434 lines
25 KiB
C
434 lines
25 KiB
C
/* Настоящие ETCP/NCD/BBR поверх независимых UDP-каналов. Строгие проверки скорости и пауз при DROP/HOLD. */
|
||
#include <stdlib.h>
|
||
#include <string.h>
|
||
#include <stdio.h>
|
||
#include "etcp.h"
|
||
#include "etcp_api.h"
|
||
#include "etcp_session.h"
|
||
#include "node_conn_direct.h"
|
||
#include "topo_node.h"
|
||
#include "utun_instance.h"
|
||
#include "dummynet.h"
|
||
#include "test_net.h"
|
||
#include "test_clock.h"
|
||
#include "../lib/mem.h"
|
||
|
||
#define CHECK(x) do { if (!(x)) { DEBUG_ERROR(DEBUG_CATEGORY_GENERAL, "FAIL %s:%d: %s", __func__, __LINE__, #x); exit(1); } } while (0)
|
||
#define PATHS 3
|
||
#define TRACE_COUNT 131072
|
||
#define CMD 253
|
||
#define SAMPLE_MS 4000
|
||
#define WARMUP_MS 3000
|
||
#define STALL_MS 3000
|
||
#define MAX_GAP_US 100000
|
||
struct profile { unsigned rtt_ms, kbps; };
|
||
struct message { uint8_t cmd, from, reserved[2]; uint32_t seq; uint64_t enqueued; uint8_t body[1008]; };
|
||
struct trace { uint64_t first_tx, first_rx; uint32_t attempts; uint8_t first_path; };
|
||
struct endpoint {
|
||
struct UTUN_INSTANCE* inst;
|
||
struct NODE_CONN_DIRECT* owner;
|
||
struct ETCP_CONN* conn;
|
||
struct ETCP_SOCKET* sockets[PATHS];
|
||
unsigned sent, received;
|
||
uint64_t bytes, last_delivery, max_gap, max_latency, latency_hist[10001], samples;
|
||
};
|
||
struct path_stats { uint64_t data, retries, rx_unique; };
|
||
static struct UASYNC* ua;
|
||
static struct endpoint ep[2];
|
||
static struct dummynet* channels[PATHS];
|
||
static struct profile profiles[PATHS];
|
||
static struct trace traces[2][TRACE_COUNT];
|
||
static struct path_stats traffic[2][PATHS];
|
||
static unsigned path_count, failures, enabled_mask, bidirectional, low_load, measuring;
|
||
static uint64_t started, next_message[2], measure_start, fault_start;
|
||
static uint32_t fault_seq;
|
||
static unsigned fault_trace_count;
|
||
static FILE* csv;
|
||
static const char* case_name;
|
||
static const char* pubkeys[2] = {
|
||
"b3193173def895bd0fcea6f86af077c7d77216f10395275f627ac18242ec0f01",
|
||
"ce8871f07fa056c636d297115f231b08c29cdf94e0d440fce83a07c34416d36a"
|
||
};
|
||
static const char* privkeys[2] = {
|
||
"704f2e012c8fa8768130cb0f988a997dccb628372bc5ceccacc78dcbfec5916f",
|
||
"38240cb82199e504686507f11f6eaa4f740fde6f0c425c495e49a523019a5d68"
|
||
};
|
||
|
||
/* Наблюдение не заменяет передачу/приём и не меняет состояние протокола. */
|
||
void __real_etcp_dump_pkt_sections(struct ETCP_DGRAM*, struct ETCP_LINK*, int);
|
||
void __wrap_etcp_dump_pkt_sections(struct ETCP_DGRAM* pkt, struct ETCP_LINK* link, int sending) {
|
||
int side = link->etcp->instance == ep[0].inst ? 0 : 1, path = -1;
|
||
for (unsigned p = 0; p < path_count; p++)
|
||
if (ss_get_port(&link->remote_addr) == dummynet_get_listen_port(channels[p])) path = p;
|
||
if (path >= 0) for (unsigned offset = 0; offset < pkt->data_len;) {
|
||
const uint8_t* data = pkt->data + offset;
|
||
unsigned remaining = pkt->data_len - offset, len;
|
||
switch (data[0]) {
|
||
case ETCP_SECTION_ACK: CHECK(remaining >= 8); len = 8 + 8 * data[1]; break;
|
||
case ETCP_SECTION_TIMESTAMP: len = 5; break;
|
||
case ETCP_SECTION_MEAS_TS: len = 9; break;
|
||
case ETCP_SECTION_MEAS_RESP: len = 13; break;
|
||
case ETCP_SECTION_FILLER: CHECK(remaining >= 3); len = 3 + data[1] + (data[2] << 8); break;
|
||
case ETCP_SECTION_PAYLOAD: {
|
||
CHECK(remaining >= 5);
|
||
uint32_t seq = (uint32_t)data[1] | ((uint32_t)data[2] << 8) | ((uint32_t)data[3] << 16) | ((uint32_t)data[4] << 24);
|
||
CHECK(seq < TRACE_COUNT);
|
||
struct trace* trace = &traces[sending ? side : 1-side][seq];
|
||
uint64_t now = get_time_us();
|
||
if (sending) {
|
||
if (trace->attempts++ && measuring) traffic[side][path].retries++;
|
||
if (!trace->first_tx) { trace->first_tx = now; trace->first_path = path; }
|
||
if (measuring) traffic[side][path].data += remaining - 5;
|
||
} else if (!trace->first_rx) {
|
||
trace->first_rx = now;
|
||
if (measuring) traffic[1-side][path].rx_unique += remaining - 5;
|
||
}
|
||
if (fault_start && (seq == fault_seq || (sending && trace->attempts > 1 && fault_trace_count++ < 32)))
|
||
DEBUG_INFO(DEBUG_CATEGORY_DEBUG, "FAULT trace dir=%d seq=%u path=%d tx=%d attempts=%u elapsed_us=%llu", side,
|
||
seq, path, sending, trace->attempts, (unsigned long long)(now - fault_start));
|
||
len = remaining; break;
|
||
}
|
||
default: len = remaining; break;
|
||
}
|
||
CHECK(len <= remaining); offset += len;
|
||
}
|
||
__real_etcp_dump_pkt_sections(pkt, link, sending);
|
||
}
|
||
|
||
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];
|
||
CHECK(entry->len == sizeof(struct message));
|
||
struct message msg; memcpy(&msg, entry->dgram, sizeof(msg));
|
||
CHECK(msg.cmd == CMD && msg.from == 1-side && msg.seq == dest->received);
|
||
for (unsigned i = 0; i < sizeof(msg.body); i++) CHECK(msg.body[i] == (uint8_t)(msg.seq * 3 + msg.from + i));
|
||
dest->received++; dest->bytes += sizeof(msg);
|
||
uint64_t now = get_time_us();
|
||
if (measuring) {
|
||
uint64_t gap = now - dest->last_delivery, latency = now - msg.enqueued;
|
||
if (gap > dest->max_gap) dest->max_gap = gap;
|
||
if (latency > dest->max_latency) dest->max_latency = latency;
|
||
dest->latency_hist[latency / 1000 < 10000 ? latency / 1000 : 10000]++; dest->samples++;
|
||
}
|
||
dest->last_delivery = now;
|
||
queue_dgram_free(entry); queue_entry_free(entry);
|
||
}
|
||
|
||
static void create_endpoint(unsigned side) {
|
||
char config[2048], listeners[768] = "";
|
||
for (unsigned p = 0; p < path_count; p++) {
|
||
char text[160]; snprintf(text, sizeof(text), "[server:p%u]\naddr=127.0.0.1:0\ntype=public\n", p);
|
||
strcat(listeners, text);
|
||
}
|
||
snprintf(config, sizeof(config), "[global]\nmy_node_name=multipath-%u\nmy_public_key=%s\nmy_private_key=%s\n"
|
||
"keepalive_interval=100\nkeepalive_timeout=2000\nkeepalive_adaptive=0\nauto_sockets=no\n"
|
||
"[allowed_keys]\nallow_all=1\n%s",
|
||
side, pubkeys[side], privkeys[side], listeners);
|
||
ep[side].inst = utun_instance_create_from_str(ua, config); CHECK(ep[side].inst);
|
||
CHECK(utun_core_start(ep[side].inst) == 0);
|
||
CHECK(etcp_bind(ep[side].inst, CMD, receive_message) == 0);
|
||
for (unsigned p = 0; p < path_count; p++) {
|
||
char name[16]; snprintf(name, sizeof(name), "p%u", p);
|
||
ep[side].sockets[p] = test_etcp_bound_socket(ep[side].inst, name);
|
||
}
|
||
}
|
||
|
||
static void setup(const struct profile* settings, unsigned count, unsigned mask) {
|
||
memset(ep, 0, sizeof(ep)); memset(channels, 0, sizeof(channels));
|
||
memset(traces, 0, sizeof(traces)); memset(traffic, 0, sizeof(traffic));
|
||
memset(next_message, 0, sizeof(next_message));
|
||
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);
|
||
create_endpoint(0); create_endpoint(1);
|
||
for (unsigned p = 0; p < count; p++) {
|
||
if (!(mask & (1u << p))) continue;
|
||
channels[p] = dummynet_create(ua, "127.0.0.1", 0); CHECK(channels[p]);
|
||
for (int dir = 0; dir < 2; dir++) CHECK(dummynet_set_direction(channels[p], dir, settings[p].rtt_ms / 2, 0,
|
||
settings[p].kbps, 512, 0, "127.0.0.1", ss_get_port(&ep[1-dir].sockets[p]->local_addr)) == 0);
|
||
struct TOPO_ADDR4 addr = { .protocol = TOPO_PROTO_UDP, .port = dummynet_get_listen_port(channels[p]) };
|
||
uint32_t ip = htonl(INADDR_LOOPBACK); memcpy(addr.addr, &ip, sizeof(ip));
|
||
struct TOPO_NODE peer = { .node_id = ep[1].inst->node_id, .node_name = "multipath-peer", .v4_addrs = &addr };
|
||
memcpy(peer.public_key, ep[1].inst->my_keys.public_key, SC_PUBKEY_SIZE);
|
||
struct NODE_CONN_DIRECT* extra = NULL;
|
||
CHECK(node_conn_direct_open_node(ep[0].inst, peer.node_id, NULL, NULL, &extra, &peer, ep[0].sockets[p]) >= 0);
|
||
if (!ep[0].owner) ep[0].owner = extra;
|
||
else node_conn_direct_close(extra);
|
||
}
|
||
uint64_t end = get_time_us() + 10000000;
|
||
int ready = 0;
|
||
while (get_time_us() < end) {
|
||
test_clock_poll(ua, 10); ready = 1;
|
||
for (int i = 0; i < 2; i++) {
|
||
ep[i].conn = instance_find_conn(ep[i].inst, ep[1-i].inst->node_id);
|
||
struct ETCP_CONN* c = ep[i].conn;
|
||
if (!c || !c->initialized || !c->links_up || c->session_required || c->reinit_pending) { ready = 0; continue; }
|
||
for (unsigned p = 0; p < count; p++) if (mask & (1u << p)) {
|
||
int found = 0;
|
||
for (struct ETCP_LINK* l = c->links; l; l = l->next)
|
||
if (ss_get_port(&l->remote_addr) == dummynet_get_listen_port(channels[p]) && l->initialized && l->link_status) found++;
|
||
if (found != 1) ready = 0;
|
||
}
|
||
}
|
||
if (ready) break;
|
||
}
|
||
CHECK(ready);
|
||
struct TOPO_NODE peer = { .node_id = ep[0].inst->node_id, .node_name = "held-peer" };
|
||
memcpy(peer.public_key, ep[0].inst->my_keys.public_key, SC_PUBKEY_SIZE);
|
||
CHECK(node_conn_direct_open_node(ep[1].inst, peer.node_id, NULL, NULL, &ep[1].owner, &peer, NULL) >= 0);
|
||
started = get_time_us();
|
||
for (int i = 0; i < 2; i++) ep[i].last_delivery = started;
|
||
}
|
||
|
||
/* Ограниченная очередь источника: latency не скрывается за неограниченным накоплением тестовых сообщений. */
|
||
static void produce(unsigned side) {
|
||
struct endpoint* source = &ep[side];
|
||
uint64_t now = get_time_us();
|
||
if (low_load && now < next_message[side]) return;
|
||
for (unsigned budget = 0; budget < 64 && source->conn->send_input_q->count < 4; budget++) {
|
||
if (source->sent - ep[1-side].received >= 2048) break;
|
||
struct ll_entry* entry = ll_alloc_lldgram(sizeof(struct message)); CHECK(entry);
|
||
struct message msg = { .cmd = CMD, .from = side, .seq = source->sent, .enqueued = now };
|
||
for (unsigned i = 0; i < sizeof(msg.body); i++) msg.body[i] = msg.seq * 3 + side + i;
|
||
memcpy(entry->dgram, &msg, sizeof(msg)); entry->len = sizeof(msg);
|
||
CHECK(etcp_send(source->conn, entry) == 0); source->sent++;
|
||
if (low_load) { next_message[side] = now + 10000; break; }
|
||
}
|
||
}
|
||
|
||
static void snapshot(const char* phase) {
|
||
uint64_t now = get_time_us();
|
||
for (int side = 0; side < 2; side++) {
|
||
struct ETCP_CONN* conn = ep[side].conn;
|
||
for (struct ETCP_LINK* l = conn->links; l; l = l->next) {
|
||
uint64_t occupied = 0; unsigned packets = 0;
|
||
for (struct ll_entry* e = conn->input_wait_ack->head; e; e = e->next) {
|
||
struct INFLIGHT_PACKET* pkt = (struct INFLIGHT_PACKET*)e;
|
||
CHECK(pkt->state == INFLIGHT_STATE_WAIT_ACK);
|
||
if (pkt->last_link == l) { occupied += pkt->ll.len; packets++; }
|
||
}
|
||
CHECK(l->inflight_bytes == occupied && l->inflight_packets == packets);
|
||
int p = -1;
|
||
for (unsigned j = 0; j < path_count; j++)
|
||
if (ss_get_port(&l->remote_addr) == dummynet_get_listen_port(channels[j])) p = j;
|
||
if (p < 0) continue;
|
||
const struct dummynet_stats* ds = dummynet_get_stats(channels[p], side);
|
||
if (csv) fprintf(csv, "%s,%x,%s,%llu,%d,%d,%u,%u,%u,%u,%u,%u,%d,%d,%llu,%llu,%llu,%u,%llu,%llu,%d,%u,%llu\n",
|
||
case_name, enabled_mask, phase, (unsigned long long)((now-started)/1000), side, p, l->rtt_last, l->bandwidth,
|
||
l->inflight_bytes, l->inflight_lim_bytes, conn->rtt_avg_10, conn->jitter, conn->input_send_q->count,
|
||
conn->recv_q->count, (unsigned long long)ds->sent_bytes, (unsigned long long)traffic[side][p].rx_unique,
|
||
(unsigned long long)traffic[side][p].retries, ds->queue_size,
|
||
(unsigned long long)ep[1-side].bytes, (unsigned long long)ep[1-side].max_gap,
|
||
ep[1-side].conn->recv_q->count, ep[1-side].conn->last_delivered_id+1, (unsigned long long)ds->dropped);
|
||
}
|
||
CHECK(conn->unacked_bytes == conn->input_send_q->total_bytes + conn->input_wait_ack->total_bytes);
|
||
}
|
||
}
|
||
|
||
static void run_for(unsigned ms, int sending, const char* phase) {
|
||
uint64_t end = get_time_us() + ms * 1000ULL, next_snapshot = get_time_us();
|
||
while (get_time_us() < end) {
|
||
if (sending) { produce(0); if (bidirectional) produce(1); }
|
||
test_clock_poll(ua, 10);
|
||
for (int i = 0; i < 2; i++) CHECK(ep[i].conn->state != 2 && !ep[i].conn->reinit_pending);
|
||
if (get_time_us() >= next_snapshot) { snapshot(phase); next_snapshot = get_time_us() + 100000; }
|
||
if (!sending && ep[0].sent == ep[1].received && ep[1].sent == ep[0].received &&
|
||
!ep[0].conn->unacked_bytes && !ep[1].conn->unacked_bytes) break;
|
||
}
|
||
}
|
||
|
||
static void begin_measure(void) {
|
||
measure_start = get_time_us(); measuring = 1;
|
||
memset(traffic, 0, sizeof(traffic));
|
||
for (int i = 0; i < 2; i++) {
|
||
ep[i].max_gap = ep[i].max_latency = ep[i].samples = 0; ep[i].last_delivery = measure_start;
|
||
memset(ep[i].latency_hist, 0, sizeof(ep[i].latency_hist));
|
||
}
|
||
for (unsigned p = 0; p < path_count; p++) if (channels[p]) dummynet_reset_stats(channels[p], -1);
|
||
}
|
||
|
||
static void expect(int passed, const char* requirement, uint64_t actual, uint64_t limit) {
|
||
if (passed) return;
|
||
failures++;
|
||
DEBUG_ERROR(DEBUG_CATEGORY_DEBUG, "FAIL %s: %s actual=%llu limit=%llu", case_name, requirement,
|
||
(unsigned long long)actual, (unsigned long long)limit);
|
||
}
|
||
|
||
static uint64_t percentile(const struct endpoint* endpoint, unsigned percent) {
|
||
uint64_t sum = 0, target = (endpoint->samples * percent + 99) / 100;
|
||
if (!target) return 0;
|
||
for (unsigned ms = 0; ms <= 10000; ms++) {
|
||
sum += endpoint->latency_hist[ms];
|
||
if (sum >= target) return ms;
|
||
}
|
||
return 10000;
|
||
}
|
||
|
||
static void teardown(void) {
|
||
measuring = 0;
|
||
// Уничтожаем и активные delay/HOLD пакеты: эмуляторы не должны пережить свои endpoints.
|
||
for (unsigned p = 0; p < path_count; p++) { dummynet_destroy(channels[p]); channels[p] = NULL; }
|
||
for (int i = 0; i < 2; i++) node_conn_direct_close(ep[i].owner);
|
||
uasync_drain_immediate(ua);
|
||
for (int i = 0; i < 2; i++) { utun_instance_destroy(ep[i].inst); ep[i].inst = NULL; }
|
||
uasync_poll(ua, 0); CHECK(ua->timer_alloc_count == ua->timer_free_count); uasync_destroy(ua, 0);
|
||
}
|
||
|
||
/* Полезные байты приложения — отдельный критерий; FILLER/повторы не могут улучшить результат. */
|
||
static uint64_t measure_rate(const struct profile* settings, unsigned count, unsigned mask) {
|
||
setup(settings, count, mask); low_load = 0;
|
||
run_for(WARMUP_MS, 1, "warmup"); begin_measure();
|
||
uint64_t before[2] = {ep[0].bytes, ep[1].bytes};
|
||
run_for(SAMPLE_MS, 1, "steady");
|
||
uint64_t elapsed = get_time_us() - measure_start, rate = 0;
|
||
for (unsigned side = 0; side <= bidirectional; side++) {
|
||
uint64_t useful = (ep[1-side].bytes-before[1-side]) * 8000 / elapsed, capacity = 0, unique_total = 0;
|
||
for (unsigned p = 0; p < count; p++) if (mask & (1u << p)) {
|
||
uint64_t wire = dummynet_get_stats(channels[p], side)->sent_bytes * 8000 / elapsed;
|
||
uint64_t unique = traffic[side][p].rx_unique * 8000 / elapsed;
|
||
unique_total += traffic[side][p].rx_unique;
|
||
capacity += settings[p].kbps;
|
||
DEBUG_INFO(DEBUG_CATEGORY_DEBUG, "RATE %s dir=%u path=%u RTT=%u BWlim=%u wire=%llu unique=%llu kbps retries=%llu",
|
||
case_name, side, p, settings[p].rtt_ms, settings[p].kbps, (unsigned long long)wire,
|
||
(unsigned long long)unique, (unsigned long long)traffic[side][p].retries);
|
||
expect(wire >= settings[p].kbps * 90 / 100, "channel wire utilization >=90%", wire, settings[p].kbps * 90 / 100);
|
||
expect(unique >= settings[p].kbps * 80 / 100, "channel unique DATA >=80%", unique, settings[p].kbps * 80 / 100);
|
||
}
|
||
uint64_t fresh = unique_total * 8000 / elapsed;
|
||
DEBUG_INFO(DEBUG_CATEGORY_DEBUG, "GOODPUT %s dir=%u application=%llu fresh_DATA=%llu capacity=%llu kbps", case_name, side,
|
||
(unsigned long long)useful, (unsigned long long)fresh, (unsigned long long)capacity);
|
||
// Нельзя пройти за счёт выдачи фрагментов, накопленных до начала измерения.
|
||
if (fresh < useful) useful = fresh;
|
||
if (!side) rate = useful;
|
||
expect(useful >= capacity * 85 / 100, "application goodput >=85% capacity", useful, capacity * 85 / 100);
|
||
}
|
||
measuring = 0;
|
||
run_for(10000, 0, "drain");
|
||
for (int i = 0; i < 2; i++) expect(ep[i].sent == ep[1-i].received, "complete ordered delivery after drain",
|
||
ep[1-i].received, ep[i].sent);
|
||
teardown(); return rate;
|
||
}
|
||
|
||
/* Не ждём восстановления отказавшего пути: проверяем паузу и скорость пока DROP/HOLD ещё включён. */
|
||
static void stalled(const struct profile* settings, unsigned count, unsigned broken, unsigned directions,
|
||
enum dummynet_mode mode, int sparse, unsigned duration_ms) {
|
||
low_load = sparse;
|
||
setup(settings, count, (1u << count)-1);
|
||
run_for(WARMUP_MS, 1, "warmup");
|
||
// Выбираем момент, когда на каждом останавливаемом пути есть действительно неподтверждённый пакет.
|
||
uint64_t deadline = get_time_us() + 2000000;
|
||
unsigned pending = 0;
|
||
fault_seq = 0;
|
||
while (get_time_us() < deadline) {
|
||
pending = 0; fault_seq = 0;
|
||
for (struct ll_entry* entry = ep[0].conn->input_wait_ack->head; entry; entry = entry->next) {
|
||
struct INFLIGHT_PACKET* pkt = (struct INFLIGHT_PACKET*)entry;
|
||
if (!pkt->last_link || ((directions & 1) && traces[0][pkt->seq].first_rx)) continue;
|
||
for (unsigned p = 0; p < count; p++) if ((broken & (1u << p)) &&
|
||
ss_get_port(&pkt->last_link->remote_addr) == dummynet_get_listen_port(channels[p])) {
|
||
// Пакет должен ещё находиться в propagation delay, а не уже лежать в сокете получателя.
|
||
if ((directions & 1) && get_time_us() - traces[0][pkt->seq].first_tx >= settings[p].rtt_ms * 500ULL) continue;
|
||
pending |= 1u << p;
|
||
if (!fault_seq || pkt->seq < fault_seq) fault_seq = pkt->seq;
|
||
}
|
||
}
|
||
if (pending == broken) break;
|
||
run_for(1, 1, "arm-fault");
|
||
}
|
||
CHECK(pending == broken && fault_seq);
|
||
begin_measure(); fault_start = get_time_us();
|
||
uint32_t epochs[2] = {ep[0].conn->reinit_count, ep[1].conn->reinit_count};
|
||
DEBUG_INFO(DEBUG_CATEGORY_DEBUG, "FAULT %s mode=%d paths=%x directions=%x seq=%u duration_ms=%u sparse=%d",
|
||
case_name, mode, broken, directions, fault_seq, duration_ms, sparse);
|
||
for (unsigned p = 0; p < count; p++) if (broken & (1u << p))
|
||
for (unsigned dir = 0; dir < 2; dir++) if (directions & (1u << dir))
|
||
CHECK(dummynet_set_mode(channels[p], dir, mode) == 0);
|
||
uint64_t before = ep[1].bytes;
|
||
run_for(duration_ms > 1000 ? 1000 : duration_ms, 1, "fault-transition");
|
||
uint64_t steady_bytes = ep[1].bytes, steady_start = get_time_us();
|
||
if (duration_ms > 1000) run_for(duration_ms - 1000, 1, "fault-steady");
|
||
uint64_t now = get_time_us(), tail_gap = now - ep[1].last_delivery;
|
||
if (tail_gap > ep[1].max_gap) ep[1].max_gap = tail_gap;
|
||
uint64_t capacity = 0;
|
||
for (unsigned p = 0; p < count; p++) if (!(broken & (1u << p))) capacity += settings[p].kbps;
|
||
uint64_t rate = now > steady_start ? (ep[1].bytes-steady_bytes) * 8000 / (now-steady_start) : 0;
|
||
DEBUG_INFO(DEBUG_CATEGORY_DEBUG, "FAULT result %s delivered=%llu gap_us=%llu p50=%llu p95=%llu p99=%llu max_latency_us=%llu "
|
||
"steady_kbps=%llu healthy_capacity=%llu missing_seq=%u first_rx_us=%llu attempts=%u", case_name,
|
||
(unsigned long long)(ep[1].bytes-before), (unsigned long long)ep[1].max_gap,
|
||
(unsigned long long)percentile(&ep[1], 50), (unsigned long long)percentile(&ep[1], 95),
|
||
(unsigned long long)percentile(&ep[1], 99), (unsigned long long)ep[1].max_latency,
|
||
(unsigned long long)rate, (unsigned long long)capacity, fault_seq,
|
||
(unsigned long long)(traces[0][fault_seq].first_rx > fault_start ? traces[0][fault_seq].first_rx-fault_start : 0),
|
||
traces[0][fault_seq].attempts);
|
||
if (duration_ms > 100) expect(ep[1].bytes > before, "delivery during active fault", ep[1].bytes-before, 1);
|
||
expect(ep[1].max_gap <= MAX_GAP_US, "maximum delivery pause <=100ms", ep[1].max_gap, MAX_GAP_US);
|
||
if (sparse) expect(ep[1].max_latency <= MAX_GAP_US, "bounded-source latency <=100ms", ep[1].max_latency, MAX_GAP_US);
|
||
else if (duration_ms > 1000) expect(rate >= capacity * 85 / 100,
|
||
"fault steady goodput >=85% healthy capacity", rate, capacity * 85 / 100);
|
||
for (int i = 0; i < 2; i++) CHECK(ep[i].conn->reinit_count == epochs[i] && ep[i].conn->links_up);
|
||
measuring = duration_ms <= 100;
|
||
for (unsigned p = 0; p < count; p++) if (broken & (1u << p))
|
||
for (unsigned dir = 0; dir < 2; dir++) CHECK(dummynet_set_mode(channels[p], dir, DUMMYNET_RUN) == 0);
|
||
run_for(1000, 1, "recovery");
|
||
if (duration_ms <= 100) {
|
||
expect(ep[1].samples > 0, "delivery after short HOLD", ep[1].samples, 1);
|
||
expect(ep[1].max_gap <= MAX_GAP_US, "short HOLD including resume: pause <=100ms", ep[1].max_gap, MAX_GAP_US);
|
||
expect(ep[1].max_latency <= MAX_GAP_US, "short HOLD including resume: latency <=100ms", ep[1].max_latency, MAX_GAP_US);
|
||
}
|
||
measuring = 0;
|
||
run_for(2000, 1, "recovery"); run_for(10000, 0, "drain");
|
||
for (int i = 0; i < 2; i++) expect(ep[i].sent == ep[1-i].received, "complete delivery after recovery",
|
||
ep[1-i].received, ep[i].sent);
|
||
teardown();
|
||
}
|
||
|
||
int main(int argc, char** argv) {
|
||
debug_config_init(); debug_set_level(DEBUG_LEVEL_ERROR); test_clock_init(10);
|
||
if (getenv("UTUN_TEST_DEBUG")) debug_set_category_level(DEBUG_CATEGORY_ETCP, DEBUG_LEVEL_DEBUG);
|
||
CHECK(socket_platform_init() == 0);
|
||
const char* output = getenv("UTUN_MULTIPATH_CSV");
|
||
if (output) { csv = fopen(output, "w"); CHECK(csv); fprintf(csv,
|
||
"case,mask,phase,time_ms,direction,path,rtt_tb,pacing_kbps,inflight,cwnd,conn_rtt_tb,jitter,send_q,local_recv_q,wire_bytes,"
|
||
"unique_bytes,retries,channel_q,app_bytes,max_gap_us,dest_recv_q,needed_seq,queue_drops\n"); }
|
||
struct profile matrix[][3] = {
|
||
{{20,8000},{20,8000}}, {{20,2000},{20,8000}}, {{20,8000},{200,8000}},
|
||
{{20,2000},{200,8000}}, {{20,8000},{200,2000}}, {{20,2000},{80,8000},{200,2000}}, {{20,2000},{80,8000}}
|
||
};
|
||
const char* names[] = {"equal", "bandwidth", "rtt", "fast-distant", "slow-distant", "three", "bidirectional"};
|
||
size_t baseline = u_get_allocated_count();
|
||
unsigned selected = 0;
|
||
for (unsigned n = 0; n < sizeof(names)/sizeof(names[0]); n++) {
|
||
if (argc > 1 && strcmp(argv[1], names[n])) continue;
|
||
selected++;
|
||
bidirectional = n == 6;
|
||
case_name = names[n]; unsigned count = n == 5 ? 3 : 2;
|
||
uint64_t solo_sum = 0;
|
||
for (unsigned p = 0; p < count; p++) solo_sum += measure_rate(matrix[n], count, 1u << p);
|
||
uint64_t aggregate = measure_rate(matrix[n], count, (1u << count)-1);
|
||
expect(aggregate >= solo_sum * 90 / 100, "aggregation >=90% sum of solo rates", aggregate, solo_sum * 90 / 100);
|
||
CHECK(u_get_allocated_count() == baseline);
|
||
}
|
||
struct { const char* name; unsigned count, broken, directions, sparse, ms; enum dummynet_mode mode; } faults[] = {
|
||
{"drop",2,2,3,1,3000,DUMMYNET_DROP}, {"hold",2,2,3,1,3000,DUMMYNET_HOLD},
|
||
{"drop-forward",2,2,1,1,3000,DUMMYNET_DROP}, {"drop-reverse",2,2,2,1,3000,DUMMYNET_DROP},
|
||
{"short-hold",2,2,3,1,30,DUMMYNET_HOLD}, {"slow-drop",2,2,3,1,3000,DUMMYNET_DROP},
|
||
{"two-stuck",3,6,3,1,3000,DUMMYNET_DROP}, {"load-drop",2,2,3,0,5000,DUMMYNET_DROP},
|
||
{"load-hold",2,2,3,0,5000,DUMMYNET_HOLD}
|
||
};
|
||
for (unsigned n = 0; n < sizeof(faults)/sizeof(faults[0]); n++) {
|
||
if (argc > 1 && strcmp(argv[1], faults[n].name)) continue;
|
||
selected++;
|
||
case_name = faults[n].name;
|
||
bidirectional = 0;
|
||
struct profile settings[] = {{20,8000},{20,8000},{20,2000}};
|
||
if (!strcmp(case_name, "slow-drop")) settings[1].rtt_ms = 200;
|
||
if (faults[n].count == 3) { settings[1].rtt_ms = 80; settings[2].rtt_ms = 200; }
|
||
stalled(settings, faults[n].count, faults[n].broken, faults[n].directions, faults[n].mode, faults[n].sparse, faults[n].ms);
|
||
CHECK(u_get_allocated_count() == baseline);
|
||
}
|
||
CHECK(selected && argc <= 2);
|
||
if (csv) CHECK(fclose(csv) == 0);
|
||
DEBUG_INFO(DEBUG_CATEGORY_DEBUG, "Multipath completed: failures=%u", failures);
|
||
test_clock_finish(); socket_platform_cleanup(); return failures ? 1 : 0;
|
||
}
|