Refactor and enhance QRM plugins support for reclaim_cores with numa_binding#741

Merged

xu282934741 merged 18 commits intokubewharf:mainfrom luomingmeng:dev/support-reclaimed-cores-numa-binding

Dec 17, 2024

+4,066-775

Commits on Dec 16, 2024

supports patching the Pod NUMA binding result for pods that are actually bound to specific NUMA nodes
luomingmeng
committed
cpu & memory qrm state and advisor svc support numa headroom
luomingmeng
committed
rename qrm cpu plugin vpa_test to resize_test
luomingmeng
committed
qrm cpu plugin support reclaimed with numa binding admission and provision
luomingmeng
committed
qrm memory plugin support reclaimed with numa binding admission and provision
luomingmeng
committed
supports response the Pod NUMA binding result resource annotation by memory plugin for reclaim pods that are actually bound to specific NUMA nodes
luomingmeng
committed
metric provision supports the cgroup path of reclaimed_cores with numa_binding pod
luomingmeng
committed
cpu idle support reclaim_cores with numa binding
luomingmeng
committed
memory guard support reclaim_cores with numa binding
luomingmeng
committed
pressure suppression eviction support numa binding reclaim pods
luomingmeng
committed
chore(*): refine malachite metric provisioner to skip not exists cgroup path
luomingmeng
committed
chore(metaserver): add reclaim cgroup path and reclaim numa cgroup path to k8s cgroup path
luomingmeng
committed
fix(qrm): refactor reserved reclaimed pool and revise reclaim pool in both applyBlocks and applyPoolsAndIsolatedInfo, align the core num reserved for reclaim to numa total count
luomingmeng
committed
fix(qrm): only update specified numa allocation for pods with explicit NUMA binding annotation
luomingmeng
committed
fix(qrm): add EnableReclaimNUMABinding config to support make all reclaim pods to non-RNB by default
luomingmeng
committed
chore(qrm): dry-run memset-binder, don't move page in handleAdvisorCPUSetMems and move page in adjustAllocationEntriesForReclaimedCores instead
luomingmeng
committed
fix GetActualNUMABindingResult add annotation check
luomingmeng
committed
chore(qrm): prefer binding reclaimed_cores as RNB, prioritizing memory availability over CPU.
luomingmeng
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Refactor and enhance QRM plugins support for reclaim_cores with numa_binding#741

Commits on Dec 16, 2024

supports patching the Pod NUMA binding result for pods that are actually bound to specific NUMA nodes

cpu & memory qrm state and advisor svc support numa headroom

rename qrm cpu plugin vpa_test to resize_test

qrm cpu plugin support reclaimed with numa binding admission and provision

qrm memory plugin support reclaimed with numa binding admission and provision

supports response the Pod NUMA binding result resource annotation by memory plugin for reclaim pods that are actually bound to specific NUMA nodes

metric provision supports the cgroup path of reclaimed_cores with numa_binding pod

cpu idle support reclaim_cores with numa binding

memory guard support reclaim_cores with numa binding

pressure suppression eviction support numa binding reclaim pods

chore(*): refine malachite metric provisioner to skip not exists cgroup path

chore(metaserver): add reclaim cgroup path and reclaim numa cgroup path to k8s cgroup path

fix(qrm): refactor reserved reclaimed pool and revise reclaim pool in both applyBlocks and applyPoolsAndIsolatedInfo, align the core num reserved for reclaim to numa total count

fix(qrm): only update specified numa allocation for pods with explicit NUMA binding annotation

fix(qrm): add EnableReclaimNUMABinding config to support make all reclaim pods to non-RNB by default

chore(qrm): dry-run memset-binder, don't move page in handleAdvisorCPUSetMems and move page in adjustAllocationEntriesForReclaimedCores instead

fix GetActualNUMABindingResult add annotation check

chore(qrm): prefer binding reclaimed_cores as RNB, prioritizing memory availability over CPU.