Skip to content

Latest commit

 

History

History
245 lines (195 loc) · 6.68 KB

CHANGELOG.md

File metadata and controls

245 lines (195 loc) · 6.68 KB

Feat

  • add timeout option waiting for all resource server ready
  • upgrade vcuda

Fix

  • virutal manager can't probe correct vm controller path
  • wrong size of device memory when app has more than 1 card
  • vcuda image repository url
  • the mismatch between gpu-manager pick up and gpu-admission predicate (#74)
  • 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

v1.1.5 - 2021-05-10

Docs

  • Add FAQ link
  • Update gpu manager yaml

Feat

  • upgrade vcuda to 1.0.3
  • Upgrade vcuda-controller to v1.0.1
  • Use host network to build image
  • Update go version to 1.14.3
  • Support CRI interface

Fix

  • kubelet 1.20 device checkpoint support (#62)
  • the mismatch between gpu-manager pick up and gpu-admission predicate (#74)
  • read QoS class from pod status first
  • report device memory when allocate more than one cards
  • 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

v1.0.9 - 2021-02-23

Feat

  • use apiserver cache to list pod

Fix

  • ignore not running container while recovering

v1.0.8 - 2021-02-22

Feat

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

Fix

  • missing recover tree data if information is retrieved from checkpoint file
  • 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

v1.1.4 - 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