Add connection monitoring (#12)

CHANGELOG.md

@@ -1,6 +1,9 @@
 # postgres-vacuum-monitor
 
-## v.0.8.0
+## v.10.0
+- Add events for connection idle time and state.
+
+## v.0.9.0
 - Add the application name in the event.
 
 ## v.0.8.0

lib/postgres/vacuum/jobs/monitor_job.rb

@@ -6,6 +6,8 @@ class MonitorJob
         AUTOVACUUM_LAGGING_EVENT = 'AutoVacuumLagging'.freeze
         LONG_TRANSACTIONS = 'LongTransactions'.freeze
         BLOCKED_QUERIES = 'BlockedQueries'.freeze
+        CONNECTION_STATE = 'ConnectionState'.freeze
+        CONNECTION_IDLE_TIME = 'ConnectionIdleTime'.freeze
 
         def perform(*)
           with_each_db_name_and_connection do |name, connection|
@@ -47,6 +49,20 @@ def perform(*)
                 current_statement_in_blocking_process: row['current_statement_in_blocking_process']
               )
             end
+
+            connection.execute(Postgres::Vacuum::Monitor::Query.connection_state).each do |row|
+              reporter_class.report_event(CONNECTION_STATE, database_name: name, state: row['state'], connection_count: row['connection_count'])
+            end
+
+            connection.execute(Postgres::Vacuum::Monitor::Query.connection_idle_time).each do |row|
+              reporter_class.report_event(
+                CONNECTION_IDLE_TIME,
+                database_name: name,
+                max: row['max'],
+                median: row['median'],
+                percentile_90: row['percentile_90']
+              )
+            end
           end
 
           true

lib/postgres/vacuum/monitor/query.rb

@@ -97,6 +97,28 @@ def blocked_queries
             WHERE NOT blocked_locks.GRANTED;
           SQL
         end
+
+        def connection_state
+          <<-SQL
+            SELECT 
+              state, count(*) as connection_count 
+            FROM pg_stat_activity 
+            GROUP BY state 
+            ORDER BY count DESC;
+          SQL
+        end
+
+        def connection_idle_time
+          <<-SQL
+            SELECT
+                max(EXTRACT(EPOCH FROM (CURRENT_TIMESTAMP - state_change))) as max,
+                percentile_cont(0.5) within GROUP (ORDER BY EXTRACT(EPOCH FROM (CURRENT_TIMESTAMP - state_change)) DESC) AS median,
+                percentile_cont(0.9) within GROUP (ORDER BY EXTRACT(EPOCH FROM (CURRENT_TIMESTAMP - state_change)) DESC) AS percentile_90
+            FROM pg_stat_activity
+            WHERE state = 'idle'
+            LIMIT 1000;
+          SQL
+        end
       end
     end
   end

lib/postgres/vacuum/monitor/version.rb

@@ -1,7 +1,7 @@
 module Postgres
   module Vacuum
     module Monitor
-      VERSION = '0.9.0'.freeze
+      VERSION = '0.10.0'.freeze
     end
   end
 end

spec/postgres/vacuum/jobs/monitor_job_spec.rb

@@ -98,6 +98,37 @@ def self.report_event(name, attributes = {})
       )
     end
 
+    it "reports connection state" do
+      allow(mock_connection).to receive(:execute).with(Postgres::Vacuum::Monitor::Query.connection_state).and_return(
+        ['connection_count' => 4, 'state' => 'idle']
+      )
+
+      job.perform
+
+      expect(TestMetricsReporter).to have_received(:report_event).with(
+        Postgres::Vacuum::Jobs::MonitorJob::CONNECTION_STATE,
+        database_name: 'postgres_vacuum_monitor_test',
+        connection_count: 4,
+        state: 'idle'
+      )
+    end
+
+    it "reports connection idle time" do
+      allow(mock_connection).to receive(:execute).with(Postgres::Vacuum::Monitor::Query.connection_idle_time).and_return(
+        ['max' => 3.1, 'median' => 222.22, 'percentile_90' => 9323.323]
+      )
+
+      job.perform
+
+      expect(TestMetricsReporter).to have_received(:report_event).with(
+        Postgres::Vacuum::Jobs::MonitorJob::CONNECTION_IDLE_TIME,
+        database_name: 'postgres_vacuum_monitor_test',
+        max: 3.1,
+        median: 222.22,
+        percentile_90: 9323.323
+      )
+    end
+
     context "with multiple connection pools" do
 
       class SecondPool < ActiveRecord::Base
@@ -107,7 +138,7 @@ class SecondPool < ActiveRecord::Base
 
       it "reports once for a single database." do
         expect(job.perform).to eq true
-        expect(mock_connection).to have_received(:execute).exactly(3)
+        expect(mock_connection).to have_received(:execute).exactly(5)
       end
 
       context "to different databases" do
@@ -119,7 +150,7 @@ class SecondPool < ActiveRecord::Base
 
         it "reports twice for two databases" do
           expect(job.perform).to eq true
-          expect(mock_connection).to have_received(:execute).exactly(6)
+          expect(mock_connection).to have_received(:execute).exactly(10)
         end
       end
     end