Don't exit the acceptor process when handling an econnaborted error

mtrudel · Dec 27, 2023 · 54d0476 · 54d0476
1 parent ab93507
commit 54d0476
Show file tree

Hide file tree

Showing 3 changed files with 25 additions and 2 deletions.
diff --git a/lib/thousand_island/acceptor.ex b/lib/thousand_island/acceptor.ex
@@ -30,7 +30,11 @@ defmodule ThousandIsland.Acceptor do
         ThousandIsland.Telemetry.span_event(span, :spawn_error)
         accept(listener_socket, connection_sup_pid, server_config, span, count + 1)
 
-      {:error, reason} when reason in [:closed, :einval, :econnaborted] ->
+      {:error, :econnaborted} ->
+        ThousandIsland.Telemetry.span_event(span, :econnaborted)
+        accept(listener_socket, connection_sup_pid, server_config, span, count + 1)
+
+      {:error, reason} when reason in [:closed, :einval] ->
         ThousandIsland.Telemetry.stop_span(span, %{connections: count})
 
       {:error, reason} ->

diff --git a/lib/thousand_island/logger.ex b/lib/thousand_island/logger.ex
@@ -21,7 +21,10 @@ defmodule ThousandIsland.Logger do
   """
   @spec attach_logger(log_level()) :: :ok | {:error, :already_exists}
   def attach_logger(:error) do
-    events = [[:thousand_island, :acceptor, :spawn_error]]
+    events = [
+      [:thousand_island, :acceptor, :spawn_error],
+      [:thousand_island, :acceptor, :econnaborted]
+    ]
 
     :telemetry.attach_many("#{__MODULE__}.error", events, &__MODULE__.log_error/4, nil)
   end

diff --git a/lib/thousand_island/telemetry.ex b/lib/thousand_island/telemetry.ex
@@ -97,6 +97,21 @@ defmodule ThousandIsland.Telemetry do
 
       * `telemetry_span_context`: A unique identifier for this span
 
+  * `[:thousand_island, :acceptor, :econnaborted]`
+
+      Thousand Island was unable to spawn a process to handle a connection since the remote end
+      closed before we could accept it. This usually occurs when it takes too long for your server
+      to start processing a connection; you may want to look at tuning OS-level TCP parameters or
+      adding more server capacity.
+
+      This event contains the following measurements:
+
+      * `monotonic_time`: The time of this event, in `:native` units
+
+      This event contains the following metadata:
+
+      * `telemetry_span_context`: A unique identifier for this span
+
   ## `[:thousand_island, :connection, *]`
 
   Represents Thousand Island handling a specific client request
@@ -281,6 +296,7 @@ defmodule ThousandIsland.Telemetry do
   @type event_name ::
           :ready
           | :spawn_error
+          | :econnaborted
           | :recv_error
           | :send_error
           | :sendfile_error