Support (automatic) operator upgrades with OLM (#67)

razvan · siegfriedweber · web-flow · commit d94eed218799 · 2024-05-06T15:32:24.000Z
* wip: added spec.replaces property to the CSV

* wip: add multi op catalog script

* fix: the multi op catalog script

* cleanup

* feat(build-manifests): --quay-release allows to diverge from --release if needed

* feat: automatic commons op upgrade

* fix: update multi-op-catalog to include multiple ops and versions

* fix: don't crash when there are no labels (some hdfs objects)

* fix: small sanity check

* feat(build-manifests): add --skips arg(s)

* fix: more fixes for the spark operator

* fix: typo

* fix: typo

* fix: update internal annotation and better usability.

* feat: add --channel argument

* Update olm/build-manifests.py

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;

* Update olm/multi-op-catalog.sh

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;

* Update olm/multi-op-catalog.sh

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;

* Update olm/multi-op-catalog.sh

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;

* Update olm/multi-op-catalog.sh

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;

* Update olm/multi-op-catalog.sh

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;

* Update olm/multi-op-catalog.sh

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;

* Added note on the purpose of the shell script.

* fix python shebang

---------

Co-authored-by: Siegfried Weber &lt;mail@siegfriedweber.net&gt;
diff --git a/olm/build-bundles.sh b/olm/build-bundles.sh
@@ -194,21 +194,22 @@ main() {
 		exit 1
 	fi
 
-	# this is the same folder that is also used by build-manifests.sh
 	if [ "$OPERATOR" == "spark-k8s" ]; then
-		cd "${OPENSHIFT_ROOT}/operators/stackable-spark-operator/${VERSION}"
-	else
-		cd "${OPENSHIFT_ROOT}/operators/stackable-${OPERATOR}-operator/${VERSION}"
+		echo "Renaming operator from spark-k8s to spark"
+		OPERATOR="spark"
 	fi
 
+	# this is the same folder that is also used by build-manifests.sh
+	cd "${OPENSHIFT_ROOT}/operators/stackable-${OPERATOR}-operator/${VERSION}"
+
 	# clean up any residual files from previous actions
 	bundle-clean
 	build-bundle
 
-	catalog-clean
-	catalog
+	#catalog-clean
+	#catalog
 
-	deploy
+	#deploy
 }
 
 main "$@"
diff --git a/olm/build-manifests.py b/olm/build-manifests.py
@@ -1,4 +1,4 @@
-#!/bin/env python
+#!/usr/bin/env python
 # vim: filetype=python syntax=python tabstop=4 expandtab
 
 import argparse
@@ -55,6 +55,26 @@ def parse_args(argv: list[str]) -> argparse.Namespace:
         required=True,
     )
 
+    parser.add_argument(
+        "--quay-release",
+        help="Use this release tag for quay images. Defaults to --release if not provided. Useful when issuing patch releases that use existing images.",
+        type=cli_parse_release,
+    )
+
+    parser.add_argument(
+        "--replaces",
+        help="CSV version that is replaced by this release. Example: 23.11.0",
+        type=cli_parse_release,
+    )
+
+    parser.add_argument(
+        "--skips",
+        nargs="*",
+        help="CSV versions that are skipped by this release. Example: 24.3.0",
+        default=list(),
+        type=cli_parse_release,
+    )
+
     parser.add_argument(
         "-o",
         "--repo-operator",
@@ -92,8 +112,17 @@ def parse_args(argv: list[str]) -> argparse.Namespace:
         action="store_true",
     )
 
+    parser.add_argument(
+        "--channel",
+        help="Channel name to use for the OLM bundle. Default: <major>.<minor> from the release number or 'alpha' for '0.0.0-dev'",
+    )
+
     args = parser.parse_args(argv)
 
+    # Default to the actual release if no quay release is given
+    if not args.quay_release:
+        args.quay_release = args.release
+
     if not args.repo_certified_operators:
         args.repo_certified_operators = (
             args.repo_operator.parent / "openshift-certified-operators"
@@ -136,21 +165,31 @@ def parse_args(argv: list[str]) -> argparse.Namespace:
             f"Certification repository path not found: {args.repo_certified_operators} or it's not a certified operator repository"
         )
 
+    ### Set bundle channel
+    if not args.channel:
+        if args.release == "0.0.0-dev":
+            args.channel = "alpha"
+        else:
+            args.channel = ".".join(args.release.split(".")[:2])
     return args
 
 
 def cli_validate_openshift_range(cli_arg: str) -> str:
     if not re.match(r"^v4\.\d{2}-v4\.\d{2}$", cli_arg):
-        raise argparse.ArgumentTypeError("Invalid OpenShift version range")
+        raise argparse.ArgumentTypeError(
+            "Invalid OpenShift version range. Example: v4.11-v4.13"
+        )
     return cli_arg
 
 
 def cli_parse_release(cli_arg: str) -> str:
-    if re.match(r"^\d{2}\.([1-9]|1[0-2])\.\d+$", cli_arg) or re.match(
+    if re.match(r"^\d{2}\.([1-9]|1[0-2])\.\d+(-\d*)?$", cli_arg) or re.match(
         r"^0\.0\.0-dev$", cli_arg
     ):
         return cli_arg
-    raise argparse.ArgumentTypeError("Invalid release")
+    raise argparse.ArgumentTypeError(
+        "Invalid version provided for release or replacement"
+    )
 
 
 def cli_log_level(cli_arg: str) -> int:
@@ -189,7 +228,7 @@ def generate_csv_related_images(
             if c["name"] == args.op_name
         ]
     else:
-        return quay_image([(args.op_name, args.release)])
+        return quay_image([(args.op_name, args.quay_release)])
 
 
 def generate_manifests(args: argparse.Namespace) -> list[dict]:
@@ -214,11 +253,20 @@ def generate_manifests(args: argparse.Namespace) -> list[dict]:
     )
 
     if not args.use_helm_images:
-        # patch the image of the operator container
-        # with the quay.io image
+        # patch the image of the operator container with the quay.io image
         for c in op_deployment["spec"]["template"]["spec"]["containers"]:
             if c["name"] == args.op_name:
                 c["image"] = related_images[0]["image"]
+        # patch the annotation image of the operator deployment with the quay.io image
+        try:
+            if op_deployment["spec"]["template"]["metadata"]["annotations"][
+                "internal.stackable.tech/image"
+            ]:
+                op_deployment["spec"]["template"]["metadata"]["annotations"][
+                    "internal.stackable.tech/image"
+                ] = related_images[0]["image"]
+        except KeyError:
+            pass
 
     owned_crds = to_owned_crds(crds)
 
@@ -358,10 +406,18 @@ def generate_csv(
 
     result = load_resource("csv.yaml")
 
+    csv_name = (
+        "spark-operator" if args.op_name == "spark-k8s-operator" else args.op_name
+    )
+
     result["spec"]["version"] = args.release
+    result["spec"]["replaces"] = (
+        f"{csv_name}.v{args.replaces}" if args.replaces else None
+    )
+    result["spec"]["skips"] = [f"{csv_name}.v{v}" for v in args.skips]
     result["spec"]["keywords"] = [args.product]
     result["spec"]["displayName"] = CSV_DISPLAY_NAME[args.product]
-    result["metadata"]["name"] = f"{args.op_name}.v{args.release}"
+    result["metadata"]["name"] = f"{csv_name}.v{args.release}"
     result["metadata"]["annotations"]["containerImage"] = related_images[0]["image"]
     result["metadata"]["annotations"]["description"] = CSV_DISPLAY_NAME[args.product]
     result["metadata"]["annotations"]["repository"] = (
@@ -437,13 +493,18 @@ def generate_helm_templates(args: argparse.Namespace) -> list[dict]:
                     }
                 )
             ### Patch the version label
-            if (
-                crv := man["metadata"]["labels"]["app.kubernetes.io/version"]
-            ) != args.release:
-                logging.warning(
-                    f"Version mismatch for '{man['metadata']['name']}'. Replacing '{crv}' with '{args.release}'"
-                )
-                man["metadata"]["labels"]["app.kubernetes.io/version"] = args.release
+            try:
+                if (
+                    crv := man["metadata"]["labels"]["app.kubernetes.io/version"]
+                ) != args.release:
+                    logging.warning(
+                        f"Version mismatch for '{man['metadata']['name']}'. Replacing '{crv}' with '{args.release}'"
+                    )
+                    man["metadata"]["labels"]["app.kubernetes.io/version"] = (
+                        args.release
+                    )
+            except KeyError:
+                pass
 
         logging.debug("finish generate_helm_templates")
 
@@ -494,7 +555,7 @@ def quay_image(images: list[tuple[str, str]]) -> list[dict[str, str]]:
             data = json.load(response)
             if not data["tags"]:
                 raise ManifestException(
-                    f"Could not find manifest digest for request {tag_url}"
+                    f"Could not find manifest digest for release '{release}' on quay.io. Pass '--use-helm-images' to use docker.stackable.tech instead."
                 )
 
             manifest_digest = [
@@ -523,6 +584,13 @@ def write_metadata(args: argparse.Namespace) -> None:
         )
         annos["annotations"]["com.redhat.openshift.versions"] = args.openshift_versions
 
+        annos["annotations"][
+            "operators.operatorframework.io.bundle.channel.default.v1"
+        ] = args.channel
+        annos["annotations"]["operators.operatorframework.io.bundle.channels.v1"] = (
+            args.channel
+        )
+
         anno_file = metadata_dir / "annotations.yaml"
         logging.info(f"Writing {anno_file}")
         anno_file.write_text(yaml.dump(annos))
diff --git a/olm/build-manifests.sh b/olm/build-manifests.sh
@@ -1,5 +1,8 @@
 #!/usr/bin/env bash
 #
+# NOTE: This script is intended to be used only with the secret and listener ops.
+# For all other operators, use the Python equivalent script: build-manifests.py
+#
 # Helper script to (re)generate OLM package manifests (skeletons).
 #
 # Usage:
diff --git a/olm/multi-op-catalog.sh b/olm/multi-op-catalog.sh
@@ -0,0 +1,103 @@
+#
+# Build an operator catalog.
+#
+# An experimental script that builds a custom operator catalog containing one or
+# more operators and one or more versions of each operator.
+#
+# The catalog is built in ./olm/catalog and doesn't actually require any other OLM manifests
+# because it assumes the bundle images have been built and pushed to a registry.
+#
+# It takes as argument the name of an operator to add to the catalog.
+#
+# The operator versions and the upgrade paths are hard coded in the script (for now).
+#
+# Call it multiple times with different operators to build a catalog with multiple operators.
+#
+# To start fresh, delete the olm/catalog directory.
+#
+# To start fresh with a single operator remove the olm/catalog/${OPERATOR} directory.
+#
+# This script also generates a catalog source yaml file that you can apply to your cluster.
+# Once the catalog is loaded by OLM, you can update it and OLM will automatically
+# pull in the new versions.
+#
+# To install operators from this catalog, use the OperatorHub UI and filter by "Source: Stackable Catalog".
+#
+# Assumes the bundle images are built and pushed to a registry as follows:
+#
+#./olm/build-manifests.py --openshift-versions v4.11-v4.15 --release 23.11.0 --repo-operator /home/razvan/repo/stackable/${OPERATOR}-operator --replaces nothing
+#./olm/build-bundles.sh -r 23.11.0 -o ${OPERATOR} -c ~/repo/stackable/openshift-certified-operators -d
+#
+# The secret and listener op bundles are special and need to be built manually.
+#
+# See: https://olm.operatorframework.io/docs/reference/catalog-templates/#example
+# for an example catalog with multiple bundles
+#
+export OPERATOR="$1"
+
+if [ -z "$OPERATOR" ]; then
+	echo "Usage: $0 <operator>"
+	exit 1
+fi
+
+rm -rf olm/catalog/${OPERATOR}
+rm -rf olm/catalog.Dockerfile
+
+mkdir -p olm/catalog/${OPERATOR}
+opm generate dockerfile olm/catalog
+
+opm init "stackable-${OPERATOR}-operator" \
+	--default-channel=24.4 \
+	--output yaml >"olm/catalog/${OPERATOR}/stackable-${OPERATOR}-operator.yaml"
+
+echo "Add operator to package: ${OPERATOR}"
+{
+	echo "---"
+	echo "schema: olm.channel"
+	echo "package: stackable-${OPERATOR}-operator"
+	echo "name: stable"
+	echo "entries:"
+	echo "- name: ${OPERATOR}-operator.v23.11.0"
+	echo "- name: ${OPERATOR}-operator.v24.3.0"
+	echo "  replaces: ${OPERATOR}-operator.v23.11.0"
+	echo "---"
+	echo "schema: olm.channel"
+	echo "package: stackable-${OPERATOR}-operator"
+	echo "name: \"24.4\""
+	echo "entries:"
+	echo "- name: ${OPERATOR}-operator.v24.4.0-1"
+	echo "  replaces: ${OPERATOR}-operator.v23.11.0"
+	echo "  skips:"
+	echo "  -  ${OPERATOR}-operator.v24.3.0"
+} >>"olm/catalog/${OPERATOR}/stackable-${OPERATOR}-operator.yaml"
+
+echo "Render operator: ${OPERATOR}"
+opm render "docker.stackable.tech/sandbox/${OPERATOR}-bundle:23.11.0" --output=yaml >>"olm/catalog/${OPERATOR}/stackable-${OPERATOR}.v23.11.0-bundle.yaml"
+opm render "docker.stackable.tech/sandbox/${OPERATOR}-bundle:24.3.0" --output=yaml >>"olm/catalog/${OPERATOR}/stackable-${OPERATOR}-v24.3.0-bundle.yaml"
+opm render "docker.stackable.tech/sandbox/${OPERATOR}-bundle:24.4.0-1" --output=yaml >>"olm/catalog/${OPERATOR}/stackable-${OPERATOR}-v24.4.0-1-bundle.yaml"
+
+echo "Validating catalog..."
+opm validate olm/catalog
+
+echo "Build catalog..."
+docker build olm -f olm/catalog.Dockerfile -t "docker.stackable.tech/sandbox/stackable-catalog:multi"
+docker push "docker.stackable.tech/sandbox/stackable-catalog:multi"
+
+export VERSION="multi"
+
+echo "Generating catalog source..."
+{
+	echo "---"
+	echo "apiVersion: operators.coreos.com/v1alpha1"
+	echo "kind: CatalogSource"
+	echo "metadata:"
+	echo "  name: stackable-catalog"
+	echo "spec:"
+	echo "  sourceType: grpc"
+	echo "  image: docker.stackable.tech/sandbox/stackable-catalog:${VERSION}"
+	echo "  displayName: Stackable Catalog"
+	echo "  publisher: Stackable GmbH"
+	echo "  updateStrategy:"
+	echo "    registryPoll:"
+	echo "      interval: 3m"
+} >catalog-source.yaml
diff --git a/olm/resources/annotations.yaml b/olm/resources/annotations.yaml
@@ -1,11 +1,11 @@
 ---
 annotations:
-  operators.operatorframework.io.bundle.channel.default.v1: stable
-  operators.operatorframework.io.bundle.channels.v1: stable
+  operators.operatorframework.io.bundle.channel.default.v1: placeholder
+  operators.operatorframework.io.bundle.channels.v1: placeholder
   operators.operatorframework.io.bundle.manifests.v1: manifests/
   operators.operatorframework.io.bundle.mediatype.v1: registry+v1
   operators.operatorframework.io.bundle.metadata.v1: metadata/
   operators.operatorframework.io.bundle.package.v1: placeholder stackable-airflow-operator
 
-  com.redhat.openshift.versions: v4.11-v4.15 placeholder
+  com.redhat.openshift.versions: placeholder
 
diff --git a/olm/resources/csv.yaml b/olm/resources/csv.yaml
@@ -4,6 +4,7 @@ kind: ClusterServiceVersion
 metadata:
     name: placeholder
     annotations:
+        operatorframework.io/suggested-namespace: stackable-operators
         features.operators.openshift.io/disconnected: "false"
         features.operators.openshift.io/fips-compliant: "false"
         features.operators.openshift.io/proxy-aware: "true"
@@ -21,7 +22,7 @@ metadata:
 spec:
     displayName: placeholder
     description: |
-        The Stackable Data Platform was designed with openness and flexibility in mind. It provides a curated selection of the best open source data apps like Apache Kafka, Apache Druid, Trino and Apache Spark. All data apps work together seamlessly, and can be added or removed in no time. Based on Kubernetes, it runs everywhere – on-prem or in the cloud.
+        The Stackable Data Platform was designed with openness and flexibility in mind. It provides a curated selection of the best open source data apps like Apache Kafka, Apache Druid, Trino and Apache Spark. All data apps work together seamlessly, and can be added or removed in no time. Based on Kubernetes, it runs everywhere - on-prem or in the cloud.
 
         For more information please visit our [web page](https://stackable.tech/en/open-source-data-platform)
     icon:
@@ -37,6 +38,7 @@ spec:
       name: Stackable GmbH
       url: https://stackable.tech
     version: placeholder
+    replaces: placeholder
     minKubeVersion: 1.23.0
 
     installModes:

Original file line number	Diff line number	Diff line change
`@@ -1,5 +1,8 @@`
`1`	`1`	`#!/usr/bin/env bash`
`2`	`2`	`#`
	`3`	`+# NOTE: This script is intended to be used only with the secret and listener ops.`
	`4`	`+# For all other operators, use the Python equivalent script: build-manifests.py`
	`5`	`+#`
`3`	`6`	`# Helper script to (re)generate OLM package manifests (skeletons).`
`4`	`7`	`#`
`5`	`8`	`# Usage:`