Deployment and lifecycle management for Nix
fix: add backoff to garden socket reconnection
Use progressive backoff delays (200ms, 500ms, 1s, 2s) when the garden
socket disconnects, instead of immediately retrying. Resets the backoff
counter on successful connection.
sow-142
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+37
-5
+37
-5
apps/garden/lib/garden/socket.ex
+37
-5
apps/garden/lib/garden/socket.ex
···
95
95
#
96
96
97
97
@impl Slipstream
98
+
def handle_connect(socket) do
99
+
Logger.info(
100
+
msg: "Connected to websocket",
101
+
authority: socket.channel_config.uri.authority,
102
+
path: socket.channel_config.uri.path
103
+
)
104
+
105
+
socket = assign(socket, :reconnect_counter, 0)
106
+
{:ok, join(socket, @lobby_topic)}
107
+
end
108
+
109
+
@impl Slipstream
98
110
def init(_args) do
99
111
case do_connect() do
100
112
{:ok, socket} ->
···
113
125
@impl Slipstream
114
126
def handle_disconnect(_reason, socket) do
115
127
Logger.warning(msg: "Disconnected from server socket")
128
+
{:ok, schedule_reconnect(socket)}
129
+
end
116
130
117
-
case do_connect(socket) do
118
-
{:ok, socket} -> {:ok, socket}
119
-
{:error, reason} -> {:stop, reason, socket}
120
-
end
131
+
defp schedule_reconnect(socket) do
132
+
counter = Map.get(socket.assigns, :reconnect_counter, 0)
133
+
134
+
backoff_times =
135
+
Application.get_env(__MODULE__, :reconnect_after_msec, [200, 500, 1_000, 2_000])
136
+
137
+
delay = Enum.at(backoff_times, counter, List.last(backoff_times))
138
+
139
+
Logger.info(msg: "Scheduling reconnect", delay_ms: delay, attempt: counter + 1)
140
+
Process.send_after(self(), :attempt_reconnect, delay)
141
+
142
+
assign(socket, :reconnect_counter, counter + 1)
121
143
end
122
144
123
145
defp do_connect(socket \\ nil) do
···
256
278
nil
257
279
end
258
280
end
259
-
260
281
261
282
defp schedule_reauthentication(socket, %{expires_in: expires_in})
262
283
when is_integer(expires_in) do
···
515
536
send(self(), :check_pending_reload)
516
537
517
538
{:noreply, %{socket | active_deployments: active_deployments}}
539
+
end
540
+
end
541
+
542
+
def handle_info(:attempt_reconnect, socket) do
543
+
case do_connect(socket) do
544
+
{:ok, socket} ->
545
+
{:noreply, socket}
546
+
547
+
{:error, reason} ->
548
+
Logger.warning(msg: "Reconnect attempt failed", error: inspect(reason))
549
+
{:noreply, schedule_reconnect(socket)}
518
550
end
519
551
end
520
552