software/monitor: Added node-monitoring

software/monitor/buildout.hash.cfg

@@ -14,7 +14,7 @@
 # not need these here).
 [template]
 filename = instance.cfg
-md5sum = 072b276e0b0dd4b4a96a50348a04c7a7
+md5sum = 1aed15ab8ef224c5a37947b7dcce25ff
 
 [template-monitor]
 _update_hash_filename_ = instance-monitor.cfg.jinja2
@@ -28,6 +28,10 @@ md5sum = 2eb5596544d9c341acf653d4f7ce2680
 _update_hash_filename_ = instance-monitor-edgetest-basic.cfg.jinja2
 md5sum = efc528296ddf7fade335d5f4241c1828
 
+[template-node-monitoring]
+_update_hash_filename_ = instance-node-monitoring.jinja2.cfg
+md5sum = a894ae223687d56bc86aaeb08d25e9e7
+
 [network-bench-cfg]
 filename = network_bench.cfg.in
 md5sum = 4f4d1a0afc225656278154e453587aea

software/monitor/instance-node-monitoring.jinja2.cfg

+[buildout]
+parts =
+  directory
+  check-disk-space.py
+  check-partition-space.py
+  check-cpu-temperature.py
+  check-ram-usage.py
+  check-network-errors.py
+  check-network-transit.py
+  check-cpu-load.py
+  publish-connection-information
+
+extends = {{ monitor_template }}
+
+eggs-directory = {{ eggs_directory }}
+develop-eggs-directory = {{ develop_eggs_directory }}
+offline = true
+
+[slap-configuration]
+recipe = slapos.cookbook:slapconfiguration.serialised
+computer = {{ slap_connection['computer-id'] }}
+partition = {{ slap_connection['partition-id'] }}
+url = {{ slap_connection['server-url'] }}
+key = {{ slap_connection['key-file'] }}
+cert = {{ slap_connection['cert-file'] }}
+
+[directory]
+recipe = slapos.cookbook:mkdirectory
+software = {{ buildout_directory }}
+home = ${buildout:directory}
+etc = ${:home}/etc
+var = ${:home}/var
+etc = ${:home}/etc
+bin = ${:home}/bin
+run = ${:var}/run
+script = ${:etc}/run
+service = ${:etc}/service
+promise = ${:etc}/promise
+log = ${:var}/log
+
+[publish-connection-information]
+recipe = slapos.cookbook:publish.serialised
+<= monitor-publish
+
+[macro.promise]
+<= monitor-promise-base
+name = ${:_buildout_section_name_}
+
+[check-disk-space.py]
+<= macro.promise
+eggs = slapos.toolbox[prediction]
+promise = check_free_disk_space
+config-collectordb = ${monitor-instance-parameter:collector-db}
+config-frequency = {{ slapparameter_dict.get("promise_free_disk_space_frequency", 3) }}
+config-threshold = {{ slapparameter_dict.get("promise_free_disk_space_threshold", 0.08) }}
+config-nb-days-predicted = {{ slapparameter_dict.get("promise_free_disk_space_nb_days_predicted", 10) }}
+config-display-partition = {{ int(slapparameter_dict.get("promise_free_disk_space_display_partition", 1)) }}
+config-display-prediction = {{ int(slapparameter_dict.get("promise_free_disk_space_display_prediction", 1)) }}
+
+[check-partition-space.py]
+<= macro.promise
+eggs = slapos.toolbox[pandas]
+promise = monitor_partition_space
+config-collectordb = ${monitor-instance-parameter:collector-db}
+config-frequency = {{ slapparameter_dict.get("promise_monitor_space_frequency", 5) }}
+config-threshold-ratio = {{ slapparameter_dict.get("promise_partition_space_threshold", 0.08) }}
+
+[check-cpu-temperature.py]
+<= macro.promise
+promise = check_cpu_temperature
+config-testing = false
+config-frequency = {{ slapparameter_dict.get("promise_cpu_temperature_frequency", 5) }}
+config-max-spot-temp = {{ slapparameter_dict.get("promise_cpu_temperature_threshold", 90) }}
+config-max-avg-temp = {{ slapparameter_dict.get("promise_cpu_avg_temperature_threshold", 80) }}
+config-avg-temp-duration = {{ slapparameter_dict.get("promise_cpu_avg_temperature_threshold_duration", 600) }}
+config-last-avg-computation-file = ${directory:var}/promise_cpu_temperature_last_avg_file
+
+[check-ram-usage.py]
+<= macro.promise
+promise = check_ram_usage
+config-frequency = {{ slapparameter_dict.get("promise_ram_available_frequency", 2) }}
+config-min-threshold-ram = {{ slapparameter_dict.get("promise_ram_available_threshold", 500) }}
+config-min-avg-ram = {{ slapparameter_dict.get("promise_ram_avg_available_threshold", 1e3) }}
+config-avg-ram-period = {{ slapparameter_dict.get("promise_ram_avg_available_threshold_duration", 600) }}
+config-last-avg-ram-file = ${directory:var}/promise_ram_space_last_avg_file
+
+[check-network-errors.py]
+<= macro.promise
+promise = check_network_errors_packets
+config-frequency = {{ slapparameter_dict.get("promise_network_errors_frequency", 5) }}
+config-max-error-messages-per-MB = {{ slapparameter_dict.get("promise_network_errors_threshold", 100) }}
+config-max-lost-packets-per-MB = {{ slapparameter_dict.get("promise_network_lost_packets_threshold", 100) }}
+
+[check-network-transit.py]
+<= macro.promise
+promise = check_network_transit
+config-frequency = {{ slapparameter_dict.get("promise_network_transit_frequency", 5) }}
+config-max-data-amount = {{ slapparameter_dict.get("promise_network_transit_max_data_threshold", 1e6) }}
+config-min-data-amount = {{ slapparameter_dict.get("promise_network_transit_min_data_threshold", 0) }}
+config-transit-period = {{ slapparameter_dict.get("promise_network_transit_duration", 600) }}
+config-last-transit-file = ${directory:var}/promise_network_last_transit_file
+
+[check-cpu-load.py]
+<= macro.promise
+promise = check_server_cpu_load
+config-cpu-load-threshold = {{ slapparameter_dict.get("promise_cpu_load_threshold", 1.5) }}

software/monitor/instance.cfg

@@ -7,10 +7,27 @@ develop-eggs-directory = ${buildout:develop-eggs-directory}
 
 [switch_softwaretype]
 recipe = slapos.cookbook:switch-softwaretype
-default = instance-base-monitor:output
+default = instance-node-monitoring:output
 edgetest-basic = instance-edgetest-basic:output
 RootSoftwareInstance = $${:default}
 
+[instance-node-monitoring]
+recipe = slapos.recipe.template:jinja2
+url = ${template-node-monitoring:target}
+output = $${buildout:directory}/template-node-monitoring.cfg
+extensions = jinja2.ext.do
+context =
+    import json_module json
+    key eggs_directory buildout:eggs-directory
+    key develop_eggs_directory buildout:develop-eggs-directory
+    raw buildout_directory ${buildout:directory}
+    section slap_connection slap-connection
+    key slapparameter_dict slap-configuration:configuration
+    raw monitor_template ${monitor2-template:output}
+    raw network_benck_cfg_output ${network-bench-cfg:output}
+    raw monitor_collect_csv_dump ${monitor-collect-csv-dump:target}
+
+
 [instance-base-monitor]
 recipe = slapos.recipe.template:jinja2
 url = ${template-monitor:target}

software/monitor/software.cfg

@@ -7,6 +7,10 @@ extends =
   ../../stack/monitor/buildout.cfg
   ../../stack/slapos.cfg
   ../../component/python3/buildout.cfg
+  ../../component/pandas/buildout.cfg
+  ../../component/scipy/buildout.cfg
+  ../../component/statsmodels/buildout.cfg
+  ../../component/defaults.cfg
 
 parts =
   slapos-cookbook
@@ -32,6 +36,10 @@ url = ${:_profile_base_location_}/${:_update_hash_filename_}
 recipe = slapos.recipe.build:download
 url = ${:_profile_base_location_}/${:_update_hash_filename_}
 
+[template-node-monitoring]
+recipe = slapos.recipe.build:download
+url = ${:_profile_base_location_}/${:_update_hash_filename_}
+
 [json-test-template]
 recipe = slapos.recipe.build:download
 url = ${:_profile_base_location_}/${:_update_hash_filename_}
@@ -69,3 +77,41 @@ miniupnpc = 2.0.2
 peewee = 3.14.4
 python-whois = 0.7.3
 future = 0.18.2
+
+# For node-monitoring
+statsmodels = 0.11.1
+patsy = 0.5.1
+
+# Build GCC with Fortran for OpenBLAS (scipy & numpy)
+[gcc]
+max_version = 0
+
+# Part below shall be removed after new promises are
+# merged into slapos.toolbox master
+
+# >>>>>>>>>>>>
+[slapos.toolbox-repository]
+recipe = slapos.recipe.build:gitclone
+# Change this url once merge to nexedi/slapos.toolbox
+repository = https://lab.nexedi.com/Just1/slapos.toolbox
+branch = json-promise
+git-executable = ${git:location}/bin/git
+
+[slapos-toolbox-dependencies]
+eggs +=
+  ${pandas:egg}
+  ${statsmodels:egg}
+  ${scipy:egg}
+
+[slapos-toolbox-dev]
+<= slapos-toolbox
+recipe = zc.recipe.egg:develop
+setup = ${slapos.toolbox-repository:location}
+
+[slapos-toolbox]
+prerequisite = ${slapos-toolbox-dev:recipe}
+eggs = slapos.toolbox
+
+[versions]
+slapos.toolbox =
+# <<<<<<<<<<<<