wangbo 3aa78d3e5c perf(worker): 按实例内存上限扩展异步容量 ...

将策略推导出的集群目标与单 Worker 内存安全上限分离，避免失活实例的全部容量转移到单个 2 GiB Pod。

生产环境单实例上限和媒体物化并发设为 24；两实例正常时总容量由 16 提升至 48，后续可通过增加 Worker 节点继续扩展。新增 0093 迁移保存实例容量上限，并兼容滚动发布中的旧实例。

风险：更高媒体并发会增加 Worker 和节点内存压力，保留单实例 24 的硬边界并由持续监控观察 RSS、OOM、队列与节点余量。

验证：go test ./... -count=1；go vet ./...；真实 PostgreSQL 分配与故障转移测试；迁移安全检查；kubectl kustomize；gofmt -l；git diff --check。

2026-07-30 15:08:54 +08:00

..

admission_test.go

feat(queue): 增加非文本模型分布式准入队列

2026-07-29 16:15:43 +08:00

admission.go

perf(queue): 限制大媒体任务内存并消除准入惊群

2026-07-30 13:13:32 +08:00

billing_mode_test.go

feat(billing): 完成异步结算与请求执行闭环

2026-07-21 00:25:53 +08:00

billing_settlement_worker_test.go

feat(billing): 完成异步结算与请求执行闭环

2026-07-21 00:25:53 +08:00

billing_settlement_worker.go

feat(billing): 完成异步结算与请求执行闭环

2026-07-21 00:25:53 +08:00

binary_results_test.go

perf(storage): 拦截任务二进制并本地暂存结果

2026-07-24 21:13:09 +08:00

binary_results.go

perf(storage): 拦截任务二进制并本地暂存结果

2026-07-24 21:13:09 +08:00

cache_affinity_test.go

fix(routing): 对齐缓存亲和力硬规则与审计

2026-07-24 23:51:27 +08:00

cache_affinity.go

fix(routing): 对齐缓存亲和力硬规则与审计

2026-07-24 23:51:27 +08:00

candidate_filter_test.go

Handle Gemini inline data in asset and size processing

2026-06-15 00:14:40 +08:00

candidate_filter.go

fix(media): 规范比例与 OpenAI 图像尺寸参数

2026-07-27 23:50:17 +08:00

candidate_platform_filter_test.go

fix(runtime): honor requested platform for routing

2026-07-24 07:56:48 +08:00

candidate_platform_filter.go

fix(runtime): honor requested platform for routing

2026-07-24 07:56:48 +08:00

credential_env_test.go

feat(gateway): 补齐桌面端高级媒体直连接口

2026-07-22 14:02:53 +08:00

execution_lease.go

fix(billing): 封住发布前计费竞态

2026-07-21 10:23:58 +08:00

helpers.go

Add runtime restore and temp asset cleanup

2026-06-05 20:28:26 +08:00

input_image_normalizer_test.go

feat(seedance): 同步输入图片约束与自动转换

2026-07-23 21:55:42 +08:00

input_image_normalizer.go

feat(seedance): 同步输入图片约束与自动转换

2026-07-23 21:55:42 +08:00

keling_audio_test.go

feat(kling): 接入 Omni 视频兼容接口

2026-07-22 00:25:05 +08:00

limits_test.go

perf(queue): 按策略动态扩缩异步 Worker

2026-07-24 12:26:56 +08:00

limits.go

feat(queue): 增加非文本模型分布式准入队列

2026-07-29 16:15:43 +08:00

media_parameter_normalization_test.go

fix(media): 规范比例与 OpenAI 图像尺寸参数

2026-07-27 23:50:17 +08:00

output_token_limit_test.go

fix(api): 缺失输出能力配置时透传请求

2026-07-23 11:07:31 +08:00

output_token_limit.go

fix(api): 缺失输出能力配置时透传请求

2026-07-23 11:07:31 +08:00

param_processor_media.go

fix(media): 规范比例与 OpenAI 图像尺寸参数

2026-07-27 23:50:17 +08:00

param_processor_message.go

Split param processor and tighten Volces frame validation

2026-05-14 00:41:06 +08:00

param_processor_script_test.go

feat(api): migrate media clients and universal scripts

2026-05-20 21:29:07 +08:00

param_processor_script.go

feat(api): migrate media clients and universal scripts

2026-05-20 21:29:07 +08:00

param_processor_test.go

feat(gateway): 补齐桌面端高级媒体直连接口

2026-07-22 14:02:53 +08:00

param_processor_utils.go

fix(media): 规范比例与 OpenAI 图像尺寸参数

2026-07-27 23:50:17 +08:00

param_processor_video_content.go

Split param processor and tighten Volces frame validation

2026-05-14 00:41:06 +08:00

param_processor.go

feat: 完善模型请求适配与输出限制

2026-07-17 13:52:00 +08:00

portrait_assets_test.go

feat(seedance): 接入真人资产管理与引用

2026-07-22 00:25:22 +08:00

portrait_assets.go

feat(seedance): 接入真人资产管理与引用

2026-07-22 00:25:22 +08:00

pricing_test.go

feat(gateway): 补齐桌面端高级媒体直连接口

2026-07-22 14:02:53 +08:00

pricing_v2_test.go

fix(billing): 视频时长按实际秒数线性计费

2026-07-21 13:32:23 +08:00

pricing_v2.go

feat(gateway): 补齐桌面端高级媒体直连接口

2026-07-22 14:02:53 +08:00

pricing.go

fix(runtime): honor requested platform for routing

2026-07-24 07:56:48 +08:00

proxy_test.go

perf(queue): 限制大媒体任务内存并消除准入惊群

2026-07-30 13:13:32 +08:00

proxy.go

perf(queue): 限制大媒体任务内存并消除准入惊群

2026-07-30 13:13:32 +08:00

queue_client_test.go

fix(queue): 自动救援失活 Worker 遗留任务

2026-07-29 23:44:29 +08:00

queue_worker_manager_test.go

fix(worker): 增强队列启动与租约到期恢复

2026-07-29 23:55:49 +08:00

queue_worker_test.go

fix(runner): 稳定异步任务重启恢复

2026-07-21 11:29:56 +08:00

queue_worker.go

perf(worker): 按实例内存上限扩展异步容量

2026-07-30 15:08:54 +08:00

recording_test.go

fix(gemini): 移除空返图占位并暴露失败原因

2026-07-29 19:50:43 +08:00

recording.go

fix(gemini): 移除空返图占位并暴露失败原因

2026-07-29 19:50:43 +08:00

request_assets_test.go

fix(audio): 兼容百炼 OpenAI 音频输入

2026-07-27 23:50:17 +08:00

request_assets.go

fix(images): 优先转换编辑图片为 multipart 二进制

2026-07-24 09:04:23 +08:00

request_context_test.go

feat(queue): 增加非文本模型分布式准入队列

2026-07-29 16:15:43 +08:00

request_context.go

feat(queue): 增加非文本模型分布式准入队列

2026-07-29 16:15:43 +08:00

responses_test.go

feat: add Responses API compatibility

2026-07-10 23:33:15 +08:00

responses.go

feat: add Responses API compatibility

2026-07-10 23:33:15 +08:00

retry_decision_test.go

refactor(runner): 收敛上游失败决策与轮转策略

2026-07-27 23:50:17 +08:00

retry_decision.go

refactor(runner): 收敛上游失败决策与轮转策略

2026-07-27 23:50:17 +08:00

runtime_policy.go

refactor(runner): 收敛上游失败决策与轮转策略

2026-07-27 23:50:17 +08:00

service_test.go

feat: 完善模型请求适配与输出限制

2026-07-17 13:52:00 +08:00

service.go

perf(worker): 按实例内存上限扩展异步容量

2026-07-30 15:08:54 +08:00

task_cancel_test.go

feat(task): 暴露任务取消控制状态

2026-06-25 17:01:32 +08:00

task_cancel.go

perf(queue): 按策略动态扩缩异步 Worker

2026-07-24 12:26:56 +08:00

task_history_workers_test.go

perf(storage): 极简化任务历史并增加保留治理

2026-07-24 18:23:22 +08:00

task_history_workers.go

perf(storage): 极简化任务历史并增加保留治理

2026-07-24 18:23:22 +08:00

trace.go

refactor(runner): 收敛上游失败决策与轮转策略

2026-07-27 23:50:17 +08:00

upload_test.go

perf(queue): 限制大媒体任务内存并消除准入惊群

2026-07-30 13:13:32 +08:00

upload.go

perf(queue): 限制大媒体任务内存并消除准入惊群

2026-07-30 13:13:32 +08:00

video_duration.go

fix: prefer generated video metadata for billing

2026-05-20 00:11:28 +08:00

voice_clone.go

perf(storage): 极简化任务历史并增加保留治理

2026-07-24 18:23:22 +08:00

wallet_execute_test.go

feat(billing): 完成异步结算与请求执行闭环

2026-07-21 00:25:53 +08:00