Enable Temporal to startup even if Elastic Search is down (#866)

A customer discovered an issue where Temporal Server will fail to start up if it is using ElasticSearch, and ElasticSearch is down. There are two separate issues:

Within the Server itself, we block Temporal from starting up if we cannot create an Elastic Search Client
Within the startup of the Docker Image, we wait indefinitely until we are able to access ElasticSearch.
This PR provides a proper fix for the first issue and a workaround for the second issue.

For the first issue, when we create the ElasticSearch client, we disable health checks, and then re-enable the healthchecks after the client is created. This ensure we can create the client even if ES is down.

For the second issue, we added another environment variable called $ES_SCHEMA_SETUP_TIMEOUT_IN_SECONDS. By default, this is 0 (meaning if ES is down, docker start.sh script waits forever). If the user finds their startup is blocked because of this, they can set the variable to any integer value. The docker script will continue after $ES_SCHEMA_SETUP_TIMEOUT_IN_SECONDS seconds if it is unable to establish a connection to Elastic Search 

We have a task at Temporal tracking the creation of a better story for the second issue.

common/elasticsearch/client.go

@@ -90,12 +90,23 @@ func NewClient(config *Config) (Client, error) {
 	client, err := elastic.NewClient(
 		elastic.SetURL(config.URL.String()),
 		elastic.SetSniff(false),
+
+		// Disable health check so we don't block client creation if ES happens to be down.
+		elastic.SetHealthcheck(false),
+
 		elastic.SetRetrier(elastic.NewBackoffRetrier(elastic.NewExponentialBackoff(128*time.Millisecond, 513*time.Millisecond))),
 		elastic.SetDecoder(&elastic.NumberDecoder{}), // critical to ensure decode of int64 won't lose precise
 	)
+
 	if err != nil {
 		return nil, err
 	}
+
+	// Re-enable the healthcheck after client has successfully been created.
+	client.Stop()
+	elastic.SetHealthcheck(true)(client)
+	client.Start()
+
 	return NewWrapperClient(client), nil
 }
 

docker/start.sh

@@ -4,6 +4,7 @@ set -x
 
 DB="${DB:-cassandra}"
 ENABLE_ES="${ENABLE_ES:-false}"
+ES_SCHEMA_SETUP_TIMEOUT_IN_SECONDS="${ES_SCHEMA_SETUP_TIMEOUT_IN_SECONDS:-0}"
 ES_PORT="${ES_PORT:-9200}"
 ES_SCHEME="${ES_SCHEME:-http}"
 RF=${RF:-1}
@@ -119,16 +120,28 @@ wait_for_postgres() {
 }
 
 
-wait_for_es() {
+setup_es() {
+    SECONDS=0
+
     server=`echo $ES_SEEDS | awk -F ',' '{print $1}'`
     URL="${ES_SCHEME}://$server:$ES_PORT"
     curl -s $URL 2>&1 > /dev/null
+
     until [ $? -eq 0 ]; do
+        duration=$SECONDS
+
+        if [ $ES_SCHEMA_SETUP_TIMEOUT_IN_SECONDS -gt 0 ] && [ $duration -ge $ES_SCHEMA_SETUP_TIMEOUT_IN_SECONDS ]; then
+            echo 'WARNING: timed out waiting for elasticsearch to start up. skipping index creation'
+            return;
+        fi
+
         echo 'waiting for elasticsearch to start up'
         sleep 1
         curl -s $URL 2>&1 > /dev/null
     done
+
     echo 'elasticsearch started'
+    setup_es_template
 }
 
 wait_for_db() {
@@ -169,8 +182,7 @@ if [ "$1" = "autosetup" ]; then
 fi
 
 if [ "$ENABLE_ES" == "true" ]; then
-    wait_for_es
-    setup_es_template
+    setup_es
 fi
 
 exec bash /start-temporal.sh