Iteration Plan for March 2019

[scarlett2018]

This plan captures our work in March. This is a 4 week iteration, and we will ship our March Update in in early April. We'll continue work we started in the previous iteration, and we started to pick up items from our roadmap.

Endgame

• March 15 20, 2019 Code freeze for the endgame
• March 29 April 3, 2019 Endgame done

The endgame details for this iteration are tracked in #2307 .

Plan Items

Below is a summary of the top level plan items.

Manageability and operation supports

• Detect and alert for unhealthy GPU Need alarms for unhealthy GPU cases #2192 @mzmssg @xudifsd
• Provide the ability to query all the jobs in a Node in PAI Web Portal Need to know all the jobs in a Node #2128 @xudifsd
• Team wise storage management support Team wise storage for NFS data storage #2204 @ydye @wangdian

Better Installation and upgrading experience

• Backward compatible upgrading PAI would officially support upgrade from previously installation after v0.10 (inclusive) #2212 @hao1939 {moved to 0.10.1}

Usability

• Simplest Job Submission Page for Job (types) that will only run on 1 node Simplified Job Templates for NFS configured OpenPAI #2238 @debuggy @Gerhut
• Provide filters for jobs list page Provide filter for jobs list table #302 @Gerhut
• Refine Job detail page Refine Job Details Page Phase I #2211 @sunqinzheng @qfyin

Debuggability

• Basic supports for reserving container's life time for debug after job completion [Job Debugging] Container reservation when the job failed due to users' error. #2213 [Job Debugging] An option for user to enable debugging reservation #2214 @ydye

Sunset supports

• Remove TensorFlow mpi example [Docs] Remove TensorFlow mpi example #2337 @abuccts

Bug Fixes

• hadoop-ai can't detect ECC error hadoop-ai can't detect ECC error #2146 @mzmssg
• Support nvidia-container-runtime for gpu isolation Support nvidia-container-runtime for gpu isolation #2352 @abuccts

---

Pending Items

N/A

Deferred Items - last updated 3/13/2019

• Display resource utility per vc/queue metrics in grafana [Close as Dup] display resource utility per vc/queue metrics #2208 @xudifsd
• Mark utilization for each job in the Node Mark utilization for each job in the Node #2250 @xudifsd
• Instead of setting the GPU/memory/core manually and separately, PAI expose simple SKUs that are available in this PAI instance for user to select As a user, I need to know the available GPU's type (sku) so that I can decide whether the available GPU are fit for my training. #2062 @debuggy @qfyin

[xudifsd]

user also request #2208 to be implemented.

[scarlett2018]

user also request #2208 to be implemented.

Thanks @xudifsd , updated.

[abuccts]

Add #2232.

[xwzheng1020]

Add #2229

[scarlett2018]

Adding #2337 to the release

[scarlett2018]

Adding #2146 fix to the release