update catalog image unpacking process #2

everettraven · 2023-02-01T20:33:21Z

Description:
Updates the catalog image unpacking process to use a Job for unpacking the image contents to STDOUT using opm render <catalog-image> and then scraping the pod logs to use during the reconciliation process and creation of child resources.

A brief overview of the changes:

A couple new functions in catalogsource_controller.go and removal of the containerdregistry approach
go mod tidy related updates
Introduction of a new UnpackPhaseError type that is used to recognize when a Job pod is not yet in the Succeeded state.
A couple manifest updates to ensure the controller has the proper permissions to run as expected
A couple helper scripts for testing:
- reset.sh deletes and recreates a KinD cluster and then applies the necessary manifests to get everything up and running
- timetest.sh can be used by running time ./timetest.sh to evaluate the time it takes for the sample CatalogSource CR to have all its child resources created on the cluster.

Motivations:

Improve the performance of unpacking the catalog image contents for consumption by the CatalogSource controller

Additional Context:
Both the docker.io/anik120/rukpak-packageserver:latest and docker.io/anik120/catalogsource-controller:latest images have been updated with these changes.

to use a Job resource to unpack image contents in parallel. This increases the speed in which catalogsources are processed and children resources are created. Signed-off-by: Bryce Palmer <everettraven@gmail.com>

everettraven · 2023-02-01T20:36:22Z

pkg/controllers/core/catalogsource_controller.go

 		return ctrl.Result{}, err
 	}

+	// TODO: Can we create these resources in parallel using goroutines?


Wasn't entirely sure if it would be worth exploring this so figured I would pose this as question - Should we explore this route?

Did some testing regarding this, and splitting the Package and BundleMetadata creation into 2 goroutines only resulted in an ~6 second increase in resource creation. I suspect that breaking the BundleMetadata creations into multiple goroutines we may make a more significant gain since it is the larger portion of creation requests.

After attempting to break the BundleMetadata CR creation into multiple goroutines, it appears that did not result in any extra speed increase.

Due to these brief findings, I think using coroutines is adding additional complexity with minimal benefit and is probably best to leave out for now.

anik120

Great work on this Bryce 🎉

I left some comments, and the only comment about the PR structure I have is that when there's a PR that requires a bunch of related changes, it's better for reviewing if there are multiple commits, instead of one commit carrying of all the changes.

For example, instead of having this section in the PR commit message's body:

A brief overview of the changes:
A couple new functions in catalogsource_controller.go and removal of the containerdregistry approach
go mod tidy related updates
Introduction of a new UnpackPhaseError type that is used to recognize when a Job pod is not yet in the Succeeded state.
A couple manifest updates to ensure the controller has the proper permissions to run as expected
A couple helper scripts for testing:
reset.sh deletes and recreates a KinD cluster and then applies the necessary manifests to get everything up and running
timetest.sh can be used by running time ./timetest.sh to evaluate the time it takes for the sample CatalogSource CR to have all its child resources created on the cluster.

If the commit is divided into several commits that are logical units of the changes, maybe something like

commit 1: update dependencies (go mod changes): commit message explains why this is needed

commit 2: code changes to controller (including RBAC changes)

commit 3: test/bootstrap scripts

then it's easier to review commit by commit + you get more of an opportunity to present the logical units of your changes in the commits themselves rather than needing to explain them in a comment in the PR (which for eg doesn't show up if you do git blame).

anik120 · 2023-02-07T20:31:35Z

config/rbac/role.yaml

+- resources:
+  - pods
+  verbs:
+  - create
+  - delete
+  - get
+  - list
+  - patch
+  - update
+  - watch
+- apiGroups:
+  - batch
+  resources:
+  - jobs
+  verbs:
+  - create
+  - delete
+  - get
+  - list
+  - patch
+  - update
+  - watch
+- apiGroups:
+  - batch
+  resources:
+  - pods/logs
+  verbs:
+  - create
+  - delete
+  - get
+  - list
+  - patch
+  - update
+  - watch


Do we need ALL of these? I.e it's better we start with the minimal set of required permission and then work our way up to everything we need

Good point. Just a side note, this file isn't even used in the workflow for spinning this PoC up - I'll just revert these changes for now.

Reverted this in 85b94cb

anik120 · 2023-02-07T20:31:41Z

config/rbac.yaml

+    resources:
+      - 'jobs'
+    verbs:
+      - '*'


Probably should just start off with just create

Set verbs to create, get, list, watch in 85b94cb

anik120 · 2023-02-07T20:43:57Z

pkg/controllers/core/catalogsource_controller.go

 }

+const opmImage = "quay.io/operator-framework/opm:v1.26"


This should ideally be configurable via a flag to the binary that is then parsed, instead of hardcoding it in code.

Made it configurable via a flag in the controller manager binary in commit f7e8156

anik120 · 2023-02-07T20:47:48Z

timetest.sh

+## Loop until all packages are created (170)
+packages=0
+
+until [ $packages -eq 170 ]


A nit but good hygiene to include a timeout here. Even though we're using a tagged image (that I own), if we use this concept of testing in the implementation repo where we'll have more official images to test, we must include a timeout since the underlying content of these images can always change (unless we use image shas to pin these down).

Fair point. This script was really just thrown together for verifying how fast the resources were created. I'm honestly not sure how useful it is on it's own so I'll probably just remove it altogether.

Script removed in 85b94cb

anik120 · 2023-02-07T20:48:28Z

reset.sh

@@ -0,0 +1,6 @@
+#!/bin/bash


Any reason this is not just a makefile target?

Honestly forgot this was included in the PR and was more so meant as an easy way for me to reset the kind cluster. If this is something we think would be useful I can create a Makefile target.

Otherwise I can just remove this script altogether.

Script removed in 85b94cb

That's fine. We'll need to do a makefile targets clean up anyway but that can be punted for when we're in the "not PoC anymore" stage.

Signed-off-by: Bryce Palmer <everettraven@gmail.com>

Signed-off-by: Bryce Palmer <bpalmer@redhat.com>

Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

…ster (#209) * Add Ingress overlay to access `Catalog` contents outside of the cluster Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com> * Address review feedback Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com> * Address review feedback #1 Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com> * Address feedback #2 Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com> --------- Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

update catalog image unpacking

0cf57ec

to use a Job resource to unpack image contents in parallel. This increases the speed in which catalogsources are processed and children resources are created. Signed-off-by: Bryce Palmer <everettraven@gmail.com>

everettraven commented Feb 1, 2023

View reviewed changes

anik120 requested changes Feb 7, 2023

View reviewed changes

everettraven and others added 2 commits February 8, 2023 09:28

address review comments

85b94cb

Signed-off-by: Bryce Palmer <everettraven@gmail.com>

make opm image configurable via a flag

f7e8156

Signed-off-by: Bryce Palmer <bpalmer@redhat.com>

anik120 merged commit 01358be into operator-framework:main Feb 9, 2023

rashmigottipati added a commit to rashmigottipati/catalogd that referenced this pull request Jun 20, 2023

Address review comments operator-framework#2

6e21bdf

Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

rashmigottipati added a commit to rashmigottipati/catalogd that referenced this pull request Jun 20, 2023

Address review comments operator-framework#2

1a8625e

Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

rashmigottipati added a commit to rashmigottipati/catalogd that referenced this pull request Jun 20, 2023

Address review comments operator-framework#2

e4f64c0

Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

rashmigottipati added a commit to rashmigottipati/catalogd that referenced this pull request Jun 20, 2023

Address review comments operator-framework#2

579ad94

Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

rashmigottipati added a commit to rashmigottipati/catalogd that referenced this pull request Jun 20, 2023

Address review comments operator-framework#2

f049cd1

Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

rashmigottipati added a commit to rashmigottipati/catalogd that referenced this pull request Nov 9, 2023

Address feedback operator-framework#2

9d68719

Signed-off-by: Rashmi Gottipati <chowdary.grashmi@gmail.com>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

update catalog image unpacking process #2

update catalog image unpacking process #2

everettraven commented Feb 1, 2023

everettraven Feb 1, 2023

everettraven Feb 1, 2023 •

edited

Loading

everettraven Feb 1, 2023

anik120 left a comment

anik120 Feb 7, 2023

everettraven Feb 7, 2023

everettraven Feb 8, 2023

anik120 Feb 7, 2023

everettraven Feb 8, 2023

anik120 Feb 7, 2023

everettraven Feb 8, 2023

anik120 Feb 7, 2023

everettraven Feb 7, 2023

everettraven Feb 8, 2023

anik120 Feb 7, 2023

everettraven Feb 7, 2023

everettraven Feb 8, 2023

anik120 Feb 9, 2023 •

edited

Loading

update catalog image unpacking process #2

update catalog image unpacking process #2

Conversation

everettraven commented Feb 1, 2023

Choose a reason for hiding this comment

everettraven Feb 1, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

anik120 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

anik120 Feb 9, 2023 • edited Loading

Choose a reason for hiding this comment

everettraven Feb 1, 2023 •

edited

Loading

anik120 Feb 9, 2023 •

edited

Loading