monasca-thresh: Allow topology check and removal in storm

Patch adds a script in the monasca-thresh image that can be used
to check if a topology exists in Storm, and optionally kill it.

This is part of a bug in kolla-ansible where topologies were
not submitted to Storm, but run locally.  This patch includes
a topology check script enabled by KOLLA_BOOTSTRAP which will exit
kolla_start if the topology exists, and optionally enables topology
removal (to allow replacement) enabled by TOPOLOGY_REPLACE.

Topology names and various timeouts may be customized.  If the
new env variables are not set, existing behavior is unchanged.

Partial-Bug: #1808805
Change-Id: If8f0730031435dda4235b7f2d2c23e5f5f767f87

docker/monasca/monasca-thresh/Dockerfile.j2

@@ -66,8 +66,14 @@ RUN cd /monasca-common-source/java \
 
 # Overwrite the script inherited from Storm
 COPY extend_start.sh /usr/local/bin/kolla_extend_start
+
+# Add bootstrap script
+COPY topology_bootstrap.sh /usr/local/bin/topology_bootstrap
+
 RUN touch /usr/local/bin/kolla_monasca_extend_start \
-    && chmod 755 /usr/local/bin/kolla_extend_start /usr/local/bin/kolla_monasca_extend_start
+    && chmod 755 /usr/local/bin/kolla_extend_start \
+    /usr/local/bin/kolla_monasca_extend_start \
+    /usr/local/bin/topology_bootstrap
 
 {% block monasca_thresh_footer %}{% endblock %}
 {% block footer %}{% endblock %}

docker/monasca/monasca-thresh/extend_start.sh

@@ -42,3 +42,9 @@ if [[ $(ls -Ab ${MONASCA_WORKER_DIR}) != "" ]]; then
 fi
 
 . /usr/local/bin/kolla_monasca_extend_start
+
+# Bootstrap and exit if KOLLA_BOOTSTRAP variable is set. This catches all cases
+# of the KOLLA_BOOTSTRAP variable being set, including empty.
+if [[ "${!KOLLA_BOOTSTRAP[@]}" ]]; then
+    . /usr/local/bin/topology_bootstrap
+fi

docker/monasca/monasca-thresh/topology_bootstrap.sh

@@ -0,0 +1,90 @@
+#!/bin/sh
+
+# This script should be sourced by kolla_extend_start when bootstrapping
+#
+# Optional env(<default>):
+#   TOPOLOGY_NAME("monasca-thresh") - topology name to check
+#   TOPOLOGY_KILL_TIMEOUT(5) - secs to wait for topology kill
+#   STORM_WAIT_RETRIES(24) - retries to check for storm
+#   STORM_WAIT_TIMEOUT(20) - secs to wait for storm list
+#   STORM_WAIT_DELAY(5) - secs between storm list attempts
+
+#  - If topology exists, then:
+#     a) if TOPOLOGY_REPLACE is set, the existing topology is killed
+#        and script falls through (topology may be added)
+#     b) otherwise script exits with 0 (topology already exists)
+#  - If topology doesn't exist, script falls through (topology may be added)
+#  - If storm cannot be reached, or kill fails, script exits with 1
+
+TOPOLOGY_NAME=${TOPOLOGY_NAME:-monasca-thresh}
+TOPOLOGY_KILL_TIMEOUT=${TOPOLOGY_KILL_TIMEOUT:-5}
+
+# defaults from monasca-thresh
+STORM_WAIT_RETRIES=${STORM_WAIT_RETRIES:-24}
+STORM_WAIT_TIMEOUT=${STORM_WAIT_TIMEOUT:-20}
+STORM_WAIT_DELAY=${STORM_WAIT_DELAY:-5}
+
+STORM="/opt/storm/bin/storm"
+
+echo "Waiting for storm to become available..."
+success="false"
+for i in $(seq "$STORM_WAIT_RETRIES"); do
+    if timeout "$STORM_WAIT_TIMEOUT" "$STORM" list; then
+        echo "Storm is available, continuing..."
+        success="true"
+        break
+    else
+        echo "Connection attempt $i of $STORM_WAIT_RETRIES failed"
+        sleep "$STORM_WAIT_DELAY"
+    fi
+done
+
+if [ "$success" != "true" ]; then
+    echo "Unable to connect to Storm! Exiting..."
+    sleep 1
+    exit 1
+fi
+
+locate_topology() { # <topology>
+    echo "Searching for topology $1 in the storm"
+    topologies=$("$STORM" list | awk '/-----/,0{if (!/-----/)print $1}')
+    found="false"
+    for topology in $topologies; do
+        if [ "$topology" = "$1" ]; then
+            echo "Found storm topology with name: $topology"
+            found="true"
+            break
+        fi
+    done
+}
+
+# search for existing topology
+locate_topology "$TOPOLOGY_NAME"
+
+if [ "$found" = "true" ]; then
+
+    if [[ ! "${!TOPOLOGY_REPLACE[@]}" ]]; then
+        echo "Topology $TOPOLOGY_NAME found, submission not necessary"
+        exit 0
+    fi
+
+    echo "Topology replacement requested, killing old one..."
+    "$STORM" kill "$TOPOLOGY_NAME" -w "$TOPOLOGY_KILL_TIMEOUT"
+
+    echo "Wait $TOPOLOGY_KILL_TIMEOUT secs for topology to reap its artifacts..."
+    sleep "$TOPOLOGY_KILL_TIMEOUT"
+
+    for i in $(seq "$STORM_WAIT_RETRIES"); do
+        locate_topology "$TOPOLOGY_NAME"
+        [ "$found" != "true" ] && break
+        echo "... wait some more..."
+        sleep "$STORM_WAIT_DELAY"
+    done
+    if [ "$found" = "true" ]; then
+        echo "Unable to kill existing topology, giving up..."
+        exit 1
+    fi
+    echo "Topology successfully killed, continuing..."
+else
+    echo "Topology not found, continuing..."
+fi

releasenotes/notes/bug-1808805-e63af01591f03506.yaml

@@ -0,0 +1,8 @@
+---
+fixes:
+  - |
+    Adds an option to the monasca-thresh container which checks
+    if the topology is currently submitted (KOLLA_BOOTSTRAP), with
+    an option to kill it (TOPOLOGY_REPLACE).  Topology names
+    and various timeouts may be customized.
+    `LP#1808805 <https://launchpad.net/bugs/1808805>`__