extern int slurm_persist_send_msg(
slurm_persist_conn_t *persist_conn, Buf buffer)
{
uint32_t msg_size, nw_size;
char *msg;
ssize_t msg_wrote;
int rc, retry_cnt = 0;
xassert(persist_conn);
if (persist_conn->fd < 0)
return EAGAIN;
if (!buffer)
return SLURM_ERROR;
rc = slurm_persist_conn_writeable(persist_conn);
if (rc == -1) {
re_open:
if (retry_cnt++ > 3)
return EAGAIN;
/* if errno is ACCESS_DENIED do not try to reopen to
connection just return that */
if (errno == ESLURM_ACCESS_DENIED)
return ESLURM_ACCESS_DENIED;
if (persist_conn->flags & PERSIST_FLAG_RECONNECT) {
slurm_persist_conn_reopen(persist_conn, true);
rc = slurm_persist_conn_writeable(persist_conn);
} else
return SLURM_ERROR;
}
if (rc < 1)
return EAGAIN;
msg_size = get_buf_offset(buffer);
nw_size = htonl(msg_size);
msg_wrote = write(persist_conn->fd, &nw_size, sizeof(nw_size));
if (msg_wrote != sizeof(nw_size))
return EAGAIN;
msg = get_buf_data(buffer);
while (msg_size > 0) {
rc = slurm_persist_conn_writeable(persist_conn);
if (rc == -1)
goto re_open;
if (rc < 1)
return EAGAIN;
msg_wrote = write(persist_conn->fd, msg, msg_size);
if (msg_wrote <= 0)
return EAGAIN;
msg += msg_wrote;
msg_size -= msg_wrote;
}
return SLURM_SUCCESS;
}
/* run_dbd_backup - this is the backup controller, it should run in standby
* mode, assuming control when the primary controller stops responding */
extern void run_dbd_backup(void)
{
slurm_persist_conn_t slurmdbd_conn;
primary_resumed = false;
memset(&slurmdbd_conn, 0, sizeof(slurm_persist_conn_t));
slurmdbd_conn.rem_host = slurmdbd_conf->dbd_addr;
slurmdbd_conn.rem_port = slurmdbd_conf->dbd_port;
slurmdbd_conn.cluster_name = "backup_slurmdbd";
slurmdbd_conn.fd = -1;
slurmdbd_conn.shutdown = &shutdown_time;
slurm_persist_conn_open_without_init(&slurmdbd_conn);
/* repeatedly ping Primary */
while (!shutdown_time) {
int writeable = slurm_persist_conn_writeable(&slurmdbd_conn);
//info("%d %d", have_control, writeable);
if (have_control && writeable == 1) {
info("Primary has come back");
primary_resumed = true;
shutdown_threads();
have_control = false;
break;
} else if (!have_control && writeable <= 0) {
have_control = true;
info("Taking Control");
break;
}
sleep(1);
if (writeable <= 0)
slurm_persist_conn_reopen(&slurmdbd_conn, false);
}
slurm_persist_conn_close(&slurmdbd_conn);
return;
}
extern Buf slurm_persist_recv_msg(slurm_persist_conn_t *persist_conn)
{
uint32_t msg_size, nw_size;
char *msg;
ssize_t msg_read, offset;
Buf buffer;
xassert(persist_conn);
if (persist_conn->fd < 0)
return NULL;
if (!_conn_readable(persist_conn))
goto endit;
msg_read = read(persist_conn->fd, &nw_size, sizeof(nw_size));
if (msg_read != sizeof(nw_size))
goto endit;
msg_size = ntohl(nw_size);
/* We don't error check for an upper limit here
* since size could possibly be massive */
if (msg_size < 2) {
error("Persistent Conn: Invalid msg_size (%u)", msg_size);
goto endit;
}
msg = xmalloc(msg_size);
offset = 0;
while (msg_size > offset) {
if (!_conn_readable(persist_conn))
break; /* problem with this socket */
msg_read = read(persist_conn->fd, (msg + offset),
(msg_size - offset));
if (msg_read <= 0) {
error("Persistent Conn: read: %m");
break;
}
offset += msg_read;
}
if (msg_size != offset) {
if (!(*persist_conn->shutdown)) {
error("Persistent Conn: only read %zd of %d bytes",
offset, msg_size);
} /* else in shutdown mode */
xfree(msg);
goto endit;
}
buffer = create_buf(msg, msg_size);
return buffer;
endit:
/* Close it since we abandoned it. If the connection does still exist
* on the other end we can't rely on it after this point since we didn't
* listen long enough for this response.
*/
if (!(*persist_conn->shutdown) &&
persist_conn->flags & PERSIST_FLAG_RECONNECT)
slurm_persist_conn_reopen(persist_conn, true);
return NULL;
}
