Shutdown node when component startup fails

CHANGELOG.md

@@ -11,6 +11,10 @@ Every entry has a category for which we use the following visual abbreviations:
 
 ## Unreleased
 
+- 🐞 VAST did not terminate when a critical component failed during startup.
+  VAST now binds the lifetime of the node to all critical components.
+  [#1028](https://github.com/tenzir/vast/pull/1028)
+
 - 🐞 VAST would overwrite existing on-disk state data when encountering a partial
   read during startup. This state-corrupting behavior no longer exists.
   [#1026](https://github.com/tenzir/vast/pull/1026)

libvast/src/system/component_registry.cpp

@@ -43,6 +43,20 @@ bool component_registry::remove(const caf::actor& comp) {
   return false;
 }
 
+const std::string*
+component_registry::find_label_for(const caf::actor& comp) const {
+  auto pred = [&](auto& x) { return x.second.actor == comp; };
+  auto i = std::find_if(components_.begin(), components_.end(), pred);
+  return i != components_.end() ? &i->first : nullptr;
+}
+
+const std::string*
+component_registry::find_type_for(const caf::actor& comp) const {
+  auto pred = [&](auto& x) { return x.second.actor == comp; };
+  auto i = std::find_if(components_.begin(), components_.end(), pred);
+  return i != components_.end() ? &i->second.type : nullptr;
+}
+
 caf::actor component_registry::find_by_label(std::string_view label) const {
   // TODO: remove string conversion in with C++20 transparent keys.
   if (auto i = components_.find(std::string{label}); i != components_.end())

libvast/src/system/node.cpp

@@ -391,6 +391,15 @@ caf::behavior node(node_actor* self, std::string name, path dir,
   self->set_down_handler([=](const down_msg& msg) {
     VAST_DEBUG(self, "got DOWN from", msg.source);
     auto component = caf::actor_cast<caf::actor>(msg.source);
+    auto type = self->state.registry.find_type_for(component);
+    // All monitored components are in the registry.
+    VAST_ASSERT(type != nullptr);
+    if (is_singleton(*type)) {
+      auto label = self->state.registry.find_label_for(component);
+      VAST_ASSERT(label != nullptr); // Per the above assertion.
+      VAST_ERROR(self, "got DOWN from", *label, "; initiating shutdown");
+      self->send_exit(self, caf::exit_reason::user_shutdown);
+    }
     self->state.registry.remove(component);
   });
   // Terminate deterministically on shutdown.

libvast/vast/system/component_registry.hpp

@@ -49,6 +49,18 @@ class component_registry {
   /// @returns `true` iff the component was deleted successfully.
   bool remove(const caf::actor& comp);
 
+  /// Finds the label of a given component actor.
+  /// @param comp The component actor.
+  /// @returns A pointer to the name of the label of *comp* or `nullptr` if
+  ///          *comp* is not known.
+  const std::string* find_label_for(const caf::actor& comp) const;
+
+  /// Finds the type of a given component actor.
+  /// @param comp The component actor.
+  /// @returns A pointer to the name of the type of *comp* or `nullptr` if
+  ///          *comp* is not known.
+  const std::string* find_type_for(const caf::actor& comp) const;
+
   /// Locates a component by label.
   /// @param label The label of the component to lookup.
   /// @returns The respective component actor if found.