perf: accept thread + dial backoff; join/peers.json (0.9.23)

Run inbound peer handling on its own thread so outbound dials cannot starve
:9100. Failed peers cool down exponentially. Also land --join (member seat)
and realm/peers.json seed discovery for Tailscale meshes.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
WHC
2026-10-29 18:16:46 +08:00
co-authored by Cursor
parent de2eaf87f7
commit c329db504d
7 changed files with 273 additions and 37 deletions
+147 -32
View File
@@ -20,8 +20,10 @@
#include <windows.h>
#else
#include <fcntl.h>
#include <pthread.h>
#include <signal.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <unistd.h>
#endif
@@ -72,7 +74,6 @@ static int read_meta(daemon_meta_t *m) {
} else if (!strncmp(line, "port=", 5))
m->port = atoi(line + 5);
else if (m->pid <= 0) {
/* Legacy single-line pidfile. */
int pid = atoi(line);
if (pid > 0)
m->pid = pid;
@@ -165,6 +166,20 @@ static void sleep_ms(int ms) {
#endif
}
static int64_t now_ms(void) {
#ifdef _WIN32
return (int64_t)GetTickCount64();
#else
struct timeval tv;
gettimeofday(&tv, NULL);
return (int64_t)tv.tv_sec * 1000 + (int64_t)tv.tv_usec / 1000;
#endif
}
#ifndef _WIN32
static pthread_mutex_t g_sync_mu = PTHREAD_MUTEX_INITIALIZER;
#endif
static int load_local_bundle(cst_peer_sync_bundle_t *b) {
memset(b, 0, sizeof(*b));
cst_parliament_meta_t m;
@@ -186,16 +201,27 @@ static int apply_if_newer(cst_peer_sync_bundle_t *b, unsigned long before) {
b->chair_instance, b->context);
}
/* Short sync budget: accept thread must not stall the listen backlog. */
static int sync_on_session(cst_peer_session_t *sess) {
cst_peer_sync_bundle_t local;
if (load_local_bundle(&local) != 0)
#ifndef _WIN32
pthread_mutex_lock(&g_sync_mu);
#endif
if (load_local_bundle(&local) != 0) {
#ifndef _WIN32
pthread_mutex_unlock(&g_sync_mu);
#endif
return -1;
}
unsigned long before = local.revision;
char err[128];
int rc = cst_peer_sync(sess, &local, 15000, err, sizeof(err));
int rc = cst_peer_sync(sess, &local, 3000, err, sizeof(err));
if (rc == 0)
apply_if_newer(&local, before);
free(local.context);
#ifndef _WIN32
pthread_mutex_unlock(&g_sync_mu);
#endif
return rc;
}
@@ -208,6 +234,72 @@ static int unlock_realm(const char *password, cst_realm_keys_t *keys) {
return cst_realm_load(dir, keys);
}
typedef struct {
os_socket_t ls;
cst_realm_cert_t cert;
uint8_t sk[64];
uint8_t realm_pk[32];
} accept_ctx_t;
#ifndef _WIN32
static void *accept_main(void *arg) {
accept_ctx_t *ac = arg;
while (!g_stop) {
cst_peer_session_t sess;
char herr[128];
if (cst_peer_accept((intptr_t)ac->ls, &ac->cert, ac->sk, ac->realm_pk, 400,
&sess, herr, sizeof(herr)) != 0)
continue;
sync_on_session(&sess);
cst_peer_close(&sess);
}
return NULL;
}
#endif
/* Per-peer dial cooldown so stuck remotes do not dominate the dial loop. */
typedef struct {
char iid[64];
int fails;
int64_t cool_until_ms;
} dial_cool_t;
static dial_cool_t *cool_find(dial_cool_t *tab, int *n, int cap,
const char *iid) {
for (int i = 0; i < *n; i++) {
if (!strcmp(tab[i].iid, iid))
return &tab[i];
}
if (*n >= cap)
return NULL;
dial_cool_t *c = &tab[(*n)++];
memset(c, 0, sizeof(*c));
snprintf(c->iid, sizeof(c->iid), "%s", iid);
return c;
}
static int cool_blocked(dial_cool_t *c, int64_t now) {
return c && c->cool_until_ms > now;
}
static void cool_fail(dial_cool_t *c, int64_t now) {
if (!c)
return;
c->fails = c->fails < 8 ? c->fails + 1 : 8;
/* 4s, 8s, 16s, … capped at 120s */
int64_t wait = 4000LL << (c->fails - 1);
if (wait > 120000)
wait = 120000;
c->cool_until_ms = now + wait;
}
static void cool_ok(dial_cool_t *c) {
if (!c)
return;
c->fails = 0;
c->cool_until_ms = 0;
}
int cst_parliament_daemon_run(const char *password, const char *instance_id,
int peer_port, char *err, size_t err_sz) {
if (err && err_sz)
@@ -272,65 +364,93 @@ int cst_parliament_daemon_run(const char *password, const char *instance_id,
signal(SIGHUP, on_sig);
#endif
g_stop = 0;
int tick = 0, idle = 0, dial_rr = 0;
accept_ctx_t ac;
memset(&ac, 0, sizeof(ac));
ac.ls = ls;
ac.cert = cert;
memcpy(ac.sk, sk, 64);
memcpy(ac.realm_pk, realm.pk, 32);
#ifndef _WIN32
pthread_t ath;
int have_ath = pthread_create(&ath, NULL, accept_main, &ac) == 0;
#else
int have_ath = 0;
#endif
int tick = 0, dial_rr = 0;
dial_cool_t cool[24];
int ncool = 0;
while (!g_stop) {
cst_peer_session_t sess;
char herr[128];
/* Prefer accept: a blocking outbound dial used to starve inbound peers
* (CLOSE-WAIT + unread hello ~385B on :9100). Drain the backlog first. */
int accepted = 0;
for (;;) {
int accept_ms = accepted > 0 ? 50 : (idle > 20 ? 800 : idle > 5 ? 400 : 200);
if (cst_peer_accept((intptr_t)ls, &cert, sk, realm.pk, accept_ms, &sess,
herr, sizeof(herr)) != 0)
#ifdef _WIN32
/* No accept thread: keep prefer-accept single-threaded path. */
for (int a = 0; a < 4; a++) {
if (cst_peer_accept((intptr_t)ls, &cert, sk, realm.pk, 200, &sess, herr,
sizeof(herr)) != 0)
break;
sync_on_session(&sess);
cst_peer_close(&sess);
accepted++;
idle = 0;
if (accepted >= 8)
break;
}
if (accepted == 0)
idle++;
#endif
int push = cst_parliament_sync_pending();
tick++;
/* At most one outbound dial per loop turn, short wall budget, so accept
* keeps pace. Lexicographic rule avoids A↔B mutual-dial deadlock unless
* we have a local push to deliver. */
int period = idle > 20 ? 30 : idle > 5 ? 12 : 6;
int period = push ? 1 : 8;
if ((push || tick % period == 0) && disc) {
cst_peer_info_t peers[16];
int n = cst_discover_peers(disc, peers, 16);
int64_t now = now_ms();
int tried = 0;
for (int k = 0; k < n; k++) {
int i = (dial_rr + k) % n;
if (peers[i].port <= 0)
continue;
if (!strcmp(peers[i].instance_id, instance_id))
continue; /* never dial ourselves */
continue;
if (!push && strcmp(instance_id, peers[i].instance_id) > 0)
continue; /* lower id initiates idle polls */
continue;
dial_cool_t *c = cool_find(cool, &ncool, 24, peers[i].instance_id);
if (!push && cool_blocked(c, now))
continue;
dial_rr = (i + 1) % n;
if (cst_peer_dial(peers[i].host, peers[i].port, &cert, sk, realm.pk,
2500, &sess, herr, sizeof(herr)) == 0) {
/* Dial budget stays short; accept thread covers inbound meanwhile. */
int ok = cst_peer_dial(peers[i].host, peers[i].port, &cert, sk,
realm.pk, 2000, &sess, herr, sizeof(herr)) == 0;
if (ok) {
sync_on_session(&sess);
cst_peer_close(&sess);
idle = 0;
cool_ok(c);
} else {
cool_fail(c, now);
}
tried = 1;
break; /* one dial then go back to accept */
break;
}
if (push && (tried || n == 0))
cst_parliament_sync_ack();
}
/* Pace dial loop; accept runs concurrently on Unix. */
sleep_ms(have_ath ? 150 : 50);
}
#ifndef _WIN32
if (have_ath) {
/* Wake accept's poll by shutting down listen sock after stop. */
os_net_shutdown(ls);
pthread_join(ath, NULL);
}
#endif
if (disc)
cst_discover_stop(disc);
os_net_close(ls);
clear_pid();
crypto_wipe(sk, sizeof(sk));
crypto_wipe(ac.sk, sizeof(ac.sk));
crypto_wipe(realm.seed, sizeof(realm.seed));
crypto_wipe(realm.sk, sizeof(realm.sk));
return 0;
@@ -380,13 +500,10 @@ int cst_parliament_daemon_start(const char *password, const char *instance_id,
(cur_port <= 0 || cur_port == want_port);
if (same)
return 0;
/* Different seat or port: take over. */
(void)cst_parliament_daemon_stop();
}
}
#ifdef _WIN32
/* Windows: run in-process foreground loop under a detached console is
* deferred; callers should use --service in a service manager. */
if (err && err_sz)
snprintf(err, err_sz, "background daemon requires --service on Windows");
return -1;
@@ -403,7 +520,6 @@ int cst_parliament_daemon_start(const char *password, const char *instance_id,
return -1;
}
if (pid > 0) {
/* Parent: wait briefly for matching pidfile. */
for (int i = 0; i < 50; i++) {
int cur_pid = 0, cur_port = 0;
char cur_id[64] = {0};
@@ -417,7 +533,6 @@ int cst_parliament_daemon_start(const char *password, const char *instance_id,
snprintf(err, err_sz, "service did not start");
return -1;
}
/* Child: detach fully so SSH/nohup callers are not held by open stdio. */
setsid();
int fd = open("/dev/null", O_RDWR);
if (fd >= 0) {