Explain retry choices between stalled multipath channels

tests/Makefile.am

@@ -1045,7 +1045,7 @@ test_dummynet_channels_LDFLAGS = $(test_clock_ldflags)
 test_etcp_multipath_SOURCES = test_etcp_multipath.c $(test_clock_sources)
 test_etcp_multipath_LDADD = $(LIBUTUN) $(COMMON_LIBS)
 test_etcp_multipath_LDFLAGS = $(test_clock_ldflags) -Wl,--wrap=etcp_dump_pkt_sections -Wl,--wrap=bbr_main \
-    -Wl,--wrap=etcp_tx_complete -Wl,--wrap=queue_data_put_with_index
+    -Wl,--wrap=etcp_tx_complete -Wl,--wrap=queue_data_put_with_index -Wl,--wrap=etcp_loadbalancer_select_link
 test_stcp_sigpipe_SOURCES = test_stcp_sigpipe.c
 test_stcp_sigpipe_LDADD = $(LIBUTUN) $(COMMON_LIBS)
 test_etcp_windows_LDADD = $(LIBUTUN) $(COMMON_LIBS)

tests/test_etcp_multipath.c

@@ -60,6 +60,7 @@ static uint64_t sample_logged[2][PATHS];
 static struct recovery_stats recovery;
 static uint32_t fault_rto_us[PATHS]; // максимум RTO каждого пути с начала отказа, независимо от измеренной паузы
 static uint32_t rx_trigger[2]; // DATA, разбор которого сейчас разблокировал выдачу сообщения
+static unsigned fault_paths;
 static FILE* csv;
 static const char* case_name;
 static const char* pubkeys[2] = {
@@ -188,6 +189,28 @@ int __wrap_queue_data_put_with_index(struct ll_queue* q, struct ll_entry* entry)
     return __real_queue_data_put_with_index(q, entry);
 }
 
+struct ETCP_LINK* __real_etcp_loadbalancer_select_link(struct ETCP_CONN*, uint32_t, const struct INFLIGHT_PACKET*);
+struct ETCP_LINK* __wrap_etcp_loadbalancer_select_link(struct ETCP_CONN* conn, uint32_t bytes, const struct INFLIGHT_PACKET* retry) {
+    struct ETCP_LINK* selected = __real_etcp_loadbalancer_select_link(conn, bytes, retry);
+    if (!measuring || !fault_start || conn != ep[0].conn || !retry || !selected || !retry->last_link) return selected;
+    int previous = -1, chosen = -1;
+    for (unsigned p = 0; p < path_count; p++) {
+        if (ss_get_port(&retry->last_link->remote_addr) == dummynet_get_listen_port(channels[p])) previous = p;
+        if (ss_get_port(&selected->remote_addr) == dummynet_get_listen_port(channels[p])) chosen = p;
+    }
+    if (previous < 0 || chosen < 0 || !(fault_paths & (1u << previous)) || !(fault_paths & (1u << chosen))) return selected;
+    for (struct ETCP_LINK* healthy = conn->links; healthy; healthy = healthy->next) for (unsigned p = 0; p < path_count; p++) {
+        if ((fault_paths & (1u << p)) || ss_get_port(&healthy->remote_addr) != dummynet_get_listen_port(channels[p])) continue;
+        int full = bytes > healthy->inflight_lim_bytes || healthy->inflight_bytes > healthy->inflight_lim_bytes - bytes;
+        DEBUG_INFO(DEBUG_CATEGORY_DEBUG, "FAULT retry-choice %s seq=%u attempt=%u previous=%d selected=%d elapsed_us=%llu "
+            "healthy=%u initialized=%u UP=%u pacing_blocked=%d window_full=%d inflight=%u cwnd=%u total_attempts=%u", case_name,
+            retry->seq, retry->attempt_count, previous, chosen, (unsigned long long)(get_time_us() - fault_start),
+            p, healthy->initialized, healthy->link_status, healthy->shaper_timer != NULL, full,
+            healthy->inflight_bytes, healthy->inflight_lim_bytes, conn->active_attempts);
+    }
+    return selected;
+}
+
 static void receive_message(struct ETCP_CONN* conn, struct ll_entry* entry) {
     int side = conn->instance == ep[0].inst ? 0 : 1;
     struct endpoint* dest = &ep[side];
@@ -240,6 +263,7 @@ static void setup(const struct profile* settings, unsigned count, unsigned mask)
     memset(sample_logged, 0, sizeof(sample_logged));
     memset(&recovery, 0, sizeof(recovery)); memset(fault_rto_us, 0, sizeof(fault_rto_us));
     memset(rx_trigger, 0, sizeof(rx_trigger));
+    fault_paths = 0;
     path_count = count; enabled_mask = mask; measuring = 0; fault_start = 0; fault_trace_count = 0;
     memcpy(profiles, settings, count * sizeof(*settings));
     ua = uasync_create(); CHECK(ua);
@@ -438,7 +462,7 @@ static void stalled(const struct profile* settings, unsigned count, unsigned bro
         run_for(1, 1, "arm-fault");
     }
     CHECK(pending == broken && fault_seq);
-    begin_measure(); fault_start = get_time_us();
+    begin_measure(); fault_start = get_time_us(); fault_paths = broken;
     for (struct ETCP_LINK* link = ep[0].conn->links; link; link = link->next)
         for (unsigned p = 0; p < count; p++)
             if (ss_get_port(&link->remote_addr) == dummynet_get_listen_port(channels[p])) fault_rto_us[p] = etcp_tx_rto_us(link);