feat(traffic): expose storage compaction health

go
codex 21 hours ago
parent 6fa13651b1
commit 2bb9bb78b6

@ -160,7 +160,7 @@ libvirt 的持久 pool 定义就是 Agent 后续使用的映射,不另写存
公网网络接入在宿主机的“网络接入”中按资源池线路独立配置,可使用 `bridge:br0`、其他 Linux 网桥或 `network:default` 这样的 libvirt network。名称以字母或数字开头,后续只能包含字母、数字、点、下划线和短横线,bridge 名称为 1~15 字符,libvirt network 名称为 1~64 字符。宿主机上的公网网桥或 libvirt network 必须由运维预先准备。Agent 只发现、验证和使用现有公网网络,不创建公网网桥、不启动 libvirt network,也不修改公网地址、上游网关或挂载。`cloud_hosts.default_network` 仅保留为兼容展示值,不参与 Agent 就绪、调度、恢复、迁移或换 IP 决策;真正的运行时权威是资源池线路、当前 Agent 代次已验证的网络接入,以及绑定该接入的公网地址池。VPC 使用下述独立的 Agent 受管 bridge 和 L3 路由,不改变这一公网边界。
当前完整 schema 版本为 `go-agent-wss-v55-traffic-checkpoint-tiers`。v48 将滥用监测拆分为 72 小时原始证据、短期终态投影、7 天去重凭据以及 5 分钟/小时趋势汇总;v49 为实例性能图表增加 5 分钟和小时汇总,并将一分钟原始点限制为 24 小时、小时趋势限制为 35 天;v50 保持分钟级 FEFO 扣减和游标更新不变,将同一小时、实例和额度的用量累积到持久结算桶,小时结束、额度耗尽、到期、债务结算或异常栅栏时再生成不可变账本事件。迟到采样使用新的桶修订,绝不改写已封存事件;v51 将性能历史改为逻辑实例主键,删除每行重复的宿主机 hash 和文本 UUID,将永久流量总数/检查点的 UUID 与 hash 改为二进制并删除未被读取路径使用的宽索引;v52 保持实时 FEFO 扣减不变,将新用量累积到 UTC 每日结算窗口,并通过 `bucket_seconds` 与 `seal_at` 继续识别和封存升级前的小时窗口;v53 增加紧凑结算凭据,冻结已经安全压缩的窗口身份、最大修订、账本事件和载荷哈希,让迟到采样无需依赖完整旧桶也能从下一修订继续;v54 启用7天安全窗后的完整桶压缩,只有账本身份、金额、载荷和哈希全部吻合时,才在同一主库事务中更新凭据并删除完整桶;v55 为每个逻辑实例写入一条 UTC 每日累计检查点,并在服务周期首次确定基准时按周期冻结精确边界锚点,维护迁移后仍沿用同一逻辑实例历史。部署既有数据库时必须先停止生命周期与滥用监测 worker,按当前 `cloud.sql`、`cloud_log.sql` 的新增表和字段制作一次性迁移并完成回填,再启动新代码;v51 涉及日志表主键与二进制身份重建,不能在采集 worker 运行时原地改列,也不能让 v50 代码与 v51 日志表混跑;v52 必须先为既有结算桶回填 `bucket_seconds = 3600` 与 `seal_at = bucket_start + 3600`,再替换唯一索引、到期索引和检查约束;v53 必须先创建 `cloud_traffic_settlement_receipts`;v54 必须增加 `cloud_traffic_settlement_retention` 索引后才能启动清理,验证失败的桶保持原样且不得人工跳过;v55 必须先给小时检查点回填可确认的逻辑 `instance_id`,再创建每日检查点和周期边界锚点表;无法唯一对应实例的旧检查点保持原样并禁止自动清理。不得让缺少汇总表、结算桶窗口字段、紧凑凭据表、每日检查点、周期锚点或 `instance_id` 的旧数据库直接进入采集或流量结算。v47 为 `cloud_offers` 增加季度、年度及对应自动备份价格;既有方案升级时必须在恢复销售前显式回填 90/365 天价格,不能让数据库默认零价进入购买目录。从 v44 升级时,必须先停止滥用观测与投影 worker,为既有 finding 回填 `occurrence_generation = 1` 与 `transition_revision = 1`,再创建 `cloud_abuse_alert_events` 和 `cloud_abuse_alert_event_items`;从 v45 升级还必须停止通知 worker,创建 `cloud_abuse_alert_deliveries`、`administrator_telegram_delivery_gate` 并写入唯一 gate 行。历史 finding 不反向合成事件;事件只从升级后主库事务内真实发生的 `opened`、`escalated` 和 `evidence_clear` 解决转移产生。完成列、约束、索引、外键和 gate 核对后,先恢复滥用观测,再恢复通知 worker。
当前完整 schema 版本为 `go-agent-wss-v55-traffic-checkpoint-tiers`。v48 将滥用监测拆分为 72 小时原始证据、短期终态投影、7 天去重凭据以及 5 分钟/小时趋势汇总;v49 为实例性能图表增加 5 分钟和小时汇总,并将一分钟原始点限制为 24 小时、小时趋势限制为 35 天;v50 保持分钟级 FEFO 扣减和游标更新不变,将同一小时、实例和额度的用量累积到持久结算桶,小时结束、额度耗尽、到期、债务结算或异常栅栏时再生成不可变账本事件。迟到采样使用新的桶修订,绝不改写已封存事件;v51 将性能历史改为逻辑实例主键,删除每行重复的宿主机 hash 和文本 UUID,将永久流量总数/检查点的 UUID 与 hash 改为二进制并删除未被读取路径使用的宽索引;v52 保持实时 FEFO 扣减不变,将新用量累积到 UTC 每日结算窗口,并通过 `bucket_seconds` 与 `seal_at` 继续识别和封存升级前的小时窗口;v53 增加紧凑结算凭据,冻结已经安全压缩的窗口身份、最大修订、账本事件和载荷哈希,让迟到采样无需依赖完整旧桶也能从下一修订继续;v54 启用7天安全窗后的完整桶压缩,只有账本身份、金额、载荷和哈希全部吻合时,才在同一主库事务中更新凭据并删除完整桶;v55 为每个逻辑实例写入一条 UTC 每日累计检查点,并在服务周期首次确定基准时按周期冻结精确边界锚点,维护迁移后仍沿用同一逻辑实例历史。部署既有数据库时必须先停止生命周期与滥用监测 worker,按当前 `cloud.sql`、`cloud_log.sql` 的新增表和字段制作一次性迁移并完成回填,再启动新代码;v51 涉及日志表主键与二进制身份重建,不能在采集 worker 运行时原地改列,也不能让 v50 代码与 v51 日志表混跑;v52 必须先为既有结算桶回填 `bucket_seconds = 3600` 与 `seal_at = bucket_start + 3600`,再替换唯一索引、到期索引和检查约束;v53 必须先创建 `cloud_traffic_settlement_receipts`;v54 必须增加 `cloud_traffic_settlement_retention` 索引后才能启动清理,验证失败的桶保持原样且不得人工跳过;v55 必须先给小时检查点回填可确认的逻辑 `instance_id`,再创建每日检查点和周期边界锚点表;无法唯一对应实例的旧检查点写入保留哨兵 `instance_id = 0`,不得伪造归属,自动清理也会忽略这些行。不得让缺少汇总表、结算桶窗口字段、紧凑凭据表、每日检查点、周期锚点或 `instance_id` 的旧数据库直接进入采集或流量结算。v47 为 `cloud_offers` 增加季度、年度及对应自动备份价格;既有方案升级时必须在恢复销售前显式回填 90/365 天价格,不能让数据库默认零价进入购买目录。从 v44 升级时,必须先停止滥用观测与投影 worker,为既有 finding 回填 `occurrence_generation = 1` 与 `transition_revision = 1`,再创建 `cloud_abuse_alert_events` 和 `cloud_abuse_alert_event_items`;从 v45 升级还必须停止通知 worker,创建 `cloud_abuse_alert_deliveries`、`administrator_telegram_delivery_gate` 并写入唯一 gate 行。历史 finding 不反向合成事件;事件只从升级后主库事务内真实发生的 `opened`、`escalated` 和 `evidence_clear` 解决转移产生。完成列、约束、索引、外键和 gate 核对后,先恢复滥用观测,再恢复通知 worker。
事件 header 与 projection receipt 在同一个主库事务中写入,且事件先于 receipt;同一实例同一采样桶的多规则、双协议族转移合并为一个 header 和多个 item。`continued` 不产生事件,`policy_replaced` 与 `owner_changed` 只关闭 finding,不能伪造“恢复”事件。item 已冻结当时的聚合 evidence、规则身份和 finding 代次,后续投递不得回读会变化的 finding。v46 为每个事件建立至多一条 Telegram 投递决策;远端调用开始后的中断或回执不确定统一进入 `unknown`,禁止自动重放。删除用户时按事件冻结的 `user_id` 显式删除 header,item 与投递账本按当前外键和账户清理规则收敛,避免实例转交后遗留原租户数据。

@ -358,3 +358,62 @@ Before marking rollout complete, retain:
If live Agent/network acceptance was not completed, report only that static
and local tests passed; do not claim the shared traffic pool is production
ready.
## 7. Traffic-storage compaction rollout
Traffic-storage compaction changes retention only. It does not change the
minute-level cumulative counter, FEFO allocation, pool projection or network
policy decision. Deploy it as a clean schema/code boundary:
1. Stop lifecycle before changing either database. Keep ordinary and recovery
workers running unless the same change also modifies their schema.
2. Back up the business and log databases. Record row counts and oldest
timestamps for settlement buckets, ledger events, hourly checkpoints,
daily traffic rows and service periods.
3. Apply the exact reviewed v55 migration. Create the compact settlement
receipt table, settlement retention index, daily checkpoint table and
service-period boundary-anchor table. Add `instance_id` and its index to
hourly checkpoints.
4. Backfill an hourly checkpoint's logical instance only from one exact,
unambiguous current or retained placement identity. Set an unmappable old
row to the reserved value `instance_id = 0`; never infer ownership from a
user, name, address or nearby timestamp. Those rows are retained for manual
review and are excluded from automatic deletion.
5. For each mapped instance/day, a reviewed one-time migration may derive one
daily row from the earliest and latest existing hourly cumulative rows.
Do not synthesize service-period anchors. Existing periods with persisted
base counters need no anchor; an open period with null bases deliberately
blocks deletion until the normal lifecycle path freezes real evidence.
6. Deploy the matching code and restart lifecycle. Open the administrator
task page and run its read-only health check. The “流量存储压缩” row exposes
due settlement buckets, due hourly rows, involved instances, cursor
attention count and oldest lag without exposing IDs or payloads.
7. Keep the pool in its existing rollout state while observing at least one
complete collection, accounting, settlement-compaction and
checkpoint-compaction pass. A backlog less than one hour beyond its
retention window remains informational; a larger lag, scan truncation or
attention cursor is a warning that must be investigated.
Acceptance requires all of the following:
- a repeated cumulative observation has no second accounting effect;
- a sealed daily bucket older than seven days is replaced by one matching
compact receipt while its immutable ledger event remains unchanged;
- a late observation starts the next bucket revision after detail compaction;
- an hourly checkpoint older than 72 hours is removed only when its daily row
contains the exact cumulative values and every relevant open period has
persisted bases or a valid anchor;
- missing daily evidence, a malformed anchor or an attention cursor retains
the hourly row and reports a warning;
- deleting an hourly row does not change the activated period baseline;
- 36-day performance-hour rows are pruned while 34-day rows and the complete
30-day chart remain readable;
- account deletion removes the user's totals, checkpoints, daily history,
anchors and performance history.
Rollback is a database restore plus matching source rollback performed while
lifecycle is stopped. Do not run old code against the v55 log schema, remove
receipts to recreate full settlement buckets, delete blockers, manufacture
anchors or mark an attention cursor ready. A retention incident is fail-safe:
leave the extra rows in place, repair the evidence or schema, and rerun the
same bounded lifecycle stage.

@ -16,9 +16,11 @@ final class CloudRuntimeHealthService
private \Closure $agentProbe;
private \Closure $redisProbe;
private \Closure $heartbeatProbe;
private \Closure $trafficStorageProbe;
public function __construct(?callable $databaseProbe = null, ?callable $agentProbe = null,
?callable $redisProbe = null, ?callable $heartbeatProbe = null)
?callable $redisProbe = null, ?callable $heartbeatProbe = null,
?callable $trafficStorageProbe = null)
{
$this->databaseProbe = \Closure::fromCallable($databaseProbe ?? [self::class, 'probeDatabase']);
$this->agentProbe = \Closure::fromCallable($agentProbe ?? static function (
@ -53,6 +55,10 @@ final class CloudRuntimeHealthService
});
$this->heartbeatProbe = \Closure::fromCallable($heartbeatProbe
?? [CloudRuntimeHeartbeat::class, 'snapshot']);
$this->trafficStorageProbe = \Closure::fromCallable(
$trafficStorageProbe
?? [CloudTrafficStorageHealthService::class, 'snapshot']
);
}
public function snapshot(array $actor): array
@ -67,9 +73,16 @@ final class CloudRuntimeHealthService
$checks[] = $this->databaseCheck('log', 'log_database', '日志数据库',
'已完成只读查询;不代表指标采集程序正在运行。',
'日志数据库读取失败,请检查连接和服务日志。');
$checks[] = $this->trafficStorageCheck();
} else {
$checks[] = $this->row('log_database', '日志数据库', 'disabled',
'未启用日志数据库;虚拟化指标依赖该配置。');
$checks[] = $this->row(
'traffic_storage',
'流量存储压缩',
'disabled',
'日志数据库未启用,未执行流量存储压缩检查。'
);
}
$checks[] = $this->check('agents', '宿主机 Agent 配置',
fn (): bool => ($this->agentProbe)($actor) === true,
@ -242,6 +255,53 @@ final class CloudRuntimeHealthService
return $this->row($key, $label, $ok ? 'ok' : 'error', $ok ? $success : $failure);
}
private function trafficStorageCheck(): array
{
try {
$result = ($this->trafficStorageProbe)();
if (!is_array($result)
|| !in_array($result['status'] ?? null,
['ok', 'warning'], true)
|| !is_string($result['message'] ?? null)
|| strlen($result['message']) > 500
|| preg_match('/[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]/',
$result['message']) === 1
|| !is_array($result['metrics'] ?? null)) {
throw new \RuntimeException('traffic_storage_health_invalid');
}
$metrics = [];
foreach ([
'settlement_buckets_due', 'hourly_checkpoints_due',
'hourly_checkpoint_instances_due',
'cursor_attention_instances', 'oldest_due_lag_seconds',
] as $key) {
$value = $result['metrics'][$key] ?? null;
if (!is_int($value) || $value < 0) {
throw new \RuntimeException(
'traffic_storage_health_invalid'
);
}
$metrics[$key] = $value;
}
$metrics['instance_scan_truncated'] =
($result['metrics']['instance_scan_truncated'] ?? null)
=== true;
return $this->row(
'traffic_storage',
'流量存储压缩',
(string) $result['status'],
(string) $result['message']
) + ['metrics' => $metrics];
} catch (\Throwable $exception) {
return $this->row(
'traffic_storage',
'流量存储压缩',
'error',
'流量存储压缩状态读取失败,请检查数据库结构与生命周期日志。'
);
}
}
private function row(string $key, string $label, string $status, string $message): array
{
return compact('key', 'label', 'status', 'message') + ['checked_at' => time()];

@ -35,6 +35,7 @@ final class CloudRuntimeHeartbeat
'abuse_monitor_retention' => '网络观测保留清理',
'traffic_sync' => '流量同步',
'traffic_settlement_compaction' => '流量结算压缩',
'traffic_checkpoint_compaction' => '流量检查点压缩',
'runtime_status_sync' => '实例状态同步',
'instance_reconcile' => '实例策略核对',
'idle' => '等待下一轮', 'stopped' => '已退出'];

@ -0,0 +1,122 @@
<?php
namespace App\Services\Cloud;
use App\Models\User;
/** Read-only bounded summary for traffic-storage compaction health. */
final class CloudTrafficStorageHealthService
{
private const WARNING_LAG_SECONDS = 3600;
private const INSTANCE_SCAN_LIMIT = 2000;
/**
* @return array{
* key:string,label:string,status:string,message:string,checked_at:int,
* metrics:array<string,int|bool>
* }
*/
public static function snapshot(?int $now = null): array
{
$now ??= time();
if ($now <= max(
CloudTrafficSettlementRetention::RETENTION_SECONDS,
CloudTrafficCheckpointRetention::RETENTION_SECONDS
)) {
throw new \InvalidArgumentException(
'traffic_storage_health_clock_invalid'
);
}
$main = User::getConnectionResolver()->connection();
$log = User::getConnectionResolver()->connection('log');
if (!$main->getSchemaBuilder()->hasTable(
'cloud_traffic_usage_settlement_buckets'
) || !$main->getSchemaBuilder()->hasTable(
'cloud_traffic_instance_cursors'
) || !$log->getSchemaBuilder()->hasTable(
'cloud_vm_traffic_checkpoints'
)) {
throw new \RuntimeException('traffic_storage_health_schema_missing');
}
$settlementCutoff = $now
- CloudTrafficSettlementRetention::RETENTION_SECONDS;
$checkpointCutoff = $now
- CloudTrafficCheckpointRetention::RETENTION_SECONDS;
$settlementQuery = $main->table(
'cloud_traffic_usage_settlement_buckets'
)->where('state', 'sealed')->where('updated_at', '<=', $settlementCutoff);
$settlementCount = (int) (clone $settlementQuery)->count();
$oldestSettlement = $settlementCount === 0 ? null
: (int) (clone $settlementQuery)->min('updated_at');
$checkpointQuery = $log->table('cloud_vm_traffic_checkpoints')
->where('instance_id', '>', 0)
->where('first_observed_at', '<=', $checkpointCutoff);
$checkpointCount = (int) (clone $checkpointQuery)->count();
$oldestCheckpoint = $checkpointCount === 0 ? null
: (int) (clone $checkpointQuery)->min('first_observed_at');
$dueInstanceIds = (clone $checkpointQuery)
->select('instance_id')->distinct()->orderBy('instance_id')
->limit(self::INSTANCE_SCAN_LIMIT + 1)->pluck('instance_id')
->map(static fn ($id): int => (int) $id)
->filter(static fn (int $id): bool => $id > 0)
->values()->all();
$scanTruncated = count($dueInstanceIds) > self::INSTANCE_SCAN_LIMIT;
if ($scanTruncated) {
$dueInstanceIds = array_slice(
$dueInstanceIds, 0, self::INSTANCE_SCAN_LIMIT
);
}
$attentionCount = $dueInstanceIds === [] ? 0
: (int) $main->table('cloud_traffic_instance_cursors')
->whereIn('instance_id', $dueInstanceIds)
->where('state', 'attention')->count();
$oldestLag = max(
$oldestSettlement === null
? 0 : $settlementCutoff - $oldestSettlement,
$oldestCheckpoint === null
? 0 : $checkpointCutoff - $oldestCheckpoint
);
$metrics = [
'settlement_buckets_due' => $settlementCount,
'hourly_checkpoints_due' => $checkpointCount,
'hourly_checkpoint_instances_due' => count($dueInstanceIds),
'cursor_attention_instances' => $attentionCount,
'oldest_due_lag_seconds' => max(0, $oldestLag),
'instance_scan_truncated' => $scanTruncated,
];
$status = $attentionCount > 0 || $scanTruncated
|| $oldestLag > self::WARNING_LAG_SECONDS
? 'warning' : 'ok';
if ($settlementCount === 0 && $checkpointCount === 0) {
$message = '结算桶和小时检查点均处于既定保留窗口内。';
} else {
$message = '待压缩结算桶 ' . $settlementCount
. ' 个,小时检查点 ' . $checkpointCount . ' 行(涉及 '
. count($dueInstanceIds) . ' 个实例);最旧超出窗口 '
. self::duration(max(0, $oldestLag)) . ',其中流量游标异常 '
. $attentionCount . ' 个实例。';
if ($scanTruncated) {
$message .= ' 实例计数已达到有界扫描上限。';
}
}
return [
'key' => 'traffic_storage',
'label' => '流量存储压缩',
'status' => $status,
'message' => $message,
'checked_at' => $now,
'metrics' => $metrics,
];
}
private static function duration(int $seconds): string
{
if ($seconds >= 86400) {
return intdiv($seconds, 86400) . ' 天';
}
if ($seconds >= 3600) {
return intdiv($seconds, 3600) . ' 小时';
}
return intdiv($seconds, 60) . ' 分钟';
}
}

@ -39,7 +39,22 @@ $service = new CloudRuntimeHealthService(
$calls[] = 'agents'; return true;
},
static function () use (&$calls): bool { $calls[] = 'redis'; return true; },
static fn (): array => $heartbeats
static fn (): array => $heartbeats,
static function () use (&$calls): array {
$calls[] = 'traffic_storage';
return [
'status' => 'warning',
'message' => '待压缩结算桶 2 个。',
'metrics' => [
'settlement_buckets_due' => 2,
'hourly_checkpoints_due' => 3,
'hourly_checkpoint_instances_due' => 1,
'cursor_attention_instances' => 0,
'oldest_due_lag_seconds' => 120,
'instance_scan_truncated' => false,
],
];
}
);
foreach ([['user_id' => 42, 'is_admin' => false], ['user_id' => 0, 'is_admin' => true], []] as $denied) {
try {
@ -52,8 +67,8 @@ foreach ([['user_id' => 42, 'is_admin' => false], ['user_id' => 0, 'is_admin' =>
$assert($calls === [], 'Unauthorized requests performed probes.');
$payload = $service->snapshot($actor);
$checks = array_column($payload['checks'], null, 'key');
$assert(count($checks) === 14
&& $calls === ['default', 'log', 'agents', 'redis'],
$assert(count($checks) === 15
&& $calls === ['default', 'log', 'traffic_storage', 'agents', 'redis'],
'Missing or duplicate probes.');
$assert($checks['agents']['status'] === 'ok' && $checks['worker']['status'] === 'busy'
&& $checks['policy']['status'] === 'unknown'
@ -63,6 +78,8 @@ $assert($checks['agents']['status'] === 'ok' && $checks['worker']['status'] ===
&& $checks['storage_publication']['status'] === 'unknown'
&& $checks['storage_health']['status'] === 'unknown'
&& $checks['abuse_monitor']['status'] === 'unknown'
&& $checks['traffic_storage']['status'] === 'warning'
&& $checks['traffic_storage']['metrics']['settlement_buckets_due'] === 2
&& $checks['lifecycle']['status'] === 'warning',
'Health or heartbeat semantics changed.');
$assert(!str_contains(json_encode($payload), 'MUST_NOT_EXPOSE'), 'Heartbeat internal fields leaked.');
@ -87,7 +104,9 @@ $calls = [];
$optional = $service->snapshot($actor);
$optionalChecks = array_column($optional['checks'], null, 'key');
$assert($calls === ['default', 'agents'] && $optionalChecks['redis']['status'] === 'disabled'
&& $optionalChecks['log_database']['status'] === 'disabled', 'Disabled services were probed.');
&& $optionalChecks['log_database']['status'] === 'disabled'
&& $optionalChecks['traffic_storage']['status'] === 'disabled',
'Disabled services were probed.');
$falseProbe = new CloudRuntimeHealthService(static fn () => false, static fn () => false,
static fn () => false, static fn () => []);
$assert(array_column($falseProbe->snapshot($actor)['checks'], 'status', 'key')['agents'] === 'error',
@ -187,7 +206,7 @@ $response = (new CloudAdminRuntimeController())->index($request, new Response())
$assert($response->getStatusCode() === 200
&& $response->getHeaderLine('Cache-Control') === 'private, no-store', 'Runtime response cached or unavailable.');
$decoded = json_decode((string) $response->getBody(), true);
$assert(count($decoded['checks']) === 14
$assert(count($decoded['checks']) === 15
&& array_column($decoded['checks'], 'status', 'key')['agents'] === 'error',
'Partial failure lost other checks or fabricated Agent configuration health.');
$routes = file_get_contents(dirname(__DIR__) . '/config/routes_cloud.php');

@ -0,0 +1,82 @@
<?php
require dirname(__DIR__) . '/vendor/autoload.php';
use App\Services\Cloud\CloudTrafficCheckpointRetention;
use App\Services\Cloud\CloudTrafficSettlementRetention;
use App\Services\Cloud\CloudTrafficStorageHealthService;
use Illuminate\Database\Capsule\Manager as Capsule;
use Illuminate\Database\Schema\Blueprint;
$assert = static function (bool $condition, string $message): void {
if (!$condition) {
throw new RuntimeException($message);
}
};
$capsule = new Capsule();
$capsule->addConnection([
'driver' => 'sqlite', 'database' => ':memory:',
], 'default');
$capsule->addConnection([
'driver' => 'sqlite', 'database' => ':memory:',
], 'log');
$capsule->setAsGlobal();
$capsule->bootEloquent();
$main = $capsule->schema('default');
$main->create('cloud_traffic_usage_settlement_buckets',
static function (Blueprint $table): void {
$table->increments('id');
$table->string('state');
$table->integer('updated_at');
});
$main->create('cloud_traffic_instance_cursors',
static function (Blueprint $table): void {
$table->integer('instance_id')->primary();
$table->string('state');
});
$capsule->schema('log')->create('cloud_vm_traffic_checkpoints',
static function (Blueprint $table): void {
$table->integer('instance_id');
$table->integer('first_observed_at');
});
$now = 20 * 86400;
$healthy = CloudTrafficStorageHealthService::snapshot($now);
$assert($healthy['status'] === 'ok'
&& $healthy['metrics']['settlement_buckets_due'] === 0
&& $healthy['metrics']['hourly_checkpoints_due'] === 0
&& $healthy['metrics']['cursor_attention_instances'] === 0,
'Empty compaction queues did not report a healthy bounded state.');
$settlementCutoff = $now
- CloudTrafficSettlementRetention::RETENTION_SECONDS;
$checkpointCutoff = $now
- CloudTrafficCheckpointRetention::RETENTION_SECONDS;
$capsule->connection('default')
->table('cloud_traffic_usage_settlement_buckets')->insert([
['state' => 'sealed', 'updated_at' => $settlementCutoff - 7200],
['state' => 'open', 'updated_at' => $settlementCutoff - 86400],
]);
$capsule->connection('log')->table('cloud_vm_traffic_checkpoints')->insert([
['instance_id' => 3, 'first_observed_at' => $checkpointCutoff - 7200],
['instance_id' => 4, 'first_observed_at' => $checkpointCutoff + 60],
]);
$capsule->connection('default')->table(
'cloud_traffic_instance_cursors'
)->insert([
'instance_id' => 3,
'state' => 'attention',
]);
$warning = CloudTrafficStorageHealthService::snapshot($now);
$assert($warning['status'] === 'warning'
&& $warning['metrics']['settlement_buckets_due'] === 1
&& $warning['metrics']['hourly_checkpoints_due'] === 1
&& $warning['metrics']['hourly_checkpoint_instances_due'] === 1
&& $warning['metrics']['cursor_attention_instances'] === 1
&& $warning['metrics']['oldest_due_lag_seconds'] === 7200
&& str_contains($warning['message'], '结算桶 1 个')
&& str_contains($warning['message'], '小时检查点 1 行'),
'Compaction backlog or attention state was hidden from operator health.');
echo "Traffic storage health summary contracts passed.\n";
Loading…
Cancel
Save