New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add doc for TF Serving #1165
Add doc for TF Serving #1165
Conversation
@lluunn: GitHub didn't allow me to request PR reviews from the following users: ynqa. Note that only kubeflow members and repo collaborators can review this PR, and authors cannot review their own PRs. In response to this: Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
/retest |
Tests timeout |
/retest |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: 0 of 1 files reviewed, 3 unresolved discussions (waiting on @lluunn and @jlewi)
docs_dev/tf_serving.md, line 3 at r1 (raw file):
# TF Serving in Kubeflow Status:
add authors and list yourself
docs_dev/tf_serving.md, line 21 at r1 (raw file):
#### HTTP proxy The [http proxy](https://github.com/kubeflow/kubeflow/tree/master/components/k8s-model-server/http-proxy)
You should explain why we have this given TF 1.8 supports it. And also mention current thinking about the plans going forward.
docs_dev/tf_serving.md, line 29 at r1 (raw file):
Check out gcr.io/kubeflow-images-public registry for other versions. #### Istio integration
I'd start a section monitoring and put this in that section. Can you also add information about promtheus integration and liveness/healthness checks?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: 0 of 1 files reviewed, 4 unresolved discussions (waiting on @lluunn and @jlewi)
docs_dev/tf_serving.md, line 35 at r1 (raw file):
See [detail](https://github.com/kubeflow/kubeflow/blob/master/components/k8s-model-server/istio-integration.md). ## Usage
Lets also add a section about Seldon. This could be as simple as an open question is how to combine TF Serving with Seldon's model management features . I think there might be some open issues about this.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: 0 of 1 files reviewed, 4 unresolved discussions (waiting on @jlewi)
docs_dev/tf_serving.md, line 3 at r1 (raw file):
Previously, jlewi (Jeremy Lewi) wrote…
add authors and list yourself
Done.
docs_dev/tf_serving.md, line 21 at r1 (raw file):
Previously, jlewi (Jeremy Lewi) wrote…
You should explain why we have this given TF 1.8 supports it. And also mention current thinking about the plans going forward.
Done.
docs_dev/tf_serving.md, line 29 at r1 (raw file):
Previously, jlewi (Jeremy Lewi) wrote…
I'd start a section monitoring and put this in that section. Can you also add information about promtheus integration and liveness/healthness checks?
Done.
docs_dev/tf_serving.md, line 35 at r1 (raw file):
Previously, jlewi (Jeremy Lewi) wrote…
Lets also add a section about Seldon. This could be as simple as an open question is how to combine TF Serving with Seldon's model management features . I think there might be some open issues about this.
Done.
Thank you, PTAL @jlewi |
docs_dev/tf_serving.md
Outdated
|
||
## SeldonIO | ||
[SeldonIO](https://github.com/SeldonIO/seldon-core) is a platform for serving ML | ||
models. It's an open question on how should we integrate TF Serving with |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
... on how should we -> Its an open question how we should
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
done
docs_dev/tf_serving.md
Outdated
Seldon to get nice features like routing and model management. | ||
|
||
## Next steps | ||
- [kubeflow/kubeflow#1000](https://github.com/kubeflow/kubeflow/issues/1000): Support logging request and response |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
How about autoscaling?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There is a benchmark issue below, which should cover it? I added some word
TF 1.8). Going forward, we will still maintain and support this http proxy as | ||
it has the ability to provide additional features such as: | ||
- exporting metrics, [kubeflow/kubeflow#1036](https://github.com/kubeflow/kubeflow/issues/1036). | ||
- request logging. See |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Are we planning on doing request logging in the http proxy? Why can't we do that in TF Serving?
Would we have to also proxy gRPC requests in order to get that functionality when gRPC is used?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think it's easier to make progress if we do it in our http proxy comparing to TF serving.
We can discuss on this thread #1000
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you add a link to the issue in the doc?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It's there
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: 0 of 1 files reviewed, 4 unresolved discussions (waiting on @jlewi and @lluunn)
docs_dev/tf_serving.md, line 64 at r2 (raw file):
Previously, lluunn (Lun-Kai Hsu) wrote…
There is a benchmark issue below, which should cover it? I added some word
Autoscaling seems different from benchmarking. Do we have an issue about autoscaling? Some of the questions are how do we take advantage of K8s autoscaling? Do we have the right metrics to trigger K8s autoscaling?
docs_dev/tf_serving.md, line 49 at r3 (raw file):
See [detail](https://github.com/kubeflow/kubeflow/blob/master/components/k8s-model-server/istio-integration.md). We are working on surfacing those metrics and integrating with Prometheus
I thought we were only planning on using Prometheus for application, TF Serving specific metrics?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: 0 of 1 files reviewed, 4 unresolved discussions (waiting on @jlewi and @lluunn)
docs_dev/tf_serving.md, line 64 at r2 (raw file):
Previously, jlewi (Jeremy Lewi) wrote…
Autoscaling seems different from benchmarking. Do we have an issue about autoscaling? Some of the questions are how do we take advantage of K8s autoscaling? Do we have the right metrics to trigger K8s autoscaling?
I see. Filed an issue and linked it.
docs_dev/tf_serving.md, line 49 at r3 (raw file):
Previously, jlewi (Jeremy Lewi) wrote…
I thought we were only planning on using Prometheus for application, TF Serving specific metrics?
I think we can do all, then we don't have to rely on Istio?
@jlewi PTAL, thanks! |
Error: No module named py. |
/retest |
1 similar comment
/retest |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: 0 of 1 files reviewed, 3 unresolved discussions (waiting on @jlewi)
docs_dev/tf_serving.md, line 49 at r3 (raw file):
(https://github.com/kubeflow/kubeflow/issues/1036
Do you mean we can do that in our proxy? Why would we want to build a custom proxy to collect telemtry metrics when this is exactly what ISTIO does?
We'd like to be able to do this for all Kubeflow microservices not just TF Serving.
/retest |
1 similar comment
/retest |
/retest |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: 0 of 1 files reviewed, 2 unresolved discussions (waiting on @jlewi)
docs_dev/tf_serving.md, line 49 at r3 (raw file):
Previously, jlewi (Jeremy Lewi) wrote…
(https://github.com/kubeflow/kubeflow/issues/1036
Do you mean we can do that in our proxy? Why would we want to build a custom proxy to collect telemtry metrics when this is exactly what ISTIO does?
We'd like to be able to do this for all Kubeflow microservices not just TF Serving.
I think the text is fine but I think we should continue to discuss whether its better to use ISTIO or not.
/lgtm |
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: jlewi 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 |
* tf serving md * fix * address comments * address review * comment
/cc @jlewi
/cc @ynqa
This change is