Add retry to RHEL registration

Occasionally we can see transient network outages when attempting
to register with the Redhat Portal or Satellite server. This causes
deployment or scaleout operations to fail. These outages are minimal
and retrying often resolves the issue. This becomes more prevelant
during testing as we deploy infrastructure far more frequently.

Change-Id: If23785fbe2eea4643918b2e68915bbc13c1b1112

diff --git a/extraconfig/pre_deploy/rhel-registration/scripts/rhel-registration b/extraconfig/pre_deploy/rhel-registration/scripts/rhel-registration
index 71ab076..2650a96 100644 (file)
--- a/extraconfig/pre_deploy/rhel-registration/scripts/rhel-registration
+++ b/extraconfig/pre_deploy/rhel-registration/scripts/rhel-registration
@@ -11,6 +11,7 @@ if [ -e $OK ] ; then
     exit 0
 fi
 
+retryCount=0
 opts=
 attach_opts=
 sat5_opts=
@@ -96,12 +97,28 @@ if [ -n "${REG_TYPE:-}" ]; then
     opts="$opts --type=$REG_TYPE"
 fi
 
+function retry() {
+  if [[ $retryCount < 3 ]]; then
+    $@
+    if ! [[ $? == 0 ]]; then
+      retryCount=$(echo $retryCount + 1 | bc)
+      echo "WARN: Failed to connect when running '$@', retrying..."
+      retry $@
+    else
+      retryCount=0
+    fi
+  else
+    echo "ERROR: Failed to connect after 3 attempts when running '$@'"
+    exit 1
+  fi
+}
+
 function detect_satellite_version {
     ping_api=$REG_SAT_URL/katello/api/ping
-    if curl -L -k -s -D - -o /dev/null $ping_api | grep "200 OK"; then
+    if curl --retry 3 --retry-delay 10 --max-time 30 -L -k -s -D - -o /dev/null $ping_api | grep "200 OK"; then
         echo Satellite 6 detected at $REG_SAT_URL
         satellite_version=6
-    elif curl -L -k -s -D - -o /dev/null $REG_SAT_URL/rhn/Login.do | grep "200 OK"; then
+    elif curl --retry 3 --retry-delay 10 --max-time 30 -L -k -s -D - -o /dev/null $REG_SAT_URL/rhn/Login.do | grep "200 OK"; then
         echo Satellite 5 detected at $REG_SAT_URL
         satellite_version=5
     else
@@ -112,29 +129,29 @@ function detect_satellite_version {
 
 case "${REG_METHOD:-}" in
     portal)
-        subscription-manager register $opts
+        retry subscription-manager register $opts
         if [ -z "${REG_AUTO_ATTACH:-}" -a -z "${REG_ACTIVATION_KEY:-}" ]; then
-            subscription-manager attach $attach_opts
+            retry subscription-manager attach $attach_opts
         fi
-        subscription-manager repos --disable '*'
-        subscription-manager $repos
+        retry subscription-manager repos --disable '*'
+        retry subscription-manager $repos
         ;;
     satellite)
         detect_satellite_version
         if [ "$satellite_version" = "6" ]; then
             repos="$repos --enable ${satellite_repo}"
-            curl -L -k -O "$REG_SAT_URL/pub/katello-ca-consumer-latest.noarch.rpm"
+            curl --retry 3 --retry-delay 10 --max-time 30 -L -k -O "$REG_SAT_URL/pub/katello-ca-consumer-latest.noarch.rpm"
             rpm -Uvh katello-ca-consumer-latest.noarch.rpm || true
-            subscription-manager register $opts
-            subscription-manager $repos
-            yum install -y katello-agent || true # needed for errata reporting to satellite6
+            retry subscription-manager register $opts
+            retry subscription-manager $repos
+            retry yum install -y katello-agent || true # needed for errata reporting to satellite6
             katello-package-upload
-            subscription-manager repos --disable ${satellite_repo}
+            retry subscription-manager repos --disable ${satellite_repo}
         else
             pushd /usr/share/rhn/
-            curl -k -O $REG_SAT_URL/pub/RHN-ORG-TRUSTED-SSL-CERT
+            curl --retry 3 --retry-delay 10 --max-time 30 -k -O $REG_SAT_URL/pub/RHN-ORG-TRUSTED-SSL-CERT
             popd
-            rhnreg_ks --serverUrl=$REG_SAT_URL/XMLRPC $sat5_opts
+            retry rhnreg_ks --serverUrl=$REG_SAT_URL/XMLRPC $sat5_opts
         fi
         ;;
     disable)