v1.1.3

bugs fix

1. judge which task a pod belongs to according to task name instead of task type (#27)

v1.1.2

bugs fix

• global cmd flag error(#23)
• wrong pod subdomain(#24)
• incorrect to get global rank(#25)

v1.1.1

1. update status replicas and task status
2. add volumes to job spec
3. update status CompletionTimestamp when job completed

see details in #22

v1.1.0

• refactor job spec definition and add spec.tasks to support multi tasks #20
• add DI_RANK to pod env and remove engineFields in job.spec #16
• add e2e test
• add validator to validate the correctness of dijob spec
• change job.phase to Pending when job replicas scaled to 0
• implement a processor to process di-server requests
• refactor project structure

see details in #21

v1.0.0

features

• remove webhook
• manage commands with cobra
• refactor orchestrator architecture inspired from adaptdl
• use gin to rewrite di-server
• update di-server http interface
  see #18

v0.2.2

bug fix

• resolve bug that job failed to submit when collector/learner missed (#14)

v0.2.1

feature

• add event handler for dijob, and mark job as Created when job submitted(#13)
• mark collector and learner as optional, only coordinator is required(653e64a)
• mark job Failed when the submitted job is incorrect(bea840a), but it's hard to test since client-go reflector decodes DIJob strictly, we have no chance to handle DIJob add event when incorrect job submitted

v0.2.0

1. change orchestrator image repository
2. version -> v0.2.0

v0.2.0-rc.0 release

• split webhook and operator
• add dockerfile.dev
• update CleanPolicyALL to CleanPolicyAll
• remove k8s service related operations from server, and operator is responsible for managing services
• add e2e test

v0.1.0 release

Features

• Define DIJob CRD to support DI jobs' submission
• Define AggregatorConfig CRD to support aggregator definition
• Add webhook to validate DIJob submission
• Provide http service for DI jobs to request for DI modules
• Docs to introduce DI-orchestrator architecture