support enable share gpu for reclaimed#12

Open

luomingmeng wants to merge 52 commits into

JustinChengLZ:dev/support-gpu-pluginsfrom

luomingmeng:dev/support-enable-share-gpu-for-reclaimed

Commits on Dec 25, 2025

feat(gpu): add static policy implementation for GPU resource management
luomingmeng
committed
refactor(topology): skip add zone node which is not a child of socket or numa zone node
luomingmeng
committed
feat(gpu): enhance GPU memory allocation with NUMA awareness
luomingmeng
committed
refactor(cpu): remove unused preferredHintIndexes variable
luomingmeng
committed
feat(resource-plugin): add associated device allocation support
luomingmeng
committed
refactor(qrm-plugins): embed UnimplementedResourcePluginServer in policy structs
luomingmeng
committed
feat(gpu): add associated device topology hints support
luomingmeng
committed
fix typo and add logs
luomingmeng
committed
refactor(gpu): remove redundant non-numa-affinity gpu allocation logic
luomingmeng
committed
feat(gpu): optimize GPU allocation by preferring NUMA nodes with most allocated memory
luomingmeng
committed
feat: refactor code into resource plugins and custom device plugins

JustinChengLZ
authored and
luomingmeng
committed
chore: add unit tests

JustinChengLZ
authored and
luomingmeng
committed
feat: introduce rdma state and allow states to share within gpu sub-plugins

JustinChengLZ
authored and
luomingmeng
committed
feat: refactor state to only be in one file

JustinChengLZ
authored and
luomingmeng
committed
feat: implement rdma custom device plugin and implement logic for accompany resource allocation

JustinChengLZ
authored and
luomingmeng
committed
feat: implement allocation of accompany resource first before device

JustinChengLZ
authored and
luomingmeng
committed
Update gpu_plugin.go

JustinChengLZ
authored and
luomingmeng
committed
refactor(gpu): restructure device plugin and resource management
luomingmeng
committed
refactor: remove unused GenerateDummyGPUTopology function
luomingmeng
committed
feat(gpu): implement strategy-based GPU allocation framework
luomingmeng
committed
refactor(gpu-strategy): reorganize gpu allocation strategy components
luomingmeng
committed
refactor(gpu-strategy): make strategy fields private and add accessors
luomingmeng
committed
feat(device): add device affinity group support
luomingmeng
committed
feat: develop device affinity binding and filtering strategies

JustinChengLZ
authored and
luomingmeng
committed
feat: implement binding strategy to prioritise device affinity during allocation

JustinChengLZ
authored and
luomingmeng
committed
refactor(gpu): restructure GPU strategy and state management
luomingmeng
committed
chore: rebase katalyst-api

JustinChengLZ
authored and
luomingmeng
committed
chore: fix unit test, format and lint issues

JustinChengLZ
authored and
luomingmeng
committed
fix: maintain affinity subgroup sequence in larger affinity groups

JustinChengLZ
authored and
luomingmeng
committed
refactor: simplify code by deleting redundant parameters and refactor function for allocating devices

JustinChengLZ
authored and
luomingmeng
committed
refactor: make allocation recursive to simplify logic

JustinChengLZ
authored and
luomingmeng
committed
fix(gpu): handle NUMA node edge cases and improve logging
luomingmeng
committed
feat(gpu): add canonical strategy implementation and refactor gpu memory strategy
luomingmeng
committed
refactor(gpu/strategy): optimize device affinity allocation algorithm
luomingmeng
committed
refactor: simplify the grouping of device affinity

JustinChengLZ
authored and
luomingmeng
committed
fix: handling of nil device req

JustinChengLZ
authored and
luomingmeng
committed
chore: add unit tests

JustinChengLZ
authored and
luomingmeng
committed
refactor(gpumemory): move nil device request check after qos validation
luomingmeng
committed
fix(gpu): skip zero requests in GetGPUCount and optimize logging
luomingmeng
committed
feat(gpumemory): add numa binding check and health status filter
luomingmeng
committed
fix(gpumemory): handle unhealthy devices and correct capacity values
luomingmeng
committed
refactor(qrm): remove unused state file directory fields
luomingmeng
committed
fix(gpumemory): handle numa topology not ready case gracefully
luomingmeng
committed
chore: add context to interface methods

JustinChengLZ
authored and
luomingmeng
committed
feat(gpu): add ShareGPUManager for device sharing eligibility
luomingmeng
committed
feat(eviction): add reclaimed GPU resources eviction plugin
luomingmeng
committed
build: update katalyst-api dependency version
luomingmeng
committed
fix(resources-eviction): add error healthz state update in threshold checks
luomingmeng
committed

Commits on Jan 12, 2026

Commits on Jan 13, 2026

fix(gpumemory): correct gpu memory allocation calculation
luomingmeng
committed

Commits on Jan 14, 2026

feat(evictionmanager): add GPU-specific threshold tolerance duration
luomingmeng
committed