fix(gateway): harden websocket streaming under client backpressure

components/maze32_web/include/maze32_gateway.h

@@ -20,7 +20,7 @@ extern "C" {
 #define MAZE32_GATEWAY_OUTGOING_LIMIT 768U
 #define MAZE32_GATEWAY_RATE_LIMIT_PER_SECOND 20U
 #define MAZE32_GATEWAY_RATE_LIMIT_CLOSE_THRESHOLD 40U
-#define MAZE32_GATEWAY_PLAYER_LATENCY_BUDGET_MS 250U
+#define MAZE32_GATEWAY_PLAYER_LATENCY_BUDGET_MS 500U
 
 typedef struct {
     int active;

components/maze32_web/maze32_gateway.c

@@ -1,5 +1,7 @@
 #include "maze32_gateway.h"
 
+#include <errno.h>
+#include <fcntl.h>
 #include <inttypes.h>
 #include <stdio.h>
 #include <string.h>
@@ -39,6 +41,30 @@ static unsigned int active_connection_count(const maze32_gateway_t* gateway) {
     return count;
 }
 
+static int gateway_send_all(httpd_handle_t server, int socket_fd, const char* buffer, size_t length,
+                            int flags) {
+    (void)server;
+    if (buffer == NULL) {
+        return HTTPD_SOCK_ERR_INVALID;
+    }
+    size_t offset = 0U;
+    while (offset < length) {
+        int sent = send(socket_fd, buffer + offset, length - offset, flags);
+        if (sent > 0) {
+            offset += (size_t)sent;
+            continue;
+        }
+        if (sent < 0 && errno == EINTR) {
+            continue;
+        }
+        if (sent < 0 && (errno == EAGAIN || errno == EWOULDBLOCK)) {
+            return HTTPD_SOCK_ERR_TIMEOUT;
+        }
+        return HTTPD_SOCK_ERR_FAIL;
+    }
+    return (int)offset;
+}
+
 static void sample_resources(maze32_gateway_t* gateway, const char* label) {
     size_t free_bytes = internal_free_bytes();
     size_t largest_block = internal_largest_block_bytes();
@@ -82,6 +108,10 @@ static maze32_gateway_connection_t* open_connection(maze32_gateway_t* gateway, i
             memset(connection, 0, sizeof(*connection));
             connection->active = 1;
             connection->socket_fd = socket_fd;
+            if (httpd_sess_set_send_override(gateway->server, socket_fd, gateway_send_all) !=
+                ESP_OK) {
+                ESP_LOGW(TAG, "Could not install bounded send callback for fd=%d", socket_fd);
+            }
             connection->rate_window_started_ms = now_ms();
             unsigned int active_count = active_connection_count(gateway);
             if (active_count > gateway->metrics.connection_high_water) {
@@ -202,6 +232,7 @@ static int drain_queue(maze32_gateway_t* gateway) {
             .len = entry->length,
         };
         int ready = 1;
+        int original_flags = -1;
         if (entry->drop_if_busy) {
             fd_set write_set;
             FD_ZERO(&write_set);
@@ -209,6 +240,12 @@ static int drain_queue(maze32_gateway_t* gateway) {
             struct timeval timeout = {0};
             ready = select(entry->socket_fd + 1, NULL, &write_set, NULL, &timeout) > 0 &&
                     FD_ISSET(entry->socket_fd, &write_set);
+            if (ready) {
+                original_flags = fcntl(entry->socket_fd, F_GETFL, 0);
+                if (original_flags >= 0) {
+                    (void)fcntl(entry->socket_fd, F_SETFL, original_flags | O_NONBLOCK);
+                }
+            }
         }
         if (!ready) {
             gateway->metrics.messages_dropped++;
@@ -229,6 +266,9 @@ static int drain_queue(maze32_gateway_t* gateway) {
                 close_slow_connection(gateway, entry->socket_fd);
             }
         }
+        if (original_flags >= 0) {
+            (void)fcntl(entry->socket_fd, F_SETFL, original_flags);
+        }
         gateway->outgoing_head = (gateway->outgoing_head + 1U) % MAZE32_GATEWAY_QUEUE_LIMIT;
         gateway->outgoing_count--;
     }
@@ -561,7 +601,9 @@ static void countdown_finished_work(void* context) {
             }
         }
     }
+    drain_queue(gateway);
     broadcast_snapshot(gateway);
+    drain_queue(gateway);
     broadcast_views(gateway, 1);
     drain_queue(gateway);
     ESP_LOGI(TAG, "Lobby became active at server time %" PRIu64 " ms",

components/maze32_web/maze32_web.c

@@ -1,5 +1,6 @@
 #include "maze32_web.h"
 
+#include <fcntl.h>
 #include <stdio.h>
 
 #include "esp_log.h"
@@ -107,7 +108,7 @@ esp_err_t maze32_web_start(httpd_handle_t* server, maze32_health_t* health,
     configuration.max_open_sockets = 9;
     configuration.lru_purge_enable = false;
     configuration.recv_wait_timeout = 5;
-    configuration.send_wait_timeout = 1;
+    configuration.send_wait_timeout = 5;
     configuration.global_user_ctx = gateway;
     configuration.global_user_ctx_free_fn = preserve_static_context;
     configuration.close_fn = maze32_gateway_close_socket;