From fe13c0a83f8511fe09b8c522490fbf7446a9e7de Mon Sep 17 00:00:00 2001
From: Anthony Lin <anthony.jclin@gmail.com>
Date: Tue, 7 Nov 2017 17:49:16 +0000
Subject: [PATCH] Update deploy_site script

It seems that there is a possibility for Airflow to respond
with 'unknown' state for some corner cases. This will cause
the deploy_site script to get into infinite loop. The codes
in Shipyard has been updated to return 'Unknown (*)' as the
state of the action lifecycle in such situations.

This P.S. is meant to update the script and to set a default time
out of 1.5 hr for the workflow. The time out can be changed by
setting a value for $2 (Note that we will need to set query_time,
i.e. $1 as well)

    $ ./deploy_site.sh 120 7200

Change-Id: Ic56ce85f55e9b5fa0c3f537c6158ddc3dc56fab2
---
 tools/deploy_site.sh | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)

diff --git a/tools/deploy_site.sh b/tools/deploy_site.sh
index 0a7c3ade..d5a0b7cf 100755
--- a/tools/deploy_site.sh
+++ b/tools/deploy_site.sh
@@ -71,13 +71,32 @@ echo
 # 3) Complete - The action has completed successfully.
 # 4) Failed - The action has encountered an error, and has failed.
 # 5) Paused - The action has been paused by a user.
+# 6) Unknown (*) - Unknown State for corner cases
 #
 # Initialize 'action_lifecycle' to 'Pending'
 action_lifecycle="Pending"
 
+# Polling for site_deploy action
+# Define 'deploy_time_out' and default to 1.5 hrs (5400 seconds) if not provided
+# Note that user will need to define query time in this case
+deploy_timeout=${2:-5400}
+deploy_counter=1
+
+check_timeout_counter() {
+
+    # Check total elapsed time
+    # The default time out is set to 1.5 hrs
+    # This value can be changed by setting $2
+    if [[ $deploy_counter -eq $deploy_timeout ]]; then
+       echo 'Deploy Site task has timed out.'
+       break
+    fi
+}
+
 while true;
 do
-    if [[ $action_lifecycle == "Complete" ]] || [[ $action_lifecycle == "Failed" ]] || [[ $action_lifecycle == "Paused" ]]; then
+    if [[ $action_lifecycle == "Complete" ]] || [[ $action_lifecycle == "Failed" ]] || \
+       [[ $action_lifecycle == "Paused" ]] || [[ $action_lifecycle == 'Unknown'* ]]; then
         # Print final results
         echo -e '\nFinal State of Deployment\n'
         cat /tmp/get_action_status.json | jq .
@@ -100,11 +119,16 @@ do
         sleep $query_time
 
         # Check Dag state
-        if [[ $action_lifecycle == "Failed" ]] || [[ $action_lifecycle == "Paused" ]]; then
+        if [[ $action_lifecycle == "Failed" ]] || [[ $action_lifecycle == "Paused" ]] || \
+           [[ $action_lifecycle == 'Unknown'* ]]; then
             echo -e "Dag Execution is in" ${RED}$action_lifecycle${NC} "state\n"
         else
             echo -e "Dag Execution is in" ${GREEN}$action_lifecycle${NC} "state\n"
         fi
+
+        # Step counter and check if deployment has timed out
+        ((deploy_counter++))
+        check_timeout_counter
     fi
 done