Fix empty gather logs due to script crash

README.md

@@ -184,13 +184,14 @@ Note: most of the resource outputs are given in 3 file types: `.json`, `.yaml`,
 
 ## Testing
 
-You can run the script locally from your workstation.
-To do that you need an OpenShift cluster and you will have to install the Red Hat GitOps Operator.
-Then you can run the script like this:
+To do that you need an OpenShift cluster, and you will have to install the Red Hat GitOps Operator.
+Then you can test how your changes affects gathered data:
 
 ```shell
-chmod +x ./gather_gitops.sh
-./gather_gitops.sh --base-collection-path .
+# You may need to create the repository on quay.io manually to make sure it is public
+make REGISTRY_USERNAME=my-non-production-org CONTAINER_IMAGE_TAG="$(git rev-parse HEAD)" push
+# Note some differences are expected, like few lines in rapidly populated logs
+./test/compare.sh registry.redhat.io/openshift-gitops-1/must-gather-rhel8:"$SOME_OLD_VERSION" quay.io/my-non-production-org/gitops-must-gather:"$(git rev-parse HEAD)"
 ```
 
 Last but not least, please make sure you run `make lint` before pushing new changes.

gather_gitops.sh

@@ -1,23 +1,32 @@
 #!/usr/bin/env bash
 
 set -eu -o pipefail
+s=declare_out_of_trap_script # Workaround for https://github.com/koalaman/shellcheck/issues/3287
+trap 's=$?; echo >&2 "$0: Error on line "$LINENO": $BASH_COMMAND"; exit $s' ERR
 
 LOGS_DIR="/must-gather"
 
 mkdir -p ${LOGS_DIR}
 
 GITOPS_CURRENT_CSV=$(oc get subscription.operators.coreos.com --ignore-not-found -A -o json | jq '.items[] | select(.metadata.name=="openshift-gitops-operator") | .status.currentCSV' -r)
+if [ -z "$GITOPS_CURRENT_CSV" ]; then
+    NON_ARGO_CRD_NAMES=()
+else
+    readarray -t NON_ARGO_CRD_NAMES < <(oc get csv --ignore-not-found "$GITOPS_CURRENT_CSV" -o json | jq '.spec.customresourcedefinitions.owned[] | select(.name | contains("argoproj.io") | not) | .name' -rj)
+fi
 
 # Gathering cluster version all the crd related to operators.coreos.com and argoproj.io
 echo "gather_gitops:$LINENO] inspecting crd, clusterversion .." | tee -a ${LOGS_DIR}/gather_gitops.log
+readarray -t UPSTREAM_CRDS < <(oc get crd -o name | grep -Ei "argoproj.io|operators.coreos.com")
 # Getting non.existent.crd is a hack to avoid getting all available crds in the cluster in case there are no owned resources that do not contain "argoproj.io"
-oc adm inspect --dest-dir=${LOGS_DIR} "$(oc get crd -o name | grep -Ei "argoproj.io|operators.coreos.com")" "$(oc get crd non.existent.crd --ignore-not-found "$(oc get csv --ignore-not-found "$GITOPS_CURRENT_CSV" -o json | jq '.spec.customresourcedefinitions.owned[] | select(.name | contains("argoproj.io") | not) | " " + .name' -rj)" -o name)" clusterversion/version > /dev/null
+readarray -t NON_ARGO_CRDS < <(oc get crd non.existent.crd --ignore-not-found "${NON_ARGO_CRD_NAMES[@]}" -o name)
+oc adm inspect --dest-dir=${LOGS_DIR} "${UPSTREAM_CRDS[@]}" "${NON_ARGO_CRDS[@]}" clusterversion/version > /dev/null
 
 # Gathering all namespaced custom resources across the cluster that contains "argoproj.io" related custom resources
 oc get crd -o json | jq -r '.items[] | select((.spec.group | contains ("argoproj.io")) and .spec.scope=="Namespaced") | .spec.group + " " + .metadata.name + " " + .spec.names.plural' |
 while read -r API_GROUP APIRESOURCE API_PLURAL_NAME; do
     echo "gather_gitops:$LINENO] collecting ${APIRESOURCE} .." | tee -a ${LOGS_DIR}/gather_gitops.log
-    NAMESPACES=$(oc get "${APIRESOURCE}" --all-namespaces=true --ignore-not-found -o jsonpath='{range .items[*]}{@.metadata.namespace}{"\n"}{end}' | uniq)
+    readarray -t NAMESPACES < <(oc get "${APIRESOURCE}" --all-namespaces=true --ignore-not-found -o jsonpath='{range .items[*]}{@.metadata.namespace}{"\n"}{end}' | uniq)
     for NAMESPACE in "${NAMESPACES[@]}"; do
         mkdir -p "${LOGS_DIR}/namespaces/${NAMESPACE}/${API_GROUP}"
         oc get "${APIRESOURCE}" -n "${NAMESPACE}" -o=yaml >"${LOGS_DIR}/namespaces/${NAMESPACE}/${API_GROUP}/${API_PLURAL_NAME}.yaml"
@@ -26,10 +35,10 @@ done
 
 # Gathering all namespaced custom resources across the cluster that are owned by gitops-operator but do not contain "argoproj.io" related customer resources
 # Getting "non.existent.crd" is a hack to be sure that the output is a list of items even if it only contains zero or a single item
-oc get crd --ignore-not-found non.existent.crd "$(oc get csv --ignore-not-found "$GITOPS_CURRENT_CSV" -o json | jq '.spec.customresourcedefinitions.owned[] | select(.name | contains("argoproj.io") | not) | " " + .name' -rj)"  -o json | jq -r '.items[] | select((.spec.group | contains ("argoproj.io")) and .spec.scope=="Namespaced") | .spec.group + " " + .metadata.name + " " + .spec.names.plural' |
+oc get crd --ignore-not-found non.existent.crd "${NON_ARGO_CRD_NAMES[@]}" -o json | jq -r '.items[] | select((.spec.group | contains ("argoproj.io")) and .spec.scope=="Namespaced") | .spec.group + " " + .metadata.name + " " + .spec.names.plural' |
 while read -r API_GROUP APIRESOURCE API_PLURAL_NAME; do
     echo "gather_gitops:$LINENO] collecting ${APIRESOURCE} .." | tee -a ${LOGS_DIR}/gather_gitops.log
-    NAMESPACES=$(oc get "${APIRESOURCE}" --all-namespaces=true --ignore-not-found -o jsonpath='{range .items[*]}{@.metadata.namespace}{"\n"}{end}' | uniq)
+    readarray -t NAMESPACES < <(oc get "${APIRESOURCE}" --all-namespaces=true --ignore-not-found -o jsonpath='{range .items[*]}{@.metadata.namespace}{"\n"}{end}' | uniq)
     for NAMESPACE in "${NAMESPACES[@]}"; do
         mkdir -p "${LOGS_DIR}/namespaces/${NAMESPACE}/${API_GROUP}"
         oc get "${APIRESOURCE}" -n "${NAMESPACE}" -o=yaml >"${LOGS_DIR}/namespaces/${NAMESPACE}/${API_GROUP}/${API_PLURAL_NAME}.yaml"
@@ -46,7 +55,7 @@ done
 
 # Gathering all cluster-scoped custom resources across the cluster that are owned by gitops-operator but do not contain "argoproj.io"
 # Getting "non.existent.crd" is a hack to be sure that the output is a list of items even if it only contains zero or a single item
-oc get crd --ignore-not-found non.existent.crd "$(oc get csv --ignore-not-found "$GITOPS_CURRENT_CSV" -o json | jq '.spec.customresourcedefinitions.owned[] | select(.name | contains("argoproj.io") | not) | " " + .name' -rj)"  -o json | jq -r '.items[] | select((.spec.group | contains ("argoproj.io")) and .spec.scope=="Namespaced") | .spec.group + " " + .metadata.name + " " + .spec.names.plural' |
+oc get crd --ignore-not-found non.existent.crd "${NON_ARGO_CRD_NAMES[@]}" -o json | jq -r '.items[] | select((.spec.group | contains ("argoproj.io")) and .spec.scope=="Namespaced") | .spec.group + " " + .metadata.name + " " + .spec.names.plural' |
 while read -r API_GROUP APIRESOURCE API_PLURAL_NAME; do
     mkdir -p "${LOGS_DIR}/cluster-scoped-resources/${API_GROUP}"
     echo "gather_gitops:$LINENO] collecting ${APIRESOURCE} .." | tee -a ${LOGS_DIR}/gather_gitops.log
@@ -55,13 +64,16 @@ done
 
 # Inspecting namespace reported in ARGOCD_CLUSTER_CONFIG_NAMESPACES, openshift-gitops and openshift-gitops-operator, and namespaces containing ArgoCD instances
 echo "gather_gitops:$LINENO] inspecting \$ARGOCD_CLUSTER_CONFIG_NAMESPACES, openshift-gitops and openshift-gitops-operator namespaces and namespaces containing ArgoCD instances .." | tee -a ${LOGS_DIR}/gather_gitops.log
-oc get ns --ignore-not-found "$(oc get subs -A --ignore-not-found -o json | jq '.items[] | select(.metadata.name=="openshift-gitops-operator") | .spec.config.env[]?|select(.name=="ARGOCD_CLUSTER_CONFIG_NAMESPACES")| " " + .value | sub(","; " ")' -rj)" "$(oc get ArgoCD,Rollout,RolloutManager -A -o json | jq '.items[] | " " + .metadata.namespace' -rj)" openshift-gitops openshift-gitops-operator -o json \
+readarray -t SUBSCRIPTIONS < <(oc get subs -A --ignore-not-found -o json | jq '.items[] | select(.metadata.name=="openshift-gitops-operator") | .spec.config.env[]?|select(.name=="ARGOCD_CLUSTER_CONFIG_NAMESPACES")| " " + .value | sub(","; " ")' -rj)
+readarray -t ARGO_CRDS < <(oc get ArgoCD,Rollout,RolloutManager -A -o json | jq '.items[] | " " + .metadata.namespace' -rj)
+oc get ns --ignore-not-found "${SUBSCRIPTIONS[@]}" "${ARGO_CRDS[@]}" openshift-gitops openshift-gitops-operator -o json \
 | jq '.items | unique |.[] | .metadata.name' -r |
 while read -r NAMESPACE; do
   echo "gather_gitops:$LINENO] inspecting namespace $NAMESPACE .." | tee -a ${LOGS_DIR}/gather_gitops.log
   oc adm inspect --dest-dir=${LOGS_DIR} "ns/$NAMESPACE" > /dev/null
   echo "gather_gitops:$LINENO] inspecting csv,sub,ip for namespace $NAMESPACE .." | tee -a ${LOGS_DIR}/gather_gitops.log
-  oc adm inspect --dest-dir=${LOGS_DIR} "$(oc get --ignore-not-found clusterserviceversions.operators.coreos.com,installplans.operators.coreos.com,subscriptions.operators.coreos.com -o name -n "$NAMESPACE")" -n "$NAMESPACE" &> /dev/null \
+  readarray -t CSVS_SUBS_IPS < <(oc get --ignore-not-found clusterserviceversions.operators.coreos.com,installplans.operators.coreos.com,subscriptions.operators.coreos.com -o name -n "$NAMESPACE")
+  oc adm inspect --dest-dir=${LOGS_DIR} "${CSVS_SUBS_IPS[@]}" -n "$NAMESPACE" &> /dev/null \
   || echo "gather_gitops:$LINENO] no csv,sub,ip found in namespace $NAMESPACE .." | tee -a ${LOGS_DIR}/gather_gitops.log
 done
 

test/compare.sh

@@ -0,0 +1,63 @@
+#!/usr/bin/env bash
+# https://github.com/olivergondza/bash-strict-mode
+set -eEuo pipefail
+s=declare_out_of_trap_script # Workaround for https://github.com/koalaman/shellcheck/issues/3287
+trap 's=$?; echo >&2 "$0: Error on line "$LINENO": $BASH_COMMAND"; exit $s' ERR
+
+function main() {
+    if [[ $# -ne 2 ]]; then
+        echo >&2 "Usage: $0 [IMAGE_A] [IMAGE_B]"
+        exit 1
+    fi
+    local img_a=$1
+    local img_b=$2
+
+    inv_a="$(mktemp -d gitops-must-gather-A-XXXX)"
+    inv_b="$(mktemp -d gitops-must-gather-B-XXXX)"
+    trap 'rm -rf "${inv_a}" "${inv_b}"' EXIT
+
+    gather "$img_a" "$inv_a"
+    gather "$img_b" "$inv_b"
+
+    diff --color=auto --recursive \
+        --ignore-matching-lines="resourceVersion: " \
+        "${inv_a}" "${inv_b}"
+}
+
+function gather() {
+    image=$1
+    dir=$2
+
+    if ! oc adm must-gather --image="$image" --dest-dir="${dir}" 2>&1 | tee "${dir}/oc-adm-output.log"; then
+        echo >&2 "Failed gathering for $image"
+        return 1
+    fi
+
+    sanitize "$image" "$dir"
+}
+
+function sanitize() {
+    image=$1
+    dir=$2
+
+    # Unify names of the directories its name is based on image name
+    mv "$dir"/*-sha256-* "$dir/__RESOURCES__"
+
+    # In log files, drop image name, generated resource names, timestamp, line numbers, and transfer metrics
+    sed -i -r \
+        -e "s~${image}~__IMAGE_TAG__~g" \
+        -e "s~must-gather-[a-z0-9]{5}~must-gather-XXXXX~g" \
+        -e 's~[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}\.[0-9]+Z~__TIMESTAMP__~g' \
+        -e 's~gather_gitops:[0-9]+~gather_gitops:LL~g' \
+        -e '/total size is .* speedup is .*/d' \
+        -e '/sent .* received .* bytes\/sec/d' \
+        "$dir/oc-adm-output.log" "$dir/must-gather.logs" "$dir/__RESOURCES__/gather.logs" "$dir/__RESOURCES__/gather_gitops.log"
+
+    # Timestamps are not going to match, just test there is the same number of them
+    ts="$(wc -l < "$dir/timestamp")"
+    echo "$ts" > "$dir/timestamp"
+    ts="$(wc -l < "$dir/__RESOURCES__/timestamp")"
+    echo "$ts" > "$dir/__RESOURCES__/timestamp"
+}
+
+main "$@"