X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/2835205c501da417ff80ce4f06dc32d14c8c38bb..b7c12630cb9de017d76fe60ff296534d6d061399:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index 9066c2ec8..4b9058c8e 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -19,9 +19,8 @@
 
 include_recipe "apache"
 include_recipe "apt::grafana"
-include_recipe "apt::timescaledb"
+include_recipe "awscli"
 include_recipe "networking"
-include_recipe "timescaledb"
 
 passwords = data_bag_item("prometheus", "passwords")
 tokens = data_bag_item("prometheus", "tokens")
@@ -46,11 +45,32 @@ prometheus_exporter "statuscake" do
   environment "STATUSCAKE_APIKEY" => tokens["statuscake"]
 end
 
+template "/etc/prometheus/cloudwatch.yml" do
+  source "cloudwatch.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
+prometheus_exporter "cloudwatch" do
+  address "127.0.0.1"
+  port 5000
+  listen_switch "listen-address"
+  options %w[
+    --config.file=/etc/prometheus/cloudwatch.yml
+    --enable-feature=aws-sdk-v2
+    --enable-feature=always-return-info-metrics
+  ]
+  environment "AWS_ACCESS_KEY_ID" => "AKIASQUXHPE7JHG37EA6",
+              "AWS_SECRET_ACCESS_KEY" => tokens["cloudwatch"]
+  subscribes :restart, "template[/etc/prometheus/cloudwatch.yml]"
+end
+
 cache_dir = Chef::Config[:file_cache_path]
 
-prometheus_version = "2.41.0"
-alertmanager_version = "0.24.0"
-karma_version = "0.111"
+prometheus_version = "2.45.0"
+alertmanager_version = "0.25.0"
+karma_version = "0.114"
 
 directory "/opt/prometheus-server" do
   owner "root"
@@ -58,129 +78,63 @@ directory "/opt/prometheus-server" do
   mode "755"
 end
 
-remote_file "#{cache_dir}/prometheus.linux-amd64.tar.gz" do
-  source "https://github.com/prometheus/prometheus/releases/download/v#{prometheus_version}/prometheus-#{prometheus_version}.linux-amd64.tar.gz"
+prometheus_arch = if arm?
+                    "arm64"
+                  else
+                    "amd64"
+                  end
+
+remote_file "#{cache_dir}/prometheus.linux.tar.gz" do
+  source "https://github.com/prometheus/prometheus/releases/download/v#{prometheus_version}/prometheus-#{prometheus_version}.linux-#{prometheus_arch}.tar.gz"
   owner "root"
   group "root"
   mode "644"
   backup false
 end
 
-archive_file "#{cache_dir}/prometheus.linux-amd64.tar.gz" do
+archive_file "#{cache_dir}/prometheus.linux.tar.gz" do
   action :nothing
   destination "/opt/prometheus-server/prometheus"
   overwrite true
   strip_components 1
   owner "root"
   group "root"
-  subscribes :extract, "remote_file[#{cache_dir}/prometheus.linux-amd64.tar.gz]", :immediately
+  subscribes :extract, "remote_file[#{cache_dir}/prometheus.linux.tar.gz]", :immediately
 end
 
-remote_file "#{cache_dir}/alertmanager.linux-amd64.tar.gz" do
-  source "https://github.com/prometheus/alertmanager/releases/download/v#{alertmanager_version}/alertmanager-#{alertmanager_version}.linux-amd64.tar.gz"
+remote_file "#{cache_dir}/alertmanager.linux.tar.gz" do
+  source "https://github.com/prometheus/alertmanager/releases/download/v#{alertmanager_version}/alertmanager-#{alertmanager_version}.linux-#{prometheus_arch}.tar.gz"
   owner "root"
   group "root"
   mode "644"
   backup false
 end
 
-archive_file "#{cache_dir}/alertmanager.linux-amd64.tar.gz" do
+archive_file "#{cache_dir}/alertmanager.linux.tar.gz" do
   action :nothing
   destination "/opt/prometheus-server/alertmanager"
   overwrite true
   strip_components 1
   owner "root"
   group "root"
-  subscribes :extract, "remote_file[#{cache_dir}/alertmanager.linux-amd64.tar.gz]", :immediately
+  subscribes :extract, "remote_file[#{cache_dir}/alertmanager.linux.tar.gz]", :immediately
 end
 
-remote_file "#{cache_dir}/karma-linux-amd64.tar.gz" do
-  source "https://github.com/prymitive/karma/releases/download/v#{karma_version}/karma-linux-amd64.tar.gz"
+remote_file "#{cache_dir}/karma-linux.tar.gz" do
+  source "https://github.com/prymitive/karma/releases/download/v#{karma_version}/karma-linux-#{prometheus_arch}.tar.gz"
   owner "root"
   group "root"
   mode "644"
   backup false
 end
 
-archive_file "#{cache_dir}/karma-linux-amd64.tar.gz" do
+archive_file "#{cache_dir}/karma-linux.tar.gz" do
   action :nothing
   destination "/opt/prometheus-server/karma"
   overwrite true
   owner "root"
   group "root"
-  subscribes :extract, "remote_file[#{cache_dir}/karma-linux-amd64.tar.gz]", :immediately
-end
-
-promscale_version = "0.17.0"
-promscale_extension_version = "0.8.0-1"
-
-database_version = node[:timescaledb][:database_version]
-database_cluster = "#{database_version}/main"
-
-package %W[
-  prometheus
-  promscale-extension-postgresql-#{database_version}
-]
-
-package "promscale-extension-postgresql-#{database_version}" do
-  version promscale_extension_version
-end
-
-apt_preference "promscale-extension-postgresql" do
-  pin "version #{promscale_extension_version}"
-  pin_priority "1100"
-end
-
-postgresql_user "prometheus" do
-  cluster database_cluster
-  superuser true
-end
-
-postgresql_database "promscale" do
-  cluster database_cluster
-  owner "prometheus"
-end
-
-directory "/opt/promscale" do
-  owner "root"
-  group "root"
-  mode "755"
-end
-
-directory "/opt/promscale/bin" do
-  owner "root"
-  group "root"
-  mode "755"
-end
-
-remote_file "/opt/promscale/bin/promscale" do
-  action :create
-  source "https://github.com/timescale/promscale/releases/download/#{promscale_version}/promscale_#{promscale_version}_Linux_x86_64"
-  owner "root"
-  group "root"
-  mode "755"
-end
-
-systemd_service "promscale" do
-  description "Promscale Connector"
-  type "simple"
-  user "prometheus"
-  exec_start "/opt/promscale/bin/promscale --db.uri postgresql:///promscale?host=/run/postgresql&port=5432 --db.connections-max 400"
-  limit_nofile 16384
-  sandbox :enable_network => true
-  restrict_address_families "AF_UNIX"
-end
-
-if node[:prometheus][:promscale]
-  service "promscale" do
-    action [:enable, :start]
-    subscribes :restart, "remote_file[/opt/promscale/bin/promscale]"
-    subscribes :restart, "systemd_service[promscale]"
-  end
-else
-  service "promscale" do
-    action [:disable, :stop]
-  end
+  subscribes :extract, "remote_file[#{cache_dir}/karma-linux.tar.gz]", :immediately
 end
 
 search(:node, "roles:gateway") do |gateway|
@@ -194,6 +148,7 @@ search(:node, "roles:gateway") do |gateway|
 end
 
 jobs = {}
+junos_targets = []
 snmp_targets = []
 
 search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client|
@@ -210,6 +165,7 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
       name = exporter[:name]
       address = exporter[:address]
       sni = exporter[:sni]
+      labels = Array(exporter[:labels])
       scrape_interval = exporter[:scrape_interval]
       scrape_timeout = exporter[:scrape_timeout]
       metric_relabel = exporter[:metric_relabel] || []
@@ -217,6 +173,7 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
       name = key
       address = exporter
       sni = nil
+      labels = []
       scrape_interval = nil
       scrape_timeout = nil
       metric_relabel = []
@@ -227,12 +184,22 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
       :address => address,
       :sni => sni,
       :instance => client.name.split(".").first,
+      :labels => labels,
       :scrape_interval => scrape_interval,
       :scrape_timeout => scrape_timeout,
       :metric_relabel => metric_relabel
     }
   end
 
+  Hash(client[:prometheus][:junos]).each do |instance, details|
+    junos_targets << {
+      :instance => instance,
+      :target => details[:address],
+      :address => client[:prometheus][:addresses]["junos"],
+      :labels => Array(details[:labels])
+    }
+  end
+
   Hash(client[:prometheus][:snmp]).each do |instance, details|
     snmp_targets << {
       :instance => instance,
@@ -270,10 +237,12 @@ prometheus_exporter "ssl" do
   register_target false
 end
 
+package "prometheus"
+
 systemd_service "prometheus-executable" do
   service "prometheus"
   dropin "executable"
-  exec_start "/opt/prometheus-server/prometheus/prometheus --config.file=/etc/prometheus/prometheus.yml --web.external-url=https://prometheus.openstreetmap.org/prometheus --storage.tsdb.path=/var/lib/prometheus/metrics2 --storage.tsdb.retention.time=30d"
+  exec_start "/opt/prometheus-server/prometheus/prometheus --config.file=/etc/prometheus/prometheus.yml --web.enable-admin-api --web.external-url=https://prometheus.openstreetmap.org/prometheus --storage.tsdb.path=/var/lib/prometheus/metrics2 --storage.tsdb.retention.time=540d"
   timeout_stop_sec 300
   notifies :restart, "service[prometheus]"
 end
@@ -283,7 +252,7 @@ template "/etc/prometheus/prometheus.yml" do
   owner "root"
   group "root"
   mode "644"
-  variables :jobs => jobs, :snmp_targets => snmp_targets, :certificates => certificates
+  variables :jobs => jobs, :junos_targets => junos_targets, :snmp_targets => snmp_targets, :certificates => certificates
 end
 
 template "/etc/prometheus/alert_rules.yml" do
@@ -297,7 +266,7 @@ service "prometheus" do
   action [:enable, :start]
   subscribes :reload, "template[/etc/prometheus/prometheus.yml]"
   subscribes :reload, "template[/etc/prometheus/alert_rules.yml]"
-  subscribes :restart, "archive_file[#{cache_dir}/prometheus.linux-amd64.tar.gz]"
+  subscribes :restart, "archive_file[#{cache_dir}/prometheus.linux.tar.gz]"
 end
 
 systemd_service "prometheus-alertmanager" do
@@ -332,7 +301,7 @@ service "prometheus-alertmanager" do
   action [:enable, :start]
   subscribes :reload, "template[/etc/prometheus/alertmanager.yml]"
   subscribes :restart, "systemd_service[prometheus-alertmanager]"
-  subscribes :restart, "archive_file[#{cache_dir}/alertmanager.linux-amd64.tar.gz]"
+  subscribes :restart, "archive_file[#{cache_dir}/alertmanager.linux.tar.gz]"
 end
 
 directory "/etc/amtool" do
@@ -362,7 +331,7 @@ end
 systemd_service "prometheus-karma" do
   description "Alert dashboard for Prometheus Alertmanager"
   user "prometheus"
-  exec_start "/opt/prometheus-server/karma/karma-linux-amd64 --config.file=/etc/prometheus/karma.yml"
+  exec_start "/opt/prometheus-server/karma/karma-linux-#{prometheus_arch} --config.file=/etc/prometheus/karma.yml"
   sandbox :enable_network => true
   restart "on-failure"
 end
@@ -370,7 +339,7 @@ end
 service "prometheus-karma" do
   action [:enable, :start]
   subscribes :restart, "template[/etc/prometheus/karma.yml]"
-  subscribes :restart, "archive_file[#{cache_dir}/karma-linux-amd64.tar.gz]"
+  subscribes :restart, "archive_file[#{cache_dir}/karma-linux.tar.gz]"
   subscribes :restart, "systemd_service[prometheus-karma]"
 end
 
@@ -391,9 +360,10 @@ end
 
 apache_module "alias"
 apache_module "proxy_http"
+apache_module "proxy_wstunnel"
 
 ssl_certificate "prometheus.openstreetmap.org" do
-  domains ["prometheus.openstreetmap.org", "prometheus.osm.org"]
+  domains ["prometheus.openstreetmap.org", "prometheus.osm.org", "munin.openstreetmap.org", "munin.osm.org"]
   notifies :reload, "service[apache2]"
 end
 
@@ -408,3 +378,49 @@ template "/etc/cron.daily/prometheus-backup" do
   group "root"
   mode "750"
 end
+
+package %w[
+  curl
+  jq
+]
+
+directory "/var/lib/prometheus/.aws" do
+  user "prometheus"
+  group "prometheus"
+  mode "755"
+end
+
+template "/var/lib/prometheus/.aws/credentials" do
+  source "aws-credentials.erb"
+  user "prometheus"
+  group "prometheus"
+  mode "600"
+  variables :passwords => passwords
+end
+
+template "/usr/local/bin/prometheus-backup-data" do
+  source "backup-data.erb"
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+systemd_service "prometheus-backup-data" do
+  description "Backup prometheus data to S3"
+  user "prometheus"
+  exec_start "/usr/local/bin/prometheus-backup-data"
+  read_write_paths %w[
+    /var/lib/prometheus/.aws
+    /var/lib/prometheus/metrics2/snapshots
+  ]
+  sandbox :enable_network => true
+end
+
+systemd_timer "prometheus-backup-data" do
+  description "Backup prometheus data to S3"
+  on_calendar "03:11"
+end
+
+service "prometheus-backup-data.timer" do
+  action [:enable, :start]
+end