frameworkcontroller.yaml

# Put it directly under frameworkcontroller's current working directory.
# For the full config setting and usage, see ./pkg/apis/frameworkcontroller/v1/config.go

# This is the default config for frameworkcontroller, so most settings are commented out.

#kubeApiServerAddress: http://10.10.10.10:8080
#kubeConfigFilePath: ''

# Based on the --max-mutating-requests-inflight is 500.
#kubeClientQps: 200
#kubeClientBurst: 300
#workerNumber: 500

#largeFrameworkCompression: true

#frameworkCompletedRetainSec: 2592000

#frameworkMinRetryDelaySecForTransientConflictFailed: 60
#frameworkMaxRetryDelaySecForTransientConflictFailed: 900

podFailureSpec:
################################################################################
# [-1199, -1000]: K8S issued failures
################################################################################
- code: -1000
  phrase: PodEvicted
  type:
    attributes: [Transient]
  podPatterns:
  - reasonRegex: '(?i)^Evicted$'
    messageRegex: '(?ms).*'
- code: -1001
  phrase: PodNodeLost
  type:
    attributes: [Transient]
  podPatterns:
  - reasonRegex: '(?i)^NodeLost$'
    messageRegex: '(?ms).*'
- code: -1002
  phrase: PodScheduledToInsufficientResourceNode
  type:
    attributes: [Transient]
  podPatterns:
  - reasonRegex: '(?i)^OutOf\S+$'
    messageRegex: '(?ms).*'
- code: -1003
  phrase: PodPreemptedForCriticalPod
  type:
    attributes: [Transient]
  podPatterns:
  - reasonRegex: '(?i)^Preempting$'
    messageRegex: '(?ms).*'
- code: -1004
  phrase: PodDeadlineExceeded
  type:
    attributes: [Permanent]
  podPatterns:
  - reasonRegex: '(?i)^DeadlineExceeded$'
    messageRegex: '(?ms).*'
- code: -1005
  phrase: PodNodeAdmissionForbidden
  type:
    attributes: [Permanent]
  podPatterns:
  - reasonRegex: '(?i)^Forbidden$'
    messageRegex: '(?ms).*'
- code: -1006
  phrase: PodNodeAdmissionUnexpectedError
  type:
    attributes: [Transient]
  podPatterns:
  - reasonRegex: '(?i)^UnexpectedAdmissionError$'
    messageRegex: '(?ms).*'
  - reasonRegex: '(?i)^UnknownReason$'
    messageRegex: '(?ms).*'
  - reasonRegex: '(?i)^InvalidNodeInfo$'
    messageRegex: '(?ms).*'
  - reasonRegex: '(?i)^UnexpectedPredicateFailureType$'
    messageRegex: '(?ms).*'

################################################################################
# [-1399, -1200]: Docker issued failures
################################################################################
- code: -1200
  phrase: ContainerDockerOOMKilled
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - reasonRegex: '(?i)^OOMKilled$'
      codeRange: {min: 1}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: -1201
  phrase: ContainerDockerCreationError
  podPatterns:
  - containers:
    - reasonRegex: '(?i)^ContainerCannotRun$'
      codeRange: {min: 125, max: 125}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: -1202
  phrase: ContainerDockerStartCmdPermissionDenied
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - reasonRegex: '(?i)^ContainerCannotRun$'
      codeRange: {min: 126, max: 126}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: -1203
  phrase: ContainerDockerStartCmdNotFound
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - reasonRegex: '(?i)^ContainerCannotRun$'
      codeRange: {min: 127, max: 127}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
  - containers:
    - reasonRegex: '(?i)^ContainerCannotRun$'
      codeRange: {min: 128, max: 128}
      nameRegex: '(?ms).*'
      messageRegex: '(?msi).*(not found|cannot find|no such).*'
- code: -1204
  phrase: ContainerDockerStartUnknownError
  podPatterns:
  - containers:
    - reasonRegex: '(?i)^ContainerCannotRun$'
      codeRange: {min: 128, max: 128}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'

################################################################################
# [1, 255]: User Container issued failures
################################################################################
# [129, 192]: Involuntary failures caused by OS Signal
- code: 130
  phrase: ContainerSigIntReceived
  type:
    attributes: [Transient]
  podPatterns:
  - containers:
    - codeRange: {min: 130, max: 130}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 131
  phrase: ContainerSigQuitReceived
  type:
    attributes: [Transient]
  podPatterns:
  - containers:
    - codeRange: {min: 131, max: 131}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 132
  phrase: ContainerSigIllReceived
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - codeRange: {min: 132, max: 132}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 134
  phrase: ContainerSigAbrtReceived
  podPatterns:
  - containers:
    - codeRange: {min: 134, max: 134}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 135
  phrase: ContainerSigBusReceived
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - codeRange: {min: 135, max: 135}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 136
  phrase: ContainerSigFpeReceived
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - codeRange: {min: 136, max: 136}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 137
  phrase: ContainerSigKillReceived
  type:
    attributes: [Transient]
  podPatterns:
  - containers:
    - codeRange: {min: 137, max: 137}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 139
  phrase: ContainerSigSegvReceived
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - codeRange: {min: 139, max: 139}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 141
  phrase: ContainerSigPipeReceived
  type:
    attributes: [Permanent]
  podPatterns:
  - containers:
    - codeRange: {min: 141, max: 141}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'
- code: 143
  phrase: ContainerSigTermReceived
  type:
    attributes: [Transient]
  podPatterns:
  - containers:
    - codeRange: {min: 143, max: 143}
      nameRegex: '(?ms).*'
      messageRegex: '(?ms).*'

# [1, 255] - [129, 192]: Voluntary failures caused by Container itself
# [200, 219]: Reserved Codes
# [1, 255] - [129, 192] - [200, 219]: Custom Codes
# Example: Directly forwarding Container code and just adding type info.
#- code: 220
#  phrase: Container220Failed
#  type:
#    attributes: [Permanent]
#  podPatterns:
#  - containers:
#    - codeRange: {min: 220, max: 220}
#      nameRegex: '(?ms).*'
#      messageRegex: '(?ms).*'

# Example: Classification only based on Container termination message.
#- code: 221
#  phrase: ContainerTensorflowOOMKilled
#  type:
#    attributes: [Permanent]
#  podPatterns:
#  - containers:
#    - messageRegex: '(?msi)tensorflow.*ResourceExhaustedError.*OOM.*'
#      codeRange: {min: 1}
#      nameRegex: '(?ms).*'
#- code: 222
#  phrase: ContainerMPISegvFault
#  type:
#    attributes: [Permanent]
#  podPatterns:
#  - containers:
#    - messageRegex: '(?msi)Signal code: Address not mapped.*'
#      codeRange: {min: 1}
#      nameRegex: '(?ms).*'
#- code: 223
#  phrase: ContainerCudaUncorrectableECCError
#  type:
#    attributes: [Transient]
#  podPatterns:
#  - containers:
#    - messageRegex: '(?msi)CUDA_ERROR_ECC_UNCORRECTABLE.*'
#      codeRange: {min: 1}
#      nameRegex: '(?ms).*'

# Example: Redirect all unknown failures to a single comparable code.
#- code: 255
#  phrase: ContainerUnknownFailed
#  podPatterns:
#  - containers:
#    - codeRange: {min: 1}
#      nameRegex: '(?ms).*'
#      messageRegex: '(?ms).*'