[esphome][core] Give a lost OTA chunk ack time to be retransmitted (#19041)

esphome/components/esphome/ota/ota_esphome.cpp

@@ -41,7 +41,10 @@ const noise::NoiseContext &ESPHomeOTAComponent::noise_context_() const {
 #endif
 static constexpr uint16_t OTA_BLOCK_SIZE = 8192;
 static constexpr uint32_t OTA_SOCKET_TIMEOUT_HANDSHAKE = 20000;  // milliseconds for initial handshake
-static constexpr uint32_t OTA_SOCKET_TIMEOUT_DATA = 90000;       // milliseconds for data transfer
+// Milliseconds for data transfer. Covers the lwIP retransmit run seen in
+// practice for a lost chunk ack (1.5 + 3 + 6 + 12 + 24 + 48 s); the CLI waits
+// longer (espota2.DATA_PHASE_TIMEOUT) so the device is free before it retries
+static constexpr uint32_t OTA_SOCKET_TIMEOUT_DATA = 105000;
 
 // Single-instance pointer — multi-port configs are rejected in final_validate.
 // NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)

esphome/espota2.py

@@ -96,6 +96,10 @@ UPLOAD_BUFFER_SIZE = UPLOAD_BLOCK_SIZE * 8
 # across the addresses on top of that.
 EXTRA_UPLOAD_ATTEMPTS = 2
 UPLOAD_RETRY_DELAY = 5.0
+# Data phase timeout; must stay longer than the device's OTA_SOCKET_TIMEOUT_DATA
+# (105 s) so a stalled session is gone before a retry, and long enough for lwIP
+# to get a lost chunk ack through after the retransmit run seen in practice
+DATA_PHASE_TIMEOUT = 160.0
 
 _LOGGER = logging.getLogger(__name__)
 
@@ -694,8 +698,7 @@ def perform_ota(
 
     _LOGGER.info("Handshake complete")
 
-    # Timeout must match device-side OTA_SOCKET_TIMEOUT_DATA to prevent premature failures
-    sock.settimeout(90.0)
+    sock.settimeout(DATA_PHASE_TIMEOUT)
 
     if extended_proto:
         send_check(sock, ota_type, "ota type")
@@ -854,7 +857,7 @@ def run_ota_impl_(
     # clean up a half-open connection (its handshake watchdog runs at 20s);
     # moving on to the next address family stays immediate. Known limitation:
     # a silent mid-transfer drop with no reset can wedge the device until its
-    # 90s data timeout, which outlasts this budget; the retries target the
+    # 105s data timeout, which outlasts this budget; the retries target the
     # common failures where the device resets or closes the link promptly.
     total_attempts = len(res) + EXTRA_UPLOAD_ATTEMPTS
     last_error = ""

tests/unit_tests/test_espota2.py

@@ -416,6 +416,9 @@ def test_perform_ota_no_auth(
         "Update took 14.00 seconds (prepare 2.00, upload 5.00, commit 7.00)"
         in caplog.text
     )
+    # The data phase timeout must outlast the device's 105 s data timeout
+    mock_socket.settimeout.assert_any_call(espota2.DATA_PHASE_TIMEOUT)
+    assert espota2.DATA_PHASE_TIMEOUT > 105.0
 
 
 @pytest.mark.usefixtures("mock_time")