perf(queue): 按策略动态扩缩异步 Worker

将 River 执行容量按平台模型与用户组的有效并发策略动态调整,并为长任务续租、并发租约原子抢占和限流退避补充保护。\n\n统一平台模型限流继承语义,兼容历史 platformLimits/modelLimits,并为三个迁移图像模型建立独立并发租约。补充管理端显式继承/覆盖配置、指标、单元测试及隔离 PostgreSQL 验收。\n\n验证:go test ./... -count=1;pnpm lint;pnpm test;pnpm build;隔离 PostgreSQL 并发原子性/续租测试;128 任务与三分钟长任务动态 Worker 验收。
This commit is contained in:
2026-07-24 12:26:56 +08:00
parent 290b8c1854
commit 6c5daf29ca
37 changed files with 2365 additions and 205 deletions
@@ -409,13 +409,13 @@ function ModelBindingPolicy(props: { form: PlatformWizardForm; onChange: (value:
</Label>
</>
)}
<ToggleField checked={form.modelOverrideRateLimit} label="覆盖模型限流策略" onChange={(checked) => onChange({ ...form, modelOverrideRateLimit: checked })} />
<ToggleField checked={form.modelOverrideRateLimit} label="覆盖模型限流策略" onChange={(checked) => onChange({ ...form, modelOverrideRateLimit: checked, modelRateLimitOverrideTouched: true })} />
{form.modelOverrideRateLimit && (
<>
<Label> RPM<Input value={form.modelRpmLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelRpmLimit: event.target.value })} /></Label>
<Label> RPS<Input value={form.modelRpsLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelRpsLimit: event.target.value })} /></Label>
<Label> TPM<Input value={form.modelTpmLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelTpmLimit: event.target.value })} /></Label>
<Label><Input value={form.modelConcurrencyLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelConcurrencyLimit: event.target.value })} /></Label>
<Label> RPM<Input value={form.modelRpmLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelRpmLimit: event.target.value, modelRateLimitOverrideTouched: true })} /></Label>
<Label> RPS<Input value={form.modelRpsLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelRpsLimit: event.target.value, modelRateLimitOverrideTouched: true })} /></Label>
<Label> TPM<Input value={form.modelTpmLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelTpmLimit: event.target.value, modelRateLimitOverrideTouched: true })} /></Label>
<Label><Input value={form.modelConcurrencyLimit} placeholder="不填则不限制" inputMode="numeric" onChange={(event) => onChange({ ...form, modelConcurrencyLimit: event.target.value, modelRateLimitOverrideTouched: true })} /></Label>
</>
)}
</div>
@@ -1004,6 +1004,7 @@ function platformToForm(
const rateLimitPolicy = platform.rateLimitPolicy ?? {};
const networkProxy = readNetworkProxyConfig(config);
const currentModels = platformModels.filter((model) => model.platformId === platform.id);
const modelRateLimitPolicy = currentModels.find((model) => model.rateLimitPolicyMode === 'override')?.rateLimitPolicy ?? {};
return {
...createEmptyPlatformForm(platform.provider, defaults),
provider: platform.provider,
@@ -1034,6 +1035,12 @@ function platformToForm(
selectedModelIds: platformModelBaseIds(platform, baseModels, currentModels),
modelDiscountFactors: platformModelDiscountFactors(platform, baseModels, currentModels),
modelNameMappings: platformModelNameMappings(platform, baseModels, currentModels),
modelOverrideRateLimit: currentModels.length > 0 && currentModels.every((model) => model.rateLimitPolicyMode === 'override'),
modelRateLimitOverrideTouched: false,
modelRpmLimit: readLimit(modelRateLimitPolicy, 'rpm'),
modelRpsLimit: readLimit(modelRateLimitPolicy, 'rps'),
modelTpmLimit: readLimit(modelRateLimitPolicy, 'tpm_total'),
modelConcurrencyLimit: readLimit(modelRateLimitPolicy, 'concurrent'),
selectionMode: 'partial',
};
}