feat(canary): remove canary api

Author: duanhongyi

rootfs/api/migrations/0009_remove_appsettings_canaries_remove_release_canary_and_more.py

@@ -0,0 +1,34 @@
+# Generated by Django 4.2.11 on 2024-07-01 07:58
+
+from django.db import migrations
+
+
+def migration_route_rules(apps, schema_editor):
+    route_model = apps.get_model('api', 'Route')
+    for route in route_model.objects.all():
+        if isinstance(route.rules, dict):
+            route.rules = route.rules.get("stable")
+        route.save()
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('api', '0008_alter_domain_procfile_type_alter_route_procfile_type_and_more'),
+    ]
+
+    operations = [
+        migrations.RemoveField(
+            model_name='appsettings',
+            name='canaries',
+        ),
+        migrations.RemoveField(
+            model_name='release',
+            name='canary',
+        ),
+        migrations.RemoveField(
+            model_name='service',
+            name='canary',
+        ),
+        migrations.RunPython(migration_route_rules),
+    ]

rootfs/api/models/app.py

@@ -42,7 +42,7 @@ def validate_app_id(value):
     """
     Check that the value follows the kubernetes name constraints
     """
-    match = re.match(r'[a-z]([a-z0-9-]*[a-z0-9])?(?<!-canary)$', value)
+    match = re.match(r'[a-z]([a-z0-9-]*[a-z0-9])?$', value)
     if not match:
         raise ValidationError("App name must start with an alphabetic character, cannot end with a"
                               + " hyphen and can only contain a-z (lowercase), 0-9 and hyphens.")
@@ -190,11 +190,8 @@ def restart(self, **kwargs):  # noqa
          Restart deployments with the kubectl rollout api
         """
         deployments = []
-        app_settings = self.appsettings_set.latest()
         if self.structure[kwargs['type']] > 0:
-            if kwargs['type'] in app_settings.canaries:
-                deployments.append(self._get_job_id(kwargs['type'], True))
-            deployments.append(self._get_job_id(kwargs['type'], False))
+            deployments.append(self._get_job_id(kwargs['type']))
         try:
             tasks = [
                 functools.partial(
@@ -219,13 +216,6 @@ def scale(self, user, structure):
             self.log(err_msg, logging.WARNING)
             raise DryccException(err_msg)
         app_settings = self.appsettings_set.latest()
-        if release.canary:
-            self._scale(
-                user,
-                structure,
-                self.release_set.filter(failed=False, canary=False).latest(),
-                app_settings
-            )
         self._scale(user, structure, release, app_settings)
 
     def pipeline(self, release, force_deploy=False, rollback_on_failure=True):
@@ -293,9 +283,8 @@ def deploy(self, release, procfile_types=None, force_deploy=False, rollback_on_f
             if procfile_types is not None and scale_type not in procfile_types:
                 continue
             scale_type_volumes = [_ for _ in volumes if scale_type in _.path.keys()]
-            if not release.canary or scale_type in app_settings.canaries:
-                deploys[scale_type] = self._gather_app_settings(
-                    release, app_settings, scale_type, replicas, volumes=scale_type_volumes)
+            deploys[scale_type] = self._gather_app_settings(
+                release, app_settings, scale_type, replicas, volumes=scale_type_volumes)
         self._deploy(
             deploys, procfile_types, prev_release, release, force_deploy, rollback_on_failure)
         # cleanup old release objects from kubernetes
@@ -307,22 +296,12 @@ def mount(self, user, volume, structure=None):
             raise DryccException('No build associated with this release')
         release = self.release_set.filter(failed=False).latest()
         app_settings = self.appsettings_set.latest()
-        if release.canary:
-            self._mount(
-                user,
-                volume,
-                self.release_set.filter(failed=False, canary=False).latest(),
-                app_settings,
-                structure=structure,
-            )
         self._mount(user, volume, release, app_settings, structure=structure)
 
     def cleanup_old(self, procfile_types=None):
-        names, app_settings = [], self.appsettings_set.latest()
+        names = []
         for scale_type in self.structure.keys():
-            if scale_type in app_settings.canaries:
-                names.append(self._get_job_id(scale_type, True))
-            names.append(self._get_job_id(scale_type, False))
+            names.append(self._get_job_id(scale_type))
         labels = {'heritage': 'drycc'}
         if procfile_types is not None:
             labels["type__in"] = procfile_types
@@ -360,7 +339,7 @@ def pod_name(size=5, chars=string.ascii_lowercase + string.digits):
         data['restart_policy'] = 'Never'
         data['active_deadline_seconds'] = timeout
         data['ttl_seconds_after_finished'] = expires
-        name = self._get_job_id(PROCFILE_TYPE_RUN, release.canary) + '-' + pod_name()
+        name = self._get_job_id(PROCFILE_TYPE_RUN) + '-' + pod_name()
         self.log("{} on {} runs '{}'".format(user.username, name, command))
         kwargs.update(data)
         try:
@@ -509,27 +488,21 @@ def image_pull_secret(self, namespace, registry, image):
         return name
 
     def state_to_k8s(self):
-        def _load_procfile_types(canary):
-            procfile_types = set()
-            for procfile_type, scale in self.structure.items():
-                response = self.scheduler().deployment.get(
-                    self.id, self._get_job_id(procfile_type, canary),
-                    ignore_exception=True)
-                if response.status_code == 404 and scale > 0:
-                    procfile_types.add(procfile_type)
-                elif response.status_code != 200:
-                    data = response.json()
-                    self.log('get deployment status_code {}, message: {}'.format(
-                        response.status_code, data.get("message", "")), logging.ERROR)
-            return procfile_types
-
         release = self.release_set.filter(failed=False).latest()
         if release.build is None:
             self.log('the last release does not have a build, skipping deployment...')
             return
-        procfile_types = _load_procfile_types(False)
-        if release.canary:
-            procfile_types = procfile_types.union(_load_procfile_types(canary=True))
+        procfile_types = set()
+        for procfile_type, scale in self.structure.items():
+            response = self.scheduler().deployment.get(
+                self.id, self._get_job_id(procfile_type),
+                ignore_exception=True)
+            if response.status_code == 404 and scale > 0:
+                procfile_types.add(procfile_type)
+            elif response.status_code != 200:
+                data = response.json()
+                self.log('get deployment status_code {}, message: {}'.format(
+                    response.status_code, data.get("message", "")), logging.ERROR)
         if len(procfile_types) == 0:
             self.log('the cluster status is the latest, skipping deployment...')
             return
@@ -585,11 +558,8 @@ def to_measurements(self, timestamp: float):
     def __str__(self):
         return self.id
 
-    def _get_job_id(self, container_type, canary):
-        job_id = f"{self.id}-{container_type}"
-        if canary:
-            job_id = f"{job_id}-canary"
-        return job_id
+    def _get_job_id(self, container_type):
+        return f"{self.id}-{container_type}"
 
     def _clean_app_logs(self):
         """Delete application logs stored by the logger component"""
@@ -607,23 +577,22 @@ def _mount(self, user, volume, release, app_settings, structure=None):
         volumes = Volume.objects.filter(app=self)
         tasks = []
         for scale_type, replicas in structure.items() if structure else self.structure.items():
-            if scale_type != PROCFILE_TYPE_RUN and (
-                    not release.canary or scale_type in app_settings.canaries):
+            if scale_type != PROCFILE_TYPE_RUN:
                 replicas = self.structure.get(scale_type, 0)
                 scale_type_volumes = [
                     volume for volume in volumes if scale_type in volume.path.keys()]
                 data = self._gather_app_settings(
                     release, app_settings, scale_type, replicas, volumes=scale_type_volumes)
                 deployment = self.scheduler().deployment.get(
-                    self.id, self._get_job_id(scale_type, release.canary)).json()
+                    self.id, self._get_job_id(scale_type)).json()
                 spec_annotations = deployment['spec']['template']['metadata'].get(
                     'annotations', {})
                 self.set_application_config(release, scale_type)
                 # gather volume proc types to be deployed
                 tasks.append(functools.partial(
                     self.scheduler().deployment.patch,
                     namespace=self.id,
-                    name=self._get_job_id(scale_type, release.canary),
+                    name=self._get_job_id(scale_type),
                     image=release.get_deploy_image(scale_type),
                     command=release.get_deploy_command(scale_type),
                     args=release.get_deploy_args(scale_type),
@@ -653,7 +622,7 @@ def _deploy(self, deploys, procfile_types, prev_release,
                 tasks.append(functools.partial(
                     self.scheduler().deploy,
                     namespace=self.id,
-                    name=self._get_job_id(scale_type, release.canary),
+                    name=self._get_job_id(scale_type),
                     image=release.get_deploy_image(scale_type),
                     command=release.get_deploy_command(scale_type),
                     args=release.get_deploy_args(scale_type),
@@ -664,8 +633,7 @@ def _deploy(self, deploys, procfile_types, prev_release,
             except KubeException as e:
                 # Don't rollback if the previous release doesn't have a build which means
                 # this is the first build and all the previous releases are just config changes.
-                if (prev_release.canary == release.canary and
-                        rollback_on_failure and prev_release.build is not None):
+                if (rollback_on_failure and prev_release.build is not None):
                     err = 'There was a problem deploying {}. Rolling back to release {}.'.format(
                         release.version_name, prev_release.version_name)
                     # This goes in the log before the rollback starts
@@ -701,7 +669,6 @@ def _scale(self, user, structure, release, app_settings):  # noqa
         """Scale containers up or down to match requested structure."""
         # use create to make sure minimum resources are created
         self.create()
-
         # Validate structure
         try:
             for target, count in structure.copy().items():
@@ -752,7 +719,7 @@ def _scale_pods(self, scale_types, release, app_settings):
                 functools.partial(
                     self.scheduler().scale,
                     namespace=self.id,
-                    name=self._get_job_id(scale_type, release.canary),
+                    name=self._get_job_id(scale_type),
                     image=release.get_deploy_image(scale_type),
                     command=release.get_deploy_command(scale_type),
                     args=release.get_deploy_args(scale_type),
@@ -905,7 +872,7 @@ def _check_deployment_in_progress(self, deploys, release, force_deploy=False):
         if force_deploy:
             return
         for scale_type, kwargs in deploys.items():
-            name = self._get_job_id(scale_type, release.canary)
+            name = self._get_job_id(scale_type)
             # Is there an existing deployment in progress?
             in_progress, deploy_okay = self.scheduler().deployment.in_progress(
                 self.id, name, kwargs.get("deploy_timeout"), kwargs.get("deploy_batches"),

rootfs/api/models/appsettings.py

@@ -19,7 +19,6 @@ class AppSettings(UuidAuditedModel):
 
     owner = models.ForeignKey(User, on_delete=models.PROTECT)
     app = models.ForeignKey('App', on_delete=models.CASCADE)
-    canaries = models.JSONField(default=list)
     routable = models.BooleanField(default=True)
     autoscale = models.JSONField(default=dict, blank=True)
     label = models.JSONField(default=dict, blank=True)
@@ -63,20 +62,6 @@ def previous(self):
             prev_app_settings = None
         return prev_app_settings
 
-    def _update_canaries(self, previous_settings):
-        old = getattr(previous_settings, 'canaries', [])
-        new = getattr(self, 'canaries', [])
-        data = old.copy()
-        if data and not new:
-            setattr(self, 'canaries', data)
-        elif data != new:
-            for procfile_type in new:
-                if procfile_type not in data:
-                    data.append(procfile_type)
-            setattr(self, 'canaries', data)
-            self.summary += [
-                "{} add canaries for process types {}".format(self.owner, ','.join(new))]
-
     def _update_routable(self, previous_settings):
         old = getattr(previous_settings, 'routable', None)
         new = getattr(self, 'routable', None)
@@ -164,7 +149,7 @@ def _update_fields(self, ignore_update_fields=None):
             previous_settings = self.app.appsettings_set.latest()
         except AppSettings.DoesNotExist:
             pass
-        update_fields = ["canaries", "routable", "autoscale", "label"]
+        update_fields = ["routable", "autoscale", "label"]
         try:
             for update_field in update_fields:
                 if ignore_update_fields is None or update_field not in ignore_update_fields:
@@ -181,23 +166,6 @@ def _update_fields(self, ignore_update_fields=None):
         summary = ' '.join(self.summary)
         self.log('summary of app setting changes: {}'.format(summary), logging.DEBUG)
 
-    def diff_canaries(self):
-        prev_app_settings = self.previous()
-        action, canaries = None, []
-        if prev_app_settings is not None:
-            if prev_app_settings.canaries != self.canaries:
-                for procfile_type in self.canaries:  # add canary
-                    if procfile_type not in prev_app_settings.canaries:
-                        if action is None:
-                            action = "append"
-                        canaries.append(procfile_type)
-                for procfile_type in prev_app_settings.canaries:  # delete canary
-                    if procfile_type not in self.canaries:
-                        if action is None:
-                            action = "remove"
-                        canaries.append(procfile_type)
-        return prev_app_settings, action, canaries
-
     @transaction.atomic
     def save(self, ignore_update_field=None, *args, **kwargs):
         self._update_fields(ignore_update_field)

rootfs/api/models/build.py

@@ -71,15 +71,13 @@ def get_image(self, procfile_type, default_image=None):
         return default_image if default_image else self.image
 
     def create_release(self, user, *args, **kwargs):
-        app_settings = self.app.appsettings_set.latest()
         latest_release = self.app.release_set.filter(failed=False).latest()
         latest_version = self.app.release_set.latest().version
         try:
             new_release = latest_release.new(
                 user,
                 build=self,
                 config=latest_release.config,
-                canary=len(app_settings.canaries) > 0,
             )
             run_pipeline.delay(new_release)
             return new_release

rootfs/api/models/gateway.py

@@ -1,4 +1,3 @@
-import re
 import logging
 from django.db import models
 from django.conf import settings
@@ -201,35 +200,16 @@ def hostnames(self):
     @property
     def default_rules(self):
         service = get_object_or_404(self.app.service_set, procfile_type=self.procfile_type)
-        stable_backend_refs, canary_backend_refs = [], []
+        backend_refs = []
         for item in service.ports:
             if item["port"] == self.port:
-                stable_backend = {
+                backend_refs.append({
                     "kind": "Service",
                     "name": str(service),
                     "port": item["port"],
                     "weight": 100,
-                }
-                canary_backend = {
-                    "kind": "Service",
-                    "name": "%s-canary" % str(service),
-                    "port": item["port"],
-                    "weight": 0,
-                }
-                stable_backend_refs.append(stable_backend)
-                canary_backend_refs.append(stable_backend)
-                canary_backend_refs.append(canary_backend)
-        return {
-            "stable": [{"backendRefs": stable_backend_refs}],
-            "canary": [{"backendRefs": canary_backend_refs}],
-        }
-
-    @property
-    def current_rules(self):
-        app_settings = self.app.appsettings_set.latest()
-        if self.procfile_type in app_settings.canaries:
-            return self.rules["canary"]
-        return self.rules["stable"]
+                })
+        return [{"backendRefs": backend_refs}]
 
     def log(self, message, level=logging.INFO):
         """Logs a message in the context of this service.
@@ -246,13 +226,11 @@ def check_rules(self):
         service = self.app.service_set.filter(
             procfile_type=self.procfile_type).first()
         ports = [item["port"] for item in service.ports]
-        for rules in self.rules.values():
-            for rule in rules:
-                for backend_ref in rule["backendRefs"]:
-                    port = backend_ref["port"]
-                    name = re.sub(r'(.*)-canary', r'\1', backend_ref["name"])
-                    if port not in ports or name != str(service):
-                        return False, {"detail": "backendRefs associated with incorrect service"}
+        for rule in self.rules:
+            for backend_ref in rule["backendRefs"]:
+                port = backend_ref["port"]
+                if port not in ports or backend_ref["name"] != str(service):
+                    return False, {"detail": "backendRefs associated with incorrect service"}
         return True, ""
 
     def refresh_to_k8s(self):
@@ -266,7 +244,7 @@ def refresh_to_k8s(self):
                 self.scheduler().httproute.delete(self.app.id, self._https_redirect_name)
             else:
                 parent_refs.extend(http_parent_refs)
-            self._refresh_to_k8s(self.current_rules, parent_refs)
+            self._refresh_to_k8s(self.rules, parent_refs)
         else:
             self.scheduler().httproute.delete(self.app.id, self.name)
             self.scheduler().httproute.delete(self.app.id, self._https_redirect_name)