| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
feat: introduce node-level lock to prevent concurrent pod allocation conflicts on the same node | 13 天前 | |
refactor: rename XPUDevice fields Index/Id to PhysicID/DieID | 14 天前 | |
sort the imported packages by stability and fix devil numbers | 5 个月前 | |
feat: support spread/binpack scheduling policy Core changes: - Add node/device level spread/binpack scheduling policies configured via Pod annotations: * huawei.com/vnpu-pod-node-scheduler-policy * huawei.com/vnpu-pod-device-scheduler-policy - Spread mode: node score = 200 - rawScore (soft) / 100 - rawScore (hard) - Unified device scoring function calculateDeviceScore supporting soft/hard modes - Refactor device allocation to two-phase check/apply pattern (decouple evaluation from state changes) - Support multiple device allocation per pod with ReqXPUNum Deployment: - Disable volcano default nodeorder plugin (now handled by xpu plugin) - Set enableNodeOrder: false for all volcano scheduler plugins - Disable GPU predicates (GPUSharingEnable/GPUNumberEnable) - Remove nodeorder plugin from volcano-scheduler-configmap Tests: - Add spread_test.go with comprehensive coverage (17 test functions) - Add util_test.go for scheduler policy validation Logging: - Retain 1 line node score log, remove all ===MIA=== debug logs and dead code | 16 小时前 | |
feat: support spread/binpack scheduling policy Core changes: - Add node/device level spread/binpack scheduling policies configured via Pod annotations: * huawei.com/vnpu-pod-node-scheduler-policy * huawei.com/vnpu-pod-device-scheduler-policy - Spread mode: node score = 200 - rawScore (soft) / 100 - rawScore (hard) - Unified device scoring function calculateDeviceScore supporting soft/hard modes - Refactor device allocation to two-phase check/apply pattern (decouple evaluation from state changes) - Support multiple device allocation per pod with ReqXPUNum Deployment: - Disable volcano default nodeorder plugin (now handled by xpu plugin) - Set enableNodeOrder: false for all volcano scheduler plugins - Disable GPU predicates (GPUSharingEnable/GPUNumberEnable) - Remove nodeorder plugin from volcano-scheduler-configmap Tests: - Add spread_test.go with comprehensive coverage (17 test functions) - Add util_test.go for scheduler policy validation Logging: - Retain 1 line node score log, remove all ===MIA=== debug logs and dead code | 16 小时前 | |
feat: support spread/binpack scheduling policy Core changes: - Add node/device level spread/binpack scheduling policies configured via Pod annotations: * huawei.com/vnpu-pod-node-scheduler-policy * huawei.com/vnpu-pod-device-scheduler-policy - Spread mode: node score = 200 - rawScore (soft) / 100 - rawScore (hard) - Unified device scoring function calculateDeviceScore supporting soft/hard modes - Refactor device allocation to two-phase check/apply pattern (decouple evaluation from state changes) - Support multiple device allocation per pod with ReqXPUNum Deployment: - Disable volcano default nodeorder plugin (now handled by xpu plugin) - Set enableNodeOrder: false for all volcano scheduler plugins - Disable GPU predicates (GPUSharingEnable/GPUNumberEnable) - Remove nodeorder plugin from volcano-scheduler-configmap Tests: - Add spread_test.go with comprehensive coverage (17 test functions) - Add util_test.go for scheduler policy validation Logging: - Retain 1 line node score log, remove all ===MIA=== debug logs and dead code | 16 小时前 | |
feat: introduce node-level lock to prevent concurrent pod allocation conflicts on the same node | 13 天前 | |
fix: add pipeline config | 15 天前 | |
volcano-xpu-plugin support hard mode | 4 个月前 | |
feat: support spread/binpack scheduling policy Core changes: - Add node/device level spread/binpack scheduling policies configured via Pod annotations: * huawei.com/vnpu-pod-node-scheduler-policy * huawei.com/vnpu-pod-device-scheduler-policy - Spread mode: node score = 200 - rawScore (soft) / 100 - rawScore (hard) - Unified device scoring function calculateDeviceScore supporting soft/hard modes - Refactor device allocation to two-phase check/apply pattern (decouple evaluation from state changes) - Support multiple device allocation per pod with ReqXPUNum Deployment: - Disable volcano default nodeorder plugin (now handled by xpu plugin) - Set enableNodeOrder: false for all volcano scheduler plugins - Disable GPU predicates (GPUSharingEnable/GPUNumberEnable) - Remove nodeorder plugin from volcano-scheduler-configmap Tests: - Add spread_test.go with comprehensive coverage (17 test functions) - Add util_test.go for scheduler policy validation Logging: - Retain 1 line node score log, remove all ===MIA=== debug logs and dead code | 16 小时前 | |
refactor: rename XPUDevice fields Index/Id to PhysicID/DieID | 14 天前 | |
适配修改UT | 1 个月前 | |
feat: support spread/binpack scheduling policy Core changes: - Add node/device level spread/binpack scheduling policies configured via Pod annotations: * huawei.com/vnpu-pod-node-scheduler-policy * huawei.com/vnpu-pod-device-scheduler-policy - Spread mode: node score = 200 - rawScore (soft) / 100 - rawScore (hard) - Unified device scoring function calculateDeviceScore supporting soft/hard modes - Refactor device allocation to two-phase check/apply pattern (decouple evaluation from state changes) - Support multiple device allocation per pod with ReqXPUNum Deployment: - Disable volcano default nodeorder plugin (now handled by xpu plugin) - Set enableNodeOrder: false for all volcano scheduler plugins - Disable GPU predicates (GPUSharingEnable/GPUNumberEnable) - Remove nodeorder plugin from volcano-scheduler-configmap Tests: - Add spread_test.go with comprehensive coverage (17 test functions) - Add util_test.go for scheduler policy validation Logging: - Retain 1 line node score log, remove all ===MIA=== debug logs and dead code | 16 小时前 | |
feat: support spread/binpack scheduling policy Core changes: - Add node/device level spread/binpack scheduling policies configured via Pod annotations: * huawei.com/vnpu-pod-node-scheduler-policy * huawei.com/vnpu-pod-device-scheduler-policy - Spread mode: node score = 200 - rawScore (soft) / 100 - rawScore (hard) - Unified device scoring function calculateDeviceScore supporting soft/hard modes - Refactor device allocation to two-phase check/apply pattern (decouple evaluation from state changes) - Support multiple device allocation per pod with ReqXPUNum Deployment: - Disable volcano default nodeorder plugin (now handled by xpu plugin) - Set enableNodeOrder: false for all volcano scheduler plugins - Disable GPU predicates (GPUSharingEnable/GPUNumberEnable) - Remove nodeorder plugin from volcano-scheduler-configmap Tests: - Add spread_test.go with comprehensive coverage (17 test functions) - Add util_test.go for scheduler policy validation Logging: - Retain 1 line node score log, remove all ===MIA=== debug logs and dead code | 16 小时前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 13 天前 | ||
| 14 天前 | ||
| 5 个月前 | ||
| 16 小时前 | ||
| 16 小时前 | ||
| 16 小时前 | ||
| 13 天前 | ||
| 15 天前 | ||
| 4 个月前 | ||
| 16 小时前 | ||
| 14 天前 | ||
| 1 个月前 | ||
| 16 小时前 | ||
| 16 小时前 |