Updated the image in TFMA component #1106
Conversation
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: If they are not already assigned, you can assign the PR to them by writing The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
|
Could you elaborate this change in the image? |
It's the bug fix we released on Friday: #1094 |
|
Why not update all the component image tags such that it looks more consistent |
I'm not sure it's a good idea to update all component images and component definitions every time someone make a change to a single component. |
|
Agreed that we need not build new component images but since we are releasing the components all at one time using the release script, why not keep the image tag consistent? |
I did not want any more big somewhat untested changes right before NEXT. The reality is that the images being released are not passing through testing (we release the images that are build by CloudBuild and do not go through testing). So, I was trying not to make any unnecessary sweeping changes.
|
|
The image was updated the next week using the normal release process. |
KFServer uses the number of cores on the machine to set the number of process in its tornado service which can delay the initialization time and increase the memory usage a lot. A local testing suggests that KFServer loaded 64 processes and used more than 500Mb for a simple SKLearn model. Once I set the worker number to 1, the memory usage was reduced to around 100Mb. ``` [I 200927 18:26:06 kfserver:88] Registering model: mlguild-iris-svm [I 200927 18:26:06 kfserver:77] Listening on port 8080 [I 200927 18:26:06 kfserver:79] Will fork 0 workers [I 200927 18:26:06 process:126] Starting 64 processes ```
Incorporates the fix in #1094.
This change is