fix(esm): add service, telemetry and health check (#124)

fix(esm): add service, telemetry and health check chore(esm): tweak resources and service configuration for esm fix(esm): add restart and reschedule blocks --------- Signed-off-by: Bruce Becker <[email protected]>
brucellino · Apr 1, 2024 · af8643c · af8643c
1 parent 584ab3c
commit af8643c
Showing 1 changed file with 72 additions and 4 deletions.
diff --git a/consul-esm.nomad b/consul-esm.nomad
@@ -5,8 +5,12 @@ variable "consul_esm_version" {
 
 job "consul-esm" {
   group "main" {
-
-    count = 1
+    network {
+      port "metrics" {
+        to = "9000"
+      }
+    }
+    count = 5
 
     update {
       max_parallel     = 1
@@ -25,8 +29,24 @@ job "consul-esm" {
       healthy_deadline = "5m"
     }
 
-    task "monitor" {
+    restart {
+      interval = "10m"
+      attempts = 3
+      delay    = "1m"
+      mode     = "delay"
+    }
+
+    reschedule {
+      unlimited      = true
+      delay          = "15s"
+      delay_function = "fibonacci"
+      max_delay      = "1h"
+    }
 
+    task "monitor" {
+      env {
+        log_level = "INFO"
+      }
       // scaling {
       //   enabled = true
       //   min = 0
@@ -35,10 +55,58 @@ job "consul-esm" {
 
       //   }
       // }
-
+      service {
+        port = "metrics"
+        check {
+          name     = "metrics_health"
+          type     = "http"
+          path     = "/metrics"
+          interval = "1m"
+          timeout  = "5s"
+        }
+      }
       driver = "exec"
       config {
         command = "local/consul-esm"
+        args = [
+          "-config-dir=local"
+        ]
+      }
+
+      template {
+        data        = <<EOT
+log_level = "{{ env "log_level" }}"
+enable_syslog = false
+log_json = false
+instance_id = "${uuidv4()}"
+consul_service = "consul-esm"
+consul_service_tag = ""
+consul_kv_path = "consul-esm/"
+external_node_meta {
+    "external-node" = "true"
+}
+node_reconnect_timeout = "72h"
+node_reconnect_timeout = "72h"
+node_probe_interval = "10s"
+disable_coordinate_updates = false
+http_addr = "localhost:8500"
+token = ""
+datacenter = "dc1"
+client_address = "{{ env "NOMAD_ADDR_metrics" }}"
+ping_type = "udp"
+telemetry {
+	disable_hostname = false
+ 	filter_default = false
+ 	prefix_filter = []
+ 	metrics_prefix = "/metrics"
+ 	prometheus_retention_time = "30s"
+}
+passing_threshold = 0
+critical_threshold = 0
+        EOT
+        destination = "local/config.hcl"
+        perms       = "0644"
+
       }
       artifact {
         source      = "https://releases.hashicorp.com/consul-esm/${var.consul_esm_version}/consul-esm_${var.consul_esm_version}_linux_arm64.zip"