feat(api): Add healthcheck field to Config

rootfs/api/migrations/0010_config_healthcheck.py

@@ -0,0 +1,21 @@
+# -*- coding: utf-8 -*-
+# Generated by Django 1.9.7 on 2016-06-17 20:18
+from __future__ import unicode_literals
+
+from django.db import migrations
+import jsonfield.fields
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('api', '0009_auto_20160607_2259'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='config',
+            name='healthcheck',
+            field=jsonfield.fields.JSONField(blank=True, default={}),
+        ),
+    ]

rootfs/api/models/app.py

@@ -371,7 +371,7 @@ def _scale_pods(self, scale_types):
                 'replicas': replicas,
                 'app_type': scale_type,
                 'build_type': release.build.type,
-                'healthcheck': release.config.healthcheck(),
+                'healthcheck': release.config.healthcheck,
                 'routable': routable
             }
 
@@ -427,7 +427,7 @@ def deploy(self, release):
                 'version': "v{}".format(release.version),
                 'app_type': scale_type,
                 'build_type': release.build.type,
-                'healthcheck': release.config.healthcheck(),
+                'healthcheck': release.config.healthcheck,
                 'routable': routable,
                 'batches': batches
             }
@@ -503,11 +503,12 @@ def verify_application_health(self, **kwargs):
         # Get the router host and append healthcheck path
         url = 'http://{}:{}'.format(settings.ROUTER_HOST, settings.ROUTER_PORT)
 
-        # if a health check url is available then 200 is the only acceptable status code
-        if len(kwargs['healthcheck']):
+        # if a httpGet probe is available then 200 is the only acceptable status code
+        if 'livenessProbe' in kwargs.get('healthcheck', {}) and 'httpGet' in kwargs.get('healthcheck').get('livenessProbe'):  # noqa
             allowed = [200]
-            url = urljoin(url, kwargs['healthcheck'].get('path'))
-            req_timeout = kwargs['healthcheck'].get('timeout')
+            handler = kwargs['healthcheck']['livenessProbe']['httpGet']
+            url = urljoin(url, handler.get('path', '/'))
+            req_timeout = handler.get('timeoutSeconds', 1)
         else:
             allowed = set(range(200, 599))
             allowed.remove(404)

rootfs/api/models/config.py

@@ -20,6 +20,7 @@ class Config(UuidAuditedModel):
     cpu = JSONField(default={}, blank=True)
     tags = JSONField(default={}, blank=True)
     registry = JSONField(default={}, blank=True)
+    healthcheck = JSONField(default={}, blank=True)
 
     class Meta:
         get_latest_by = 'created'
@@ -29,13 +30,13 @@ class Meta:
     def __str__(self):
         return "{}-{}".format(self.app.id, str(self.uuid)[:7])
 
-    def healthcheck(self):
+    def _migrate_legacy_healthcheck(self):
         """
         Get all healthchecks options together for use in scheduler
         """
-        # return empty dict if no healthcheck is found
+        # return if no legacy healthcheck is found
         if 'HEALTHCHECK_URL' not in self.values.keys():
-            return {}
+            return
 
         path = self.values.get('HEALTHCHECK_URL', '/')
         timeout = int(self.values.get('HEALTHCHECK_TIMEOUT', 50))
@@ -44,44 +45,17 @@ def healthcheck(self):
         success_threshold = int(self.values.get('HEALTHCHECK_SUCCESS_THRESHOLD', 1))
         failure_threshold = int(self.values.get('HEALTHCHECK_FAILURE_THRESHOLD', 3))
 
-        return {
-            'path': path,
-            'timeout': timeout,
-            'delay': delay,
-            'period_seconds': period_seconds,
-            'success_threshold': success_threshold,
-            'failure_threshold': failure_threshold,
+        self.healthcheck['livenessProbe'] = {
+            'initialDelaySeconds': delay,
+            'timeoutSeconds': timeout,
+            'periodSeconds': period_seconds,
+            'successThreshold': success_threshold,
+            'failureThreshold': failure_threshold,
+            'httpGet': {
+                'path': path,
+            }
         }
 
-    def set_healthchecks(self):
-        """Defines default values for HTTP healthchecks"""
-        if not {k: v for k, v in self.values.items() if k.startswith('HEALTHCHECK_')}:
-            return
-
-        # fetch set health values and any defaults
-        # this approach allows new health items to be added without issues
-        health = self.healthcheck()
-        if not health:
-            return
-
-        # HTTP GET related
-        self.values['HEALTHCHECK_URL'] = health['path']
-
-        # Number of seconds after which the probe times out.
-        # More info: http://releases.k8s.io/HEAD/docs/user-guide/pod-states.md#container-probes
-        self.values['HEALTHCHECK_TIMEOUT'] = health['timeout']
-        # Number of seconds after the container has started before liveness probes are initiated.
-        # More info: http://releases.k8s.io/HEAD/docs/user-guide/pod-states.md#container-probes
-        self.values['HEALTHCHECK_INITIAL_DELAY'] = health['delay']
-        # How often (in seconds) to perform the probe.
-        self.values['HEALTHCHECK_PERIOD_SECONDS'] = health['period_seconds']
-        # Minimum consecutive successes for the probe to be considered successful
-        # after having failed.
-        self.values['HEALTHCHECK_SUCCESS_THRESHOLD'] = health['success_threshold']
-        # Minimum consecutive failures for the probe to be considered failed after
-        # having succeeded.
-        self.values['HEALTHCHECK_FAILURE_THRESHOLD'] = health['failure_threshold']
-
     def set_registry(self):
         # lower case all registry options for consistency
         self.registry = {key.lower(): value for key, value in self.registry.copy().items()}
@@ -127,7 +101,7 @@ def save(self, **kwargs):
                 # usually means a totally new app
                 previous_config = self.app.config_set.latest()
 
-            for attr in ['cpu', 'memory', 'tags', 'registry', 'values']:
+            for attr in ['cpu', 'memory', 'tags', 'registry', 'values', 'healthcheck']:
                 data = getattr(previous_config, attr, {}).copy()
                 new_data = getattr(self, attr, {}).copy()
 
@@ -137,13 +111,12 @@ def save(self, **kwargs):
                         # error if unsetting non-existing key
                         if key not in data:
                             raise UnprocessableEntity('{} does not exist under {}'.format(key, attr))  # noqa
-
                         data.pop(key)
                     else:
                         data[key] = value
                 setattr(self, attr, data)
 
-            self.set_healthchecks()
+            self._migrate_legacy_healthcheck()
             self.set_registry()
             self.set_tags(previous_config)
         except Config.DoesNotExist:

rootfs/api/models/release.py

@@ -420,6 +420,23 @@ def save(self, *args, **kwargs):  # noqa
                         self.summary += ' and '
                     self.summary += "{} {}".format(self.config.owner, changes)
 
+                # if the registry information changed, log the dict diff
+                changes = []
+                old_healthcheck = old_config.healthcheck if old_config else {}
+                diff = dict_diff(self.config.healthcheck, old_healthcheck)
+                # try to be as succinct as possible
+                added = ', '.join(k for k in diff.get('added', {}))
+                added = 'added healthcheck info ' + added if added else ''
+                changed = ', '.join(k for k in diff.get('changed', {}))
+                changed = 'changed healthcheck info ' + changed if changed else ''
+                deleted = ', '.join(k for k in diff.get('deleted', {}))
+                deleted = 'deleted healthcheck info ' + deleted if deleted else ''
+                changes = ', '.join(i for i in (added, changed, deleted) if i)
+                if changes:
+                    if self.summary:
+                        self.summary += ' and '
+                    self.summary += "{} {}".format(self.config.owner, changes)
+
             if not self.summary:
                 if self.version == 1:
                     self.summary = "{} created the initial release".format(self.owner)

rootfs/api/serializers.py

@@ -4,6 +4,7 @@
 
 import json
 import re
+import jsonschema
 from urllib.parse import urlparse
 
 from django.contrib.auth.models import User
@@ -19,10 +20,77 @@
 CPUSHARE_MATCH = re.compile(r'^(?P<cpu>[-+]?[0-9]*\.?[0-9]+[m]{0,1})$')
 TAGVAL_MATCH = re.compile(r'^(?:[a-zA-Z\d][-\.\w]{0,61})?[a-zA-Z\d]$')
 CONFIGKEY_MATCH = re.compile(r'^[a-z_]+[a-z0-9_]*$', re.IGNORECASE)
+PROBE_SCHEMA = {
+    "$schema": "http://json-schema.org/schema#",
+
+    "type": "object",
+    "properties": {
+        # Exec specifies the action to take.
+        # More info: http://kubernetes.io/docs/api-reference/v1/definitions/#_v1_execaction
+        "exec": {
+            "type": "object",
+            "properties": {
+                "command": {
+                    "type": "array",
+                    "minItems": 1,
+                    "items": {"type": "string"}
+                }
+            },
+            "required": ["command"]
+        },
+        # HTTPGet specifies the http request to perform.
+        # More info: http://kubernetes.io/docs/api-reference/v1/definitions/#_v1_httpgetaction
+        "httpGet": {
+            "type": "object",
+            "properties": {
+                "path": {"type": "string"},
+                "port": {"type": "integer"},
+                "host": {"type": "string"},
+                "scheme": {"type": "string"},
+                "httpHeaders": {
+                    "type": "array",
+                    "minItems": 0,
+                    "items": {
+                        "type": "object",
+                        "properties": {
+                            "name": {"type": "string"},
+                            "value": {"type": "string"},
+                        }
+                    }
+                }
+            },
+            "required": ["port"]
+        },
+        # TCPSocket specifies an action involving a TCP port.
+        # More info: http://kubernetes.io/docs/api-reference/v1/definitions/#_v1_tcpsocketaction
+        "tcpSocket": {
+            "type": "object",
+            "properties": {
+                "port": {"type": "integer"},
+            },
+            "required": ["port"]
+        },
+        # Number of seconds after the container has started before liveness probes are initiated.
+        # More info: http://releases.k8s.io/HEAD/docs/user-guide/pod-states.md#container-probes
+        "initialDelaySeconds": {"type": "integer"},
+        # Number of seconds after which the probe times out.
+        # More info: http://releases.k8s.io/HEAD/docs/user-guide/pod-states.md#container-probes
+        "timeoutSeconds": {"type": "integer"},
+        # How often (in seconds) to perform the probe.
+        "periodSeconds": {"type": "integer"},
+        # Minimum consecutive successes for the probe to be considered successful
+        # after having failed.
+        "successThreshold": {"type": "integer"},
+        # Minimum consecutive failures for the probe to be considered
+        # failed after having succeeded.
+        "failureThreshold": {"type": "integer"},
+    }
+}
 
 
 class JSONFieldSerializer(serializers.JSONField):
     def __init__(self, *args, **kwargs):
+        self.convert_to_str = kwargs.pop('convert_to_str', True)
         super(JSONFieldSerializer, self).__init__(*args, **kwargs)
 
     def to_internal_value(self, data):
@@ -40,7 +108,8 @@ def to_representation(self, obj):
                 continue
 
             try:
-                obj[k] = str(v)
+                if self.convert_to_str:
+                    obj[k] = str(v)
             except ValueError:
                 obj[k] = v
                 # Do nothing, the validator will catch this later
@@ -128,6 +197,7 @@ class ConfigSerializer(serializers.ModelSerializer):
     cpu = JSONFieldSerializer(required=False, binary=True)
     tags = JSONFieldSerializer(required=False, binary=True)
     registry = JSONFieldSerializer(required=False, binary=True)
+    healthcheck = JSONFieldSerializer(convert_to_str=False, required=False, binary=True)
 
     class Meta:
         """Metadata options for a :class:`ConfigSerializer`."""
@@ -252,6 +322,22 @@ def validate_registry(self, data):
 
         return data
 
+    def validate_healthcheck(self, data):
+        for key, value in data.items():
+            if value is None:
+                continue
+
+            if key not in ['livenessProbe', 'readinessProbe']:
+                raise serializers.ValidationError(
+                    "Healthcheck keys must be either livenessProbe or readinessProbe")
+            try:
+                jsonschema.validate(value, PROBE_SCHEMA)
+            except jsonschema.ValidationError as e:
+                raise serializers.ValidationError(
+                    "could not validate {}: {}".format(value, e.message))
+
+        return data
+
 
 class ReleaseSerializer(serializers.ModelSerializer):
     """Serialize a :class:`~api.models.Release` model."""