Merge pull request #2 from arshiagg/netmaster

Netmaster Sidecar Added

contiv-grafana.yml

@@ -75,4 +75,5 @@ spec:
   ports:
     - protocol: TCP
       port: 3000
-      nodePort: 30001
+      nodePort: 32701
+

contiv-prometheus.yml

@@ -82,4 +82,4 @@ spec:
   ports:
     - protocol: TCP
       port: 9090
-      nodePort: 30000
+      nodePort: 32700

exporter.rb

@@ -7,19 +7,43 @@
 OVS_DB_PORT = 6640
 REQUIRED_KEYS = %w[external_ids name statistics]
 
+def netplugin?
+  ENV.fetch("EXPORTER_MODE") == "netplugin"
+end
+
+def netmaster?
+  !netplugin?
+end
+
 set :bind, '0.0.0.0'
+set :port, netplugin? ? 9004 : 9005
 
 get '/metrics' do
   # get etcd IP
   etcd = ENV.fetch("CONTIV_ETCD").split("//").last
 
   #get netmaster IP
   puts "fetching leading address"
-  netmaster = JSON.parse(HTTParty.get("http://#{etcd}/v2/keys/contiv.io/lock/netmaster/leader").body)["node"]["value"]
+  netmaster_addr = JSON.parse(HTTParty.get("http://#{etcd}/v2/keys/contiv.io/lock/netmaster/leader").body)["node"]["value"]
 
   # Get a list of networks
   puts "fetching networks"
-  raw_networks = JSON.parse(HTTParty.get("http://#{netmaster}/api/v1/networks/").body)
+  raw_networks = JSON.parse(HTTParty.get("http://#{netmaster_addr}/api/v1/networks/").body)
+
+  to_display = []
+
+  if netplugin?
+    to_display = netplugin_stats(netmaster_addr, raw_networks)
+  end
+
+  if netmaster?
+    to_display = netmaster_stats(netmaster_addr, raw_networks)
+  end
+
+  to_display.join("\n") + "\n"
+end
+
+def netplugin_stats(netmaster_addr, raw_networks)
 
   networks = []
   epInfo = {}
@@ -34,7 +58,7 @@
   # Get endpoints and endpoint info for each network
   networks.each do |net|
     puts "fetching #{net} network data"
-    raw_epstats = JSON.parse(HTTParty.get("http://#{netmaster}/api/v1/inspect/networks/#{net}/").body)
+    raw_epstats = JSON.parse(HTTParty.get("http://#{netmaster_addr}/api/v1/inspect/networks/#{net}/").body)
 
     tenant = raw_epstats["Config"]["tenantName"]
     network = raw_epstats["Config"]["networkName"]
@@ -44,21 +68,25 @@
       endptID = ep["endpointID"]
       host = ep["homingHost"]
       container = ep["containerName"]
+      epg = ep["serviceName"]
 
       # create hash of endpointID to hash of endpoint info
       epInfo[endptID] = {
         "tenant": tenant,
         "network": network,
-        "endpointID": endptID,
         "host": host,
         "containerName": container,
       }
+
+      # add epg if endpoint is part of one
+      if epg
+        epInfo[endptID]["endpointGroup"] = epg
+      end
     end
   end
 
   puts "epInfo:"
   puts epInfo.inspect
-
 
   # get ovs stats
   cmd = "ovs-vsctl --db=tcp:127.0.0.1:#{OVS_DB_PORT} list interface | egrep '^name|external_ids|statistics'"
@@ -100,6 +128,22 @@
         # get stats into hash
         epstats = interface["statistics"].split(":").last.scan(/(\w+)=(\d+)/).to_h
 
+
+        # OVS outputs data from it's point of view, not the container point of view
+        # so rx and tx must be reversed
+        new_hash = {}
+
+        epstats.keys.each do |key|
+          if key.start_with?("rx_")
+            new_hash[key.gsub("rx_", "tx_")] = epstats[key]
+          elsif key.start_with?("tx_")
+            new_hash[key.gsub("tx_", "rx_")] = epstats[key]
+          else
+            new_hash[key] = epstats[key]
+          end
+        end
+
+        epstats = new_hash
         #create key-value pairs and store into array
         info = "{" + epInfo[key].map{|k,v| "#{k}=\"#{v}\""}.join(", ") + "}"
         epstats.each do |metric, value|
@@ -109,6 +153,46 @@
     end
   end
 
-  # return key-value pairs
-  records.join("\n") + "\n"
+  records
 end
+
+def netmaster_stats(netmaster_addr, raw_networks)
+
+  records = []
+  # Get a list of networks
+  puts "fetching data"
+  raw_tenants = JSON.parse(HTTParty.get("http://#{netmaster_addr}/api/v1/tenants/").body)
+  raw_epg = JSON.parse(HTTParty.get("http://#{netmaster_addr}/api/v1/endpointGroups/").body)
+
+  records << "count_of_tenants #{raw_tenants.length}"
+
+  # count of networks per tenant
+  raw_tenants.each do |tenant_block|
+    tenant = tenant_block["tenantName"]
+    data = tenant_block["link-sets"]
+    if data["Networks"]
+      records << "count_of_networks{tenant=\"#{tenant}\"} #{data["Networks"].length}"
+    end
+  end
+
+  # count of epg per (tenant, network)
+  raw_networks.each do |network_block|
+    data = network_block["link-sets"]
+    if data["EndpointGroups"]
+      records << "count_of_endpointGroups{tenant=\"#{network_block["tenantName"]}\", network=\"#{network_block["networkName"]}\"} #{data["EndpointGroups"].length}"
+    end
+  end
+
+  # count of policies per (tenant, network, epg)
+  raw_epg.each do |epg_block|
+    data = epg_block["link-sets"]
+    if data["Policies"]
+      records << "count_of_policies{tenant=\"#{epg_block["tenantName"]}\", network=\"#{epg_block["networkName"]}\", endpointGroup=\"#{epg_block["groupName"]}\"} #{data["Policies"].length}"
+    end
+    if data["NetProfiles"]
+      records << "count_of_netprofiles{tenant=\"#{epg_block["tenantName"]}\", network=\"#{epg_block["networkName"]}\", endpointGroup=\"#{epg_block["groupName"]}\"} #{data["NetProfiles"].length}"
+    end
+  end
+
+  records
+end

prometheus.yml

@@ -1,8 +1,8 @@
 # Prometheus configuration to scrape all netplugin endpoints every 5 seconds
 scrape_configs:
 - job_name: 'kubernetes-pods'
-  scrape_interval: 15s
-  scrape_timeout: 10s
+  scrape_interval: 5s
+  scrape_timeout: 5s
 
   kubernetes_sd_configs:
   - role: pod