SynapseML
SynapseML copied to clipboard
microsoft
Adding dynamic allocation (autoscale) to LightGBM
Canceled! — The build has canceled. (cc1c0d57)
Fail! — The build has failed. (4c8bd373)
Pass! — The build has succeeded. (4c8bd373)
MMLSpark 0.13.dev34+1.g4c8bd37
This is a build for Github PR #369, changes:
- [4c8bd37] Ilya Matiach Adding dynamic allocation (autoscale) to LightGBM
-
Source: imatiach-msft/mmlspark, ilmat/dynamic-alloc at revision 4c8bd373 (by Ilya Matiach [email protected]).
-
Build: MMLSpark, 1464207 (built by elbarzil-vm on elbarzil-vm, 2018-09-10 22:30)
-
Info:
0.13.dev34+1.g4c8bd37: imatiach-msft/mmlspark/ilmat/dynamic-alloc@4c8bd373; MMLSpark#1464207 -
Queued by: Ilya Matiach for Ilya Matiach
-
Maven package uploaded, use
--packages com.microsoft.ml.spark:mmlspark_2.11:0.13.dev34+1.g4c8bd37and--repositories https://mmlspark.azureedge.net/maven. -
PIP package uploaded.
-
R package uploaded.
-
HDInsight: Copy the link to this Script Action to setup this build on a cluster.
-
ARM Template: Use this ARM Template to setup a cluster with a gpu (in the portal).
-
HDI Sub-Template: Use the HDI sub-template for just the hdi deployment (in the portal).
-
GPU VM Sub-Template: Use the GPU sub-template for just the gpu vm deployment (in the portal).
-
Convenient Deployment Script: Download this bash script or this powershell script, create a paramers file based on this template, and run as
./deploy-arm.sh ... -p <your-parameters>
-
-
Documentation uploaded.
@mhamilton723 nope I need to test it more/refactor more, sorry, will let you know when
Pass! — The build has succeeded. (9034b114)
MMLSpark 0.14.dev14+1.g9034b114
This is a build for Github PR #369, changes:
- [9034b114] Ilya Matiach Adding dynamic allocation (autoscale) to LightGBM
-
Source: imatiach-msft/mmlspark, ilmat/dynamic-alloc at revision 9034b114 (by Ilya Matiach [email protected]).
-
Build: MMLSpark, 1725480 (built by elbarzil-vm on elbarzil-vm, 2018-10-26 04:27)
-
Info:
0.14.dev14+1.g9034b114: imatiach-msft/mmlspark/ilmat/dynamic-alloc@9034b114; MMLSpark#1725480 -
Queued by: Ilya Matiach for Ilya Matiach
-
Maven package uploaded, use
--packages com.microsoft.ml.spark:mmlspark_2.11:0.14.dev14+1.g9034b114and--repositories https://mmlspark.azureedge.net/maven. -
PIP package uploaded.
-
R package uploaded.
-
HDInsight: Copy the link to this Script Action to setup this build on a cluster.
-
ARM Template: Use this ARM Template to setup a cluster with a gpu (in the portal).
-
HDI Sub-Template: Use the HDI sub-template for just the hdi deployment (in the portal).
-
GPU VM Sub-Template: Use the GPU sub-template for just the gpu vm deployment (in the portal).
-
Convenient Deployment Script: Download this bash script or this powershell script, create a paramers file based on this template, and run as
./deploy-arm.sh ... -p <your-parameters>
-
-
Documentation uploaded.
Pass! — The build has succeeded. (cdf42f3f)
MMLSpark 0.14.dev18+1.gcdf42f3f
This is a build for Github PR #369, changes:
- [cdf42f3f] Ilya Matiach Adding dynamic allocation (autoscale) to LightGBM
-
Source: imatiach-msft/mmlspark, ilmat/dynamic-alloc at revision cdf42f3f (by Ilya Matiach [email protected]).
-
Build: MMLSpark, 1744188 (built by elbarzil-vm on elbarzil-vm, 2018-10-30 05:08)
-
Info:
0.14.dev18+1.gcdf42f3f: imatiach-msft/mmlspark/ilmat/dynamic-alloc@cdf42f3f; MMLSpark#1744188 -
Queued by: Ilya Matiach for Ilya Matiach
-
Maven package uploaded, use
--packages com.microsoft.ml.spark:mmlspark_2.11:0.14.dev18+1.gcdf42f3fand--repositories https://mmlspark.azureedge.net/maven. -
PIP package uploaded.
-
R package uploaded.
-
HDInsight: Copy the link to this Script Action to setup this build on a cluster.
-
ARM Template: Use this ARM Template to setup a cluster with a gpu (in the portal).
-
HDI Sub-Template: Use the HDI sub-template for just the hdi deployment (in the portal).
-
GPU VM Sub-Template: Use the GPU sub-template for just the gpu vm deployment (in the portal).
-
Convenient Deployment Script: Download this bash script or this powershell script, create a paramers file based on this template, and run as
./deploy-arm.sh ... -p <your-parameters>
-
-
Documentation uploaded.
@peay yes but this involves a lot more testing and validation? I was hoping to use barrier execution mode from spark 2.4 instead of this PR but it doesn't support dynamic allocation, so it would just make things worse.
I see. Is this branch working? Are there some known issues or things you'd like to have tested? I can try and see if I can test it on my end if that is helpful
@peay sorry, this branch isn't close to a working state and can't be tested yet. I plan to look into this at some point though. Are you not able to disable dynamic allocation on your cluster?
Static allocation is fine for batch jobs, but is trickier for interactive work through notebooks, as it leads to ridiculously low resource utilization when we can't remove idle executors. At any rate, happy to help and do some preliminary testing once you've made more progress here!