Skip to content

Latest commit

 

History

History
174 lines (136 loc) · 4.39 KB

CHANGELOG.md

File metadata and controls

174 lines (136 loc) · 4.39 KB
Raw

Unreleased

Feat

  • upgrade vcuda

Fix

  • read QoS class from pod status first
  • kubelet 1.20 device checkpoint support (#62)
  • report device memory when allocate more than one cards
  • DeviceGetTopologyCommonAncestor get a zero value on multi-gpu board
  • gpu-manager lost checkpoint data file
  • preserve attribute
  • read cgroup.procs files recursively
  • wait server until it's ready

v1.1.4 - 2021-02-05

Fix

  • read QoS class from pod status first

v1.1.3 - 2021-02-02

Feat

  • upgrade vcuda to 1.0.3

Fix

  • report device memory when allocate more than one cards

v1.1.2 - 2020-12-09

Docs

  • Add FAQ link
  • Update gpu manager yaml

Feat

  • Upgrade vcuda-controller to v1.0.1
  • Use host network to build image
  • Update go version to 1.14.3
  • Support CRI interface

Fix

  • gpu-manager lost checkpoint data file
  • DeviceGetTopologyCommonAncestor get a zero value on multi-gpu board
  • preserve attribute
  • read cgroup.procs files recursively
  • wait server until it's ready
  • Revert using vendor directory
  • Allow non-root user to communicate with gpu manager
  • Change ius rpm broken link
  • skip symlink when copy bin to |${NV_DIR}|. (#15)

Refact

  • Use vendor directory
  • Refact gpu-manager code

v1.0.7 - 2020-12-09

Feat

  • Upgrade vcuda-controller to v1.0.2
  • Use host network to build image
  • Upgrade vcuda-controller to v1.0.1

Fix

  • gpu-manager lost checkpoint data file
  • DeviceGetTopologyCommonAncestor get a zero value on multi-gpu board
  • preserve attribute
  • upgrade go to 1.15
  • wait server until it's ready
  • Change ius rpm broken link
  • Allow non-root user to communicate with gpu manager

Refact

  • only watch pod belong this node

v1.1.1 - 2020-12-02

Docs

  • Add FAQ link
  • Update gpu manager yaml

Feat

  • Upgrade vcuda-controller to v1.0.1
  • Use host network to build image
  • Update go version to 1.14.3
  • Support CRI interface

Fix

  • DeviceGetTopologyCommonAncestor get a zero value on multi-gpu board
  • preserve attribute
  • read cgroup.procs files recursively
  • wait server until it's ready
  • Revert using vendor directory
  • Allow non-root user to communicate with gpu manager
  • Change ius rpm broken link
  • skip symlink when copy bin to |${NV_DIR}|. (#15)

Refact

  • Use vendor directory
  • Refact gpu-manager code

v1.0.6 - 2020-12-02

Fix

  • DeviceGetTopologyCommonAncestor get a zero value on multi-gpu board
  • preserve attribute

v1.0.5 - 2020-08-28

Feat

  • Upgrade vcuda-controller to v1.0.2

Fix

  • upgrade go to 1.15
  • wait server until it's ready

Refact

  • only watch pod belong this node

v1.0.4 - 2020-05-21

Feat

  • Use host network to build image
  • Upgrade vcuda-controller to v1.0.1

Fix

  • Change ius rpm broken link
  • Allow non-root user to communicate with gpu manager

v1.1.0 - 2020-05-21

Docs

  • Add FAQ link
  • Update gpu manager yaml

Feat

  • Upgrade vcuda-controller to v1.0.1
  • Use host network to build image
  • Update go version to 1.14.3
  • Support CRI interface

Fix

  • Revert using vendor directory
  • Allow non-root user to communicate with gpu manager
  • Change ius rpm broken link
  • skip symlink when copy bin to |${NV_DIR}|. (#15)

Refact

  • Use vendor directory
  • Refact gpu-manager code

v1.0.3 - 2019-12-17