FIELDS 一致)
|
||
*
|
||
* @throws \RuntimeException 文件无法打开或 stat/fseek 失败时抛出(交由上层 try-catch)
|
||
*/
|
||
public function readCsv(string $filePath, int $maxRows = 100000): \Generator
|
||
{
|
||
// per-run 状态复位(与 read() 同一套状态位,CSV 路径 droppedLines 恒 0)
|
||
$this->reachedEof = false;
|
||
$this->droppedLines = 0;
|
||
$this->currentLineHash = null;
|
||
$this->parseFails = 0;
|
||
$this->failSamples = [];
|
||
|
||
$fp = @fopen($filePath, 'rb');
|
||
if ($fp === false) {
|
||
throw new \RuntimeException("DebugLogReaderService: 无法打开文件 {$filePath}");
|
||
}
|
||
|
||
try {
|
||
clearstatcache(true, $filePath);
|
||
$stat = fstat($fp);
|
||
if ($stat === false) {
|
||
throw new \RuntimeException("DebugLogReaderService: fstat 失败 {$filePath}");
|
||
}
|
||
$currentInode = (int) $stat['ino'];
|
||
$currentSize = (int) $stat['size'];
|
||
$this->currentInode = $currentInode;
|
||
|
||
// 读 position(无则默认 offset=0)
|
||
$offset = 0;
|
||
$position = $this->getLastPosition($filePath);
|
||
if ($position !== null) {
|
||
$offset = (int) $position['offset'];
|
||
|
||
// 轮转检测:inode 变化 OR size < offset(防御性归零)
|
||
if ((int) $position['inode'] !== $currentInode || $currentSize < $offset) {
|
||
$offset = 0;
|
||
}
|
||
}
|
||
$this->currentOffset = $offset;
|
||
|
||
if ($offset > 0 && fseek($fp, $offset) !== 0) {
|
||
throw new \RuntimeException("DebugLogReaderService: fseek 失败 offset={$offset} file={$filePath}");
|
||
}
|
||
|
||
// 仅从文件头(offset==0)开始读时才做表头判定:续读起点不会是表头
|
||
$checkHeader = ($offset === 0);
|
||
$yielded = 0;
|
||
|
||
while (true) {
|
||
// 显式传全部参数:separator/enclosure/escape 与遗留 fputcsv 写出参数一致
|
||
// (PHP 8.4+ 省略 escape 会触发 deprecation,显式传可前向兼容)
|
||
$fields = fgetcsv($fp, null, ',', '"', '\\');
|
||
if ($fields === false) {
|
||
// fgetcsv 返 false 即 EOF(干净收尾,无半行概念)
|
||
$this->reachedEof = true;
|
||
|
||
return;
|
||
}
|
||
|
||
// 每读出一条逻辑记录就把安全偏移推进到记录末尾(含引号内换行占的
|
||
// 多物理行)——表头/失败行同样推进(毒行不重读),保证事务内
|
||
// savePosition 与已消费记录严格对齐
|
||
$this->currentOffset = (int) ftell($fp);
|
||
|
||
if ($checkHeader) {
|
||
$checkHeader = false;
|
||
if (DebugLogToolkit::isCsvHeader($this->normalizeCsvFields($fields))) {
|
||
// 表头行:跳过不产出(offset 已推进)
|
||
continue;
|
||
}
|
||
}
|
||
|
||
if (count($fields) !== 8) {
|
||
$this->parseFails++;
|
||
$this->recordFailSample('csv-fields=' . count($fields) . ': ' . implode(',', $this->stringifyCsvFields($fields)));
|
||
|
||
continue;
|
||
}
|
||
|
||
$row = [];
|
||
foreach (DebugLogToolkit::FIELDS as $i => $field) {
|
||
$value = $fields[$i] ?? '';
|
||
// fgetcsv 对空字段可能返回 null,统一落为 ''(与 JSONL 行的空串语义对齐)
|
||
$row[$field] = $value === null ? '' : $value;
|
||
}
|
||
// create_time 数字字符串转 int(时间戳恒为非负整数);title 保留字符串
|
||
if (ctype_digit((string) $row['create_time'])) {
|
||
$row['create_time'] = (int) $row['create_time'];
|
||
}
|
||
|
||
$this->currentLineHash = md5(serialize($fields));
|
||
|
||
yield $row;
|
||
$yielded++;
|
||
|
||
if ($yielded >= $maxRows) {
|
||
return;
|
||
}
|
||
}
|
||
} finally {
|
||
if (is_resource($fp)) {
|
||
fclose($fp);
|
||
}
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 当前安全读取位置(read/readCsv 期间/结束后可调).
|
||
*
|
||
* @return array{inode: int, offset: int, last_line_hash: string|null}
|
||
*/
|
||
public function getCurrentPosition(): array
|
||
{
|
||
return [
|
||
'inode' => $this->currentInode,
|
||
'offset' => $this->currentOffset,
|
||
'last_line_hash' => $this->currentLineHash,
|
||
];
|
||
}
|
||
|
||
/**
|
||
* 本轮读取是否干净到达 EOF(JSONL:尾部无半行;CSV:fgetcsv 返 false).
|
||
*
|
||
* 导入侧据此判断"文件已导完":reachedEof 且 offset >= filesize 才允许 unlink。
|
||
*/
|
||
public function reachedEof(): bool
|
||
{
|
||
return $this->reachedEof;
|
||
}
|
||
|
||
/**
|
||
* 本轮 read 丢弃的超长行数(由导入侧并入 fails 统计;CSV 路径恒 0).
|
||
*/
|
||
public function getDroppedLines(): int
|
||
{
|
||
return $this->droppedLines;
|
||
}
|
||
|
||
/**
|
||
* 本轮读取解析失败计数(JSONL 坏行 + CSV 字段数不符,由导入侧传给 savePosition).
|
||
*/
|
||
public function getParseFails(): int
|
||
{
|
||
return $this->parseFails;
|
||
}
|
||
|
||
/**
|
||
* 解析失败样本(json 数组字符串或 null,由导入侧传给 savePosition).
|
||
*
|
||
* 每条样本截断至 FAIL_SAMPLE_LENGTH 字符,最多 MAX_FAIL_SAMPLES 条。
|
||
*/
|
||
public function getFailSamples(): ?string
|
||
{
|
||
if ($this->failSamples === []) {
|
||
return null;
|
||
}
|
||
|
||
$json = json_encode($this->failSamples, JSON_UNESCAPED_UNICODE);
|
||
|
||
return $json === false ? null : $json;
|
||
}
|
||
|
||
/**
|
||
* 查询文件的上次读取位置(按 node_id 隔离).
|
||
*
|
||
* @return array|null 命中时返回 [inode, offset, last_line_hash, parse_fail_count, parse_fail_samples],无记录返回 null
|
||
*/
|
||
public function getLastPosition(string $filePath): ?array
|
||
{
|
||
$nodeId = HostService::getNodeId();
|
||
|
||
$row = DebugLogPosition::where('file_path', $filePath)
|
||
->where('node_id', $nodeId)
|
||
->find();
|
||
if ($row === null) {
|
||
return null;
|
||
}
|
||
|
||
return [
|
||
'inode' => (int) $row->getData('inode'),
|
||
'offset' => (int) $row->getData('offset'),
|
||
'last_line_hash' => $row->getData('last_line_hash'),
|
||
'parse_fail_count' => (int) $row->getData('parse_fail_count'),
|
||
'parse_fail_samples' => $row->getData('parse_fail_samples'),
|
||
];
|
||
}
|
||
|
||
/**
|
||
* 保存读取位置(upsert,按 node_id 隔离).
|
||
*
|
||
* 由导入任务在"insertBatch 同一事务"内显式调用(本类不自动保存),
|
||
* 保证崩溃时入库与进度要么同时生效、要么同时回滚,消重复导入窗口。
|
||
*
|
||
* @param string $filePath 日志文件绝对路径
|
||
* @param int $inode 当前文件 inode
|
||
* @param int $offset 新的字节偏移(最后一条已消费记录末尾)
|
||
* @param string|null $lastLineHash 最后一条已产出记录 md5
|
||
* @param int $failCount 本批解析失败累计次数
|
||
* @param string|null $failSamples 解析失败样本(json 或 null)
|
||
*/
|
||
public function savePosition(string $filePath, int $inode, int $offset, ?string $lastLineHash, int $failCount = 0, ?string $failSamples = null): void
|
||
{
|
||
$now = time();
|
||
$nodeId = HostService::getNodeId();
|
||
|
||
$existing = DebugLogPosition::where('file_path', $filePath)
|
||
->where('node_id', $nodeId)
|
||
->find();
|
||
if ($existing !== null) {
|
||
$existing->save([
|
||
'inode' => $inode,
|
||
'offset' => $offset,
|
||
'last_read_time' => $now,
|
||
'last_line_hash' => $lastLineHash,
|
||
'parse_fail_count' => $failCount,
|
||
'parse_fail_samples' => $failSamples,
|
||
]);
|
||
|
||
return;
|
||
}
|
||
|
||
DebugLogPosition::create([
|
||
'node_id' => $nodeId,
|
||
'file_path' => $filePath,
|
||
'inode' => $inode,
|
||
'offset' => $offset,
|
||
'last_read_time' => $now,
|
||
'last_line_hash' => $lastLineHash,
|
||
'parse_fail_count' => $failCount,
|
||
'parse_fail_samples' => $failSamples,
|
||
'create_time' => $now,
|
||
'update_time' => $now,
|
||
]);
|
||
}
|
||
|
||
/**
|
||
* 记录一条解析失败样本(截断至 FAIL_SAMPLE_LENGTH,超量丢弃).
|
||
*/
|
||
protected function recordFailSample(string $raw): void
|
||
{
|
||
if (count($this->failSamples) >= self::MAX_FAIL_SAMPLES) {
|
||
return;
|
||
}
|
||
|
||
$this->failSamples[] = mb_substr($raw, 0, self::FAIL_SAMPLE_LENGTH);
|
||
}
|
||
|
||
/**
|
||
* CSV 字段数组归一化后交 isCsvHeader 判定(null 字段落为 '').
|
||
*
|
||
* @param array $fields fgetcsv 原始返回值
|
||
*/
|
||
protected function normalizeCsvFields(array $fields): array
|
||
{
|
||
return array_map(static function ($value) {
|
||
return $value === null ? '' : $value;
|
||
}, $fields);
|
||
}
|
||
|
||
/**
|
||
* CSV 字段数组字符串化(失败样本拼接用,null 落为空串占位).
|
||
*
|
||
* @param array $fields fgetcsv 原始返回值
|
||
*
|
||
* @return string[]
|
||
*/
|
||
protected function stringifyCsvFields(array $fields): array
|
||
{
|
||
return array_map(static function ($value) {
|
||
return $value === null ? '' : (string) $value;
|
||
}, $fields);
|
||
}
|
||
}
|