PAI 0.18.0 Release and Endgame Plan

[hzy46]

This issue contains our release plan for OpenPAI 0.18.0.

Release Manager

@hzy46

Endgame

• License (3.20)
• Publish sub repos (3.23, except hivedscheduler)
  • hivedscheduler: v0.3.2 @yqwang-ms
  • openpai-protocol: v2.0.0-alpha @abuccts
  • openpai-runtime: v0.1.0 @Binyang2014
  • openpaisdk: v0.1.0 @yiyione @mydmdm
  • openpaimarketplace: v1.2.0 @debuggy
  • openpaivscode: v0.3.0 @yiyione
• Remove code (1 ~ 2 days)
  • remove hivedscheduler: PR: [Hived]: Move Hived from OpenPAI to dedicated repo #4319
  • remove protocol: PR Migrate protocol spec #4307
  • remove runtime: PR [kube-runtime] remove kube-runtime #4311
  • remove openpaivscode: PR [VS Code] Remove VS Code extension code #4317
  • remove openpaisdk
  • remove openpaimarketplace PR: Remove out-dated marketplace #4324
• build process
  • sub repo CI/CD
    • hivedscheduler
    • openpai-protocol
    • openpai-runtime
    • openpaisdk
    • openpaimarketplace
    • openpaivscode
  • sub repo image build
  • OpenPAI build based on sub repo
• every sub repo has a stable release
  • hivedscheduler
  • openpai-protocol
  • openpai-runtime
  • openpaisdk
  • openpaimarketplace
  • openpaivscode
• Code freeze, create release branch (4.17)
• Endgame done (5.6)

Plan Items

• Schema change:
  • Modify gpuType to support CPU and GPU
  • Job priority
• Refine Rest-server API Issue: [REST API] refine the spec #4337 PR: refine rest swagger and codes #4355
• upgrade verification @hzy46 issue upgrading (safe deploy) & clean data doc  #1838
  • upgrade doc in PR Update docs for v0.18.0 #4238
• check installation for separate repos @hzy46
  • check dockerd config and exclude from k8s deployment
  • kubespray quick-start make registries configurable
  • kubeconfig copy issue
  • refine error message in hived config generation
  • quick-start change tag to release tag
  • include gcr.io images (k8s dashboard, scheduler) in service image build
  • registry mirrors do not work when pulling service images
  • wait for device plugin until it is ready
  • inform user of pai cluster id, configuration, username, password PR [Kubespray] Inform user of pai cluster id, configuration, username, password #4267
  • clean docker image in service-boot.sh PR Delete quick start docker images when error happens #4248
  • add FAQs and troubleshooting PR Add FAQs and Troubleshooting for kubespray #4249
• security review @qfyin
• license @qfyin
• publish separate repos
  • hivedscheduler @yqwang-ms
  • openpai-protocol @abuccts
  • openpai-runtime (ETA by 2/28) @Binyang2014
  • openpaisdk @yiyione @mydmdm
    • publish package to NPM
  • openpaimarketplace (ETA by 3/5) @debuggy
    • [ ] cherry-pick synchronizer @hzy46
  • openpaivscode @yiyione
  • openpai-nb @mydmdm @hzy46
• apt package cache in kube-runtime @hzy46 issue Add offline apt package cache in kube runtime #4211 PR Add offline apt package cache in kube runtime  #4226
• Refactor storage by leveraging persistent volumes @abuccts PR [Rest Server] Refactor storage by leveraging persistent volumes #4157
• elastic search alternative option (Azure search or DB search as switchable options) @hzy46 / @mydmdm P0.5 - elastic search alternative option (Azure search or DB search as switchable options)  #4028 PR: Use Postgresql as an alternative storage for job history #4164
• remove elasticsearch PR: Remove Elasticsearch #4304

Documentation

• Full scan of E2E document, split into small tasks for sub review if needed. @scarlett2018 @hzy46 @mydmdm P0 - Documentation ready for Beta Release #4023

BugFix

• Wrong retry log path in job history issue The path of retry log in job history is wrong #4232 PR: Fix log url in job retry page #4237 @debuggy
• API server overloaded by job detail page when containers are too many PAI become unavailable when users open too many job detail-pages  #4263 PR [Rest Server] Use list pods with label selector in getting job detail #4270 [Rest Server] Remove list pods from getting job detail #4279 @yiyione

Candidate Items

• Show gpu metrics in Grafana @Binyang2014
• hivedscheduler metric @yqwang-ms
• experience enhancement of connecting to container (ssh or other way)
  • (at least) concatenate the command e.g. ssh -i <keyfile> -p <port> root@<ip> (now user needs to copy port / ip respectively and sometimes forget root) PR Concatenate SSH command for user #4250
  • (optional) ssh key management [Edu Version] SSH Support #4094
  • (optional) connect to container directly from webportal (ssh / exec ..)
• Provide standard deep learning docker images @hzy46 issue Provide standard deep learning docker images #4227
• unit test for kube-runtime plugins
• Job API not return correct appLaunchedTime Issue Job API not return correct appLaunchedTime #4295

[VitoRazor]

Hello, thanks for your continuous efforts on openpai, which has helped us a lot.
May I ask if there is a more accurate update of v0.18.0? We can't wait to get it.

[scarlett2018]

Hello, thanks for your continuous efforts on openpai, which has helped us a lot.
May I ask if there is a more accurate update of v0.18.0? We can't wait to get it.

@VitoRazor - thanks for the comments and supports. we are working on the end game, the release is coming soon in the following week with a clear release note about the updates =).

Would you mind share more info about how your organization is currently using OpenPAI? this will helps to help you better. (feel free to skip questions, the more info you provide, the better we understand your usage. This info is only for OpenPAI to better understand the current users, we will not use this info for other purpose.).

Organization Name:

Short summary about the issue/question:

Brief what process you are following:

How to reproduce it:

OpenPAI Environment:

OpenPAI version:
Cloud provider or hardware configuration:
OS (e.g. from /etc/os-release):
Kernel (e.g. uname -a):
Hardware (e.g. core number, memory size, storage size, GPU type etc.):
Others:
Anything else we need to know: