fix: resolve session salts at runtime so prod release boots

The @session_options module attribute used Application.compile_env, which
baked compile-time placeholders into the endpoint while runtime.exs set
the real values from SESSION_SIGNING_SALT / SESSION_ENCRYPTION_SALT env
vars. Phoenix detected the mismatch and refused to start (failing migrate
Job in k8s).

- Remove hardcoded salts from config/config.exs (no compile-time binding)
- Add stable per-env salts in dev.exs / test.exs so local + CI don't need
  the env vars
- Split static cookie opts (@static_session_options) from runtime-resolved
  opts in endpoint.ex; expose session_options/0 as an MFA tuple in socket
  connect_info so LiveView decodes sessions with the same runtime salts
- New ToweropsWeb.Plugs.RuntimeSession wraps Plug.Session, fetches salts
  from app env on first request, and caches the initialized opts in
  :persistent_term (zero per-request overhead after warm-up)

CHANGELOG.txt

@@ -1,3 +1,22 @@
+2026-05-12
+fix(session): resolve session salts at runtime so prod release boots
+  The `@session_options` module attribute used `Application.compile_env`,
+  which baked compile-time placeholder values into the endpoint while
+  `config/runtime.exs` set the real values from env vars. Phoenix detected
+  the mismatch and refused to start (the failing migrate Job in k8s).
+  - config/config.exs: removed hardcoded session salts (no compile-time
+    binding for these keys now)
+  - config/dev.exs, config/test.exs: stable per-env salts so local + CI
+    don't need SESSION_*_SALT env vars
+  - lib/towerops_web/endpoint.ex: split static cookie opts (@static_session_options)
+    from runtime-resolved opts; added session_options/0 (used as MFA tuple
+    in socket connect_info) and runtime_session_opts/0
+  - lib/towerops_web/plugs/runtime_session.ex: new wrapper that fetches
+    salts from app env on first request and caches the initialized
+    Plug.Session opts in :persistent_term
+  - SESSION_SIGNING_SALT / SESSION_ENCRYPTION_SALT in k8s/secrets.yaml are
+    now actually consumed at runtime in the release
+
 2026-05-10
 feat(insights): AI now generates its own observations from full network state
   Previously the LLM was only used to enrich existing rule-based insights

config/config.exs

@@ -162,9 +162,6 @@ config :towerops, :scopes,
     test_setup_helper: :register_and_log_in_user
   ]
 
-config :towerops, :session_encryption_salt, "vK3p8mNx"
-config :towerops, :session_signing_salt, "hrDZxLhd"
-
 # Agent Docker Image
 # Override this in runtime.exs or environment-specific config
 config :towerops,

config/dev.exs

@@ -152,12 +152,6 @@ config :towerops, Towerops.Vault,
     }
   ]
 
-# For development, we disable any cache and enable
-# debugging and code reloading.
-#
-# The watchers configuration can be used to run external
-# watchers to your application. For example, we can use it
-# to bundle .js and .css sources.
 config :towerops, ToweropsWeb.Endpoint,
   # Binding to loopback ipv4 address prevents access from other machines.
   # Change to `ip: {0, 0, 0, 0}` to allow access from other machines.
@@ -227,6 +221,15 @@ config :towerops, :mib_dirs, [
 
 # Disable rate limiting in development
 config :towerops, :rate_limiting_enabled, false
+config :towerops, :session_encryption_salt, "dev-encryption-salt-stable"
+
+# For development, we disable any cache and enable
+# debugging and code reloading.
+#
+# The watchers configuration can be used to run external
+# watchers to your application. For example, we can use it
+# to bundle .js and .css sources.
+config :towerops, :session_signing_salt, "dev-signing-salt-stable"
 
 # Redis is not configured in development - application will use stub Agent
 # If you need Redis for local development, set REDIS_HOST env var

config/test.exs

@@ -110,6 +110,8 @@ config :towerops, :agent_webhook_secret, "test-webhook-secret"
 
 # Set environment identifier for runtime checks
 config :towerops, :env, :test
+config :towerops, :session_encryption_salt, "test-encryption-salt-stable"
+config :towerops, :session_signing_salt, "test-signing-salt-stable"
 
 config :towerops,
   ping_module: Towerops.Monitoring.PingMock,

lib/towerops_web/endpoint.ex

@@ -4,24 +4,43 @@ defmodule ToweropsWeb.Endpoint do
 
   # The session will be stored in the cookie and signed,
   # this means its contents can be read but not tampered with.
-  # Set :encryption_salt if you would also like to encrypt it.
   # `http_only: true` blocks JavaScript access via `document.cookie` (XSS
   # defense). `secure` is enabled in :prod by `config/runtime.exs` so the
   # cookie is never sent over plain HTTP; left off here so local HTTP dev
   # still works.
-  @session_options [
+  #
+  # Session salts (`signing_salt`, `encryption_salt`) come from env vars at
+  # runtime via `config/runtime.exs` (see `runtime_session_opts/0` below).
+  # They are merged in by `ToweropsWeb.Plugs.RuntimeSession` and by
+  # `session_options/0` (used by LiveView/socket `connect_info`).
+  @static_session_options [
     store: :cookie,
     key: "_towerops_key",
-    signing_salt: Application.compile_env(:towerops, :session_signing_salt, "default-signing-salt-missing-env"),
-    encryption_salt: Application.compile_env(:towerops, :session_encryption_salt, "default-encryption-salt-missing-env"),
     same_site: "Lax",
-    http_only: true,
+    http_only: true
-    secure: Application.compile_env(:towerops, :secure_cookies, false)
   ]
 
+  @doc """
+  Session options resolved at runtime. Used as an MFA tuple in socket
+  `connect_info` so LiveView decodes the session using the same salts
+  that `ToweropsWeb.Plugs.RuntimeSession` uses to encode it.
+  """
+  def session_options do
+    @static_session_options ++ runtime_session_opts()
+  end
+
+  @doc false
+  def runtime_session_opts do
+    [
+      signing_salt: Application.fetch_env!(:towerops, :session_signing_salt),
+      encryption_salt: Application.fetch_env!(:towerops, :session_encryption_salt),
+      secure: Application.get_env(:towerops, :secure_cookies, false)
+    ]
+  end
+
   socket "/live", Phoenix.LiveView.Socket,
-    websocket: [connect_info: [session: @session_options]],
+    websocket: [connect_info: [session: {__MODULE__, :session_options, []}]],
-    longpoll: [connect_info: [session: @session_options]]
+    longpoll: [connect_info: [session: {__MODULE__, :session_options, []}]]
 
   socket "/socket/agent", ToweropsWeb.AgentSocket,
     websocket: [connect_info: [:peer_data]],
@@ -119,7 +138,7 @@ defmodule ToweropsWeb.Endpoint do
 
   plug Plug.MethodOverride
   plug Plug.Head
-  plug Plug.Session, @session_options
+  plug ToweropsWeb.Plugs.RuntimeSession, @static_session_options
 
   # Enable SQL Sandbox metadata for LiveView tests
   # This allows LiveView processes to access the test's sandbox connection

lib/towerops_web/plugs/runtime_session.ex

@@ -0,0 +1,33 @@
+defmodule ToweropsWeb.Plugs.RuntimeSession do
+  @moduledoc """
+  Wrapper around `Plug.Session` that resolves session salts at runtime.
+
+  The signing and encryption salts come from environment variables
+  (`SESSION_SIGNING_SALT` / `SESSION_ENCRYPTION_SALT`) via `config/runtime.exs`.
+  Because `Plug.Session` is compiled into the endpoint pipeline with its
+  options baked in, this plug defers `Plug.Session.init/1` until the first
+  request and caches the result in `:persistent_term`.
+  """
+
+  @behaviour Plug
+
+  @impl true
+  def init(static_opts), do: static_opts
+
+  @impl true
+  def call(conn, static_opts) do
+    Plug.Session.call(conn, session_opts(static_opts))
+  end
+
+  defp session_opts(static_opts) do
+    case :persistent_term.get({__MODULE__, :opts}, nil) do
+      nil ->
+        opts = Plug.Session.init(static_opts ++ ToweropsWeb.Endpoint.runtime_session_opts())
+        :persistent_term.put({__MODULE__, :opts}, opts)
+        opts
+
+      opts ->
+        opts
+    end
+  end
+end