feat(nfs): Optimising & restricting actions on auto scale servers

press/agent.py

@@ -53,7 +53,7 @@ def __init__(self, server, server_type="Server"):
 		self.server = server
 		self.port = 443 if self.server not in servers_using_alternative_port_for_communication() else 8443
 
-	def new_bench(self, bench: "Bench"):
+	def _get_registry_credentials(self) -> tuple[str, str, str]:
 		settings = frappe.db.get_value(
 			"Press Settings",
 			None,
@@ -62,15 +62,37 @@ def new_bench(self, bench: "Bench"):
 		)
 		cluster = frappe.db.get_value(self.server_type, self.server, "cluster")
 		registry_url = frappe.db.get_value("Cluster", cluster, "repository") or settings.docker_registry_url
+		return settings.docker_registry_username, settings.docker_registry_password, registry_url
+
+	def setup_bench(self, bench: "Bench", primary_server: str | None = None) -> AgentJob:
+		docker_registry_username, docker_registry_password, registry_url = self._get_registry_credentials()
+
+		data = {
+			"name": bench.name,
+			"bench_config": json.loads(bench.bench_config),
+			"registry": {
+				"url": registry_url,
+				"username": docker_registry_username,
+				"password": docker_registry_password,
+			},
+		}
+
+		if primary_server:
+			data.update({"primary_server": primary_server})
+
+		self.create_agent_job("Setup Bench", "benches/setup", data=data, bench=bench)
+
+	def new_bench(self, bench: "Bench") -> AgentJob:
+		docker_registry_username, docker_registry_password, registry_url = self._get_registry_credentials()
 
 		data = {
 			"name": bench.name,
 			"bench_config": json.loads(bench.bench_config),
 			"common_site_config": json.loads(bench.config),
 			"registry": {
 				"url": registry_url,
-				"username": settings.docker_registry_username,
-				"password": settings.docker_registry_password,
+				"username": docker_registry_username,
+				"password": docker_registry_password,
 			},
 		}
 
@@ -84,7 +106,7 @@ def new_bench(self, bench: "Bench"):
 				for m in bench.mounts
 			]
 
-		return self.create_agent_job("New Bench", "benches", data, bench=bench.name)
+		return self.create_agent_job("New Bench", "benches/new", data, bench=bench.name)
 
 	def archive_bench(self, bench):
 		return self.create_agent_job("Archive Bench", f"benches/{bench.name}/archive", bench=bench.name)

press/api/server.py

@@ -177,7 +177,7 @@ def overview(name):
 @frappe.whitelist()
 @protected(["Server", "Database Server"])
 def archive(name):
-	server = poly_get_doc(["Server", "Database Server"], name)
+	server: Server | DatabaseServer = poly_get_doc(["Server", "Database Server"], name)
 	server.drop_server()
 
 

press/fixtures/agent_job_type.json

@@ -1111,11 +1111,8 @@
   "modified": "2023-11-06 07:28:19.023648",
   "name": "New Bench",
   "request_method": "POST",
-  "request_path": "/benches",
+  "request_path": "/benches/new",
   "steps": [
-   {
-    "step_name": "Initialize Bench"
-   },
    {
     "step_name": "Update Bench Configuration"
    },
@@ -1127,6 +1124,21 @@
    }
   ]
  },
+ {
+   "disabled_auto_retry": 0,
+   "docstatus": 0,
+   "doctype": "Agent Job Type",
+   "max_retry_count": 6,
+   "modified": "2023-11-06 07:28:19.023648",
+   "name": "Setup Bench",
+   "request_method": "POST",
+   "request_path": "/benches/setup",
+   "steps": [
+    {
+     "step_name": "Setup Bench"
+    }
+   ]
+  },
  {
   "disabled_auto_retry": 0,
   "docstatus": 0,

press/playbooks/roles/agent/tasks/main.yml

@@ -21,7 +21,7 @@
 - name: Generate Agent Configuration File
   become: yes
   become_user: frappe
-  command: '/home/frappe/agent/env/bin/agent setup config --name {{ server }} --workers {{ workers }} {% if proxy_ip is defined and proxy_ip is truthy %}--proxy-ip {{ proxy_ip }}{% endif %} {% if agent_sentry_dsn is defined and agent_sentry_dsn is truthy %}--sentry-dsn {{ agent_sentry_dsn }}{% endif %}'
+  command: '/home/frappe/agent/env/bin/agent setup config --name {{ server }} --workers {{ workers }} {% if proxy_ip is defined and proxy_ip is truthy %}--proxy-ip {{ proxy_ip }}{% endif %} {% if agent_sentry_dsn is defined and agent_sentry_dsn is truthy %}--sentry-dsn {{ agent_sentry_dsn }}{% endif %} {% if is_nfs_server is defined and is_nfs_server is truthy %}--is-nfs-server{% endif %}'
   args:
     chdir: /home/frappe/agent
 

press/press/doctype/agent_job/agent_job.py

@@ -932,6 +932,7 @@ def process_job_updates(job_name: str, response_data: dict | None = None):  # no
 			process_archive_bench_job_update,
 			process_new_bench_job_update,
 			process_remove_ssh_user_job_update,
+			process_setup_bench_job_update,
 		)
 		from press.press.doctype.code_server.code_server import (
 			process_archive_code_server_job_update,
@@ -1006,6 +1007,8 @@ def process_job_updates(job_name: str, response_data: dict | None = None):  # no
 			process_site_migration_job_update(job, site_migration)
 		elif job.job_type == "Add Upstream to Proxy":
 			process_new_server_job_update(job)
+		elif job.job_type == "Setup Bench":
+			process_setup_bench_job_update(job)
 		elif job.job_type == "New Bench":
 			process_new_bench_job_update(job)
 		elif job.job_type == "Archive Bench":

press/press/doctype/bench/bench.py

@@ -459,11 +459,28 @@ def update_bench_config(self, force=False):
 			agent.update_bench_config(self)
 
 	def after_insert(self):
-		self.create_agent_request()
+		self.setup_bench()
+
+	def setup_bench(self) -> None:
+		"""Create agent request to copy data and setup bench bind on the correct server"""
+		benches_on_shared_volume = frappe.db.get_value("Server", self.server, "benches_on_shared_volume")
+		if benches_on_shared_volume:
+			nfs_server = frappe.db.get_value(
+				"NFS Volume Attachment", {"status": "Success", "primary_server": self.server}, "nfs_server"
+			)
+			Agent(
+				server=nfs_server,
+				server_type="NFS Server",
+			).setup_bench(
+				self,
+				primary_server=self.server,
+			)
+		else:
+			Agent(self.server).setup_bench(self)
 
-	def create_agent_request(self):
-		agent = Agent(self.server)
-		agent.new_bench(self)
+	def deploy_bench(self) -> None:
+		"""Create agent request run the bench on the server request `setup_bench`"""
+		Agent(self.server).new_bench(self)
 
 	def _mark_applied_patch_as_archived(self):
 		frappe.db.set_value(
@@ -1166,12 +1183,38 @@ def archive_staging_sites():
 	StagingSite.archive_expired()
 
 
-def process_new_bench_job_update(job):
+def process_setup_bench_job_update(job):
+	"""This is a prerequisite to the new bench job however bench will still change
+	its status according to this jobs. As its successor will only be fired if this job is successful.
+	"""
 	bench = Bench("Bench", job.bench)
 
+	if job.status == "Success":
+		# Deploy the bench as soon as the job is successful
+		bench.deploy_bench()
+		return
+
+	# If the status is anything except success we will update bench status
 	updated_status = {
 		"Pending": "Pending",
 		"Running": "Installing",
+		"Failure": "Broken",
+		"Delivery Failure": "Broken",
+	}[job.status]
+
+	if updated_status == bench.status:
+		# Same status do nothing
+		return
+
+	frappe.db.set_value("Bench", job.bench, "status", updated_status)
+
+
+def process_new_bench_job_update(job):
+	bench = Bench("Bench", job.bench)
+
+	updated_status = {
+		# "Pending": "Pending", # Does not make sense to backtrack our status
+		"Running": "Installing",
 		"Success": "Active",
 		"Failure": "Broken",
 		"Delivery Failure": "Broken",

press/press/doctype/nfs_server/nfs_server.json

@@ -15,6 +15,9 @@
   "virtual_machine",
   "is_server_prepared",
   "is_server_setup",
+  "billing_section",
+  "team",
+  "plan",
   "networking_section",
   "ip",
   "column_break_9",
@@ -206,6 +209,23 @@
    "fieldtype": "Link",
    "label": "Cluster",
    "options": "Cluster"
+  },
+  {
+   "fieldname": "team",
+   "fieldtype": "Link",
+   "label": "Team",
+   "options": "Team"
+  },
+  {
+   "fieldname": "billing_section",
+   "fieldtype": "Section Break",
+   "label": "Billing"
+  },
+  {
+   "fieldname": "plan",
+   "fieldtype": "Link",
+   "label": "Plan",
+   "options": "Server Plan"
   }
  ],
  "links": [
@@ -226,7 +246,7 @@
    "link_fieldname": "nfs_server"
   }
  ],
- "modified": "2025-10-15 23:20:56.419708",
+ "modified": "2025-11-10 14:17:45.639068",
  "modified_by": "Administrator",
  "module": "Press",
  "name": "NFS Server",
@@ -246,8 +266,9 @@
   }
  ],
  "row_format": "Dynamic",
+ "rows_threshold_for_grid_search": 20,
  "sort_field": "modified",
  "sort_order": "DESC",
  "states": [],
  "track_changes": 1
-}
+}

press/press/doctype/nfs_server/nfs_server.py

@@ -30,6 +30,7 @@ class NFSServer(BaseServer):
 		is_server_prepared: DF.Check
 		is_server_setup: DF.Check
 		monitoring_password: DF.Password | None
+		plan: DF.Link | None
 		private_ip: DF.Data
 		private_mac_address: DF.Data | None
 		private_vlan_id: DF.Data | None
@@ -38,13 +39,14 @@ class NFSServer(BaseServer):
 		ssh_port: DF.Int
 		ssh_user: DF.Data | None
 		status: DF.Literal["Pending", "Installing", "Active", "Broken", "Archived"]
+		team: DF.Link | None
 		tls_certificate_renewal_failed: DF.Check
 		virtual_machine: DF.Link | None
 	# end: auto-generated types
 
 	def validate(self):
 		self.validate_agent_password()
-		# self.validate_monitoring_password()
+		self.validate_monitoring_password()
 
 	def validate_monitoring_password(self):
 		if not self.monitoring_password:
@@ -66,7 +68,7 @@ def _setup_server(self):
 				port=self._ssh_port(),
 				variables={
 					"server": self.name,
-					"workers": 1,
+					"workers": 8,
 					"domain": self.domain,
 					"agent_password": agent_password,
 					"agent_repository_url": agent_repository_url,
@@ -75,6 +77,7 @@ def _setup_server(self):
 					"certificate_private_key": certificate.private_key,
 					"certificate_full_chain": certificate.full_chain,
 					"certificate_intermediate_chain": certificate.intermediate_chain,
+					"is_nfs_server": True,
 				},
 			)
 			play = ansible.run()

press/press/doctype/release_group/release_group.py

@@ -398,6 +398,14 @@ def update_config(self, config):
 		self.update_config_in_release_group(sanitized_common_site_config, sanitized_bench_config)
 		self.update_benches_config()
 
+	@property
+	def has_scaled_up_servers(self) -> bool:
+		"""Check if this release group has servers that are currently scaled up"""
+		servers = frappe.db.get_all("Release Group Server", {"parent": self.name}, pluck="server")
+		return bool(
+			frappe.db.get_all("Server", {"name": ("IN", servers), "scaled_up": True}, pluck="name"),
+		)
+
 	def update_config_in_release_group(self, common_site_config, bench_config):
 		"""Updates bench_config and common_site_config in the Release Group
 
@@ -688,6 +696,13 @@ def create_deploy_candidate(
 			return None
 
 		self.check_app_server_storage()
+
+		if self.has_scaled_up_servers:
+			frappe.throw(
+				"Unable to create a new deploy as server(s) in this bench group are currently up scaled!",
+				frappe.ValidationError,
+			)
+
 		apps = self.get_apps_to_update(apps_to_update)
 		if apps_to_update is None:
 			self.validate_dc_apps_against_rg(apps)

press/press/doctype/server/server.py

@@ -396,6 +396,19 @@ def _get_app_and_database_servers(self) -> tuple[Server, DatabaseServer]:
 	@dashboard_whitelist()
 	def drop_server(self):
 		app_server, db_server = self._get_app_and_database_servers()
+
+		if not app_server.is_primary:
+			frappe.throw(
+				"This operation is only allowed on a primary server. Secondary servers cannot be dropped directly.",
+				frappe.ValidationError,
+			)
+
+		if app_server.benches_on_shared_volume:
+			frappe.throw(
+				"This server is currently using a shared volume. Please detach it from the NFS server before proceeding.",
+				frappe.ValidationError,
+			)
+
 		app_server.archive()
 		db_server.archive()
 
@@ -2429,6 +2442,12 @@ def drop_secondary_server(self) -> None:
 	@frappe.whitelist()
 	def setup_secondary_server(self, server_plan: str):
 		"""Setup secondary server"""
+		secondary_server_platform = frappe.db.get_value("Server Plan", server_plan, "platform")
+		if self.platform != secondary_server_platform:
+			frappe.throw(
+				"Please select a plan of the same platform as the primary server", frappe.ValidationError
+			)
+
 		if self.doctype == "Database Server" or self.is_secondary:
 			return
 

press/press/doctype/virtual_machine/virtual_machine.py

@@ -382,6 +382,9 @@ def update_subscription_for_nfs_addon_storage(self, volume: "VirtualMachineVolum
 		primary_server = frappe.db.get_value(
 			"NFS Volume Attachment", {"volume_id": volume.volume_id}, "primary_server"
 		)
+		if not primary_server:
+			"""Most likely the root volume"""
+			return
 		primary_server: Server = frappe.get_doc("Server", primary_server)
 		server_plan_disk_size = frappe.db.get_value("Server Plan", primary_server.plan, "disk")