Commit 92d02148 authored by Wayne Davison's avatar Wayne Davison

Improve --timeout method to take into account all I/O that is going on.

The receiving side also switches timeout handling from the receiver to
the generator, which obviates the need for the sender to send any
keep-alive messages at all (for protocol 31 and beyond).  Given this
setup, all keep-alive messages are now sent as empty MSG_DATA messages,
with MSG_NOOP messages only being understood and (when necessary) acted
upon to forward a keep-alive event to an older receiver.  This is both
safer and more compatible with older versions.
parent 82b2a31a
......@@ -71,7 +71,6 @@ extern int io_error;
extern int flist_eof;
extern int allowed_lull;
extern int sock_f_out;
extern int ignore_timeout;
extern int protocol_version;
extern int file_total;
extern int fuzzy_basis;
......@@ -290,7 +289,7 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
rprintf(FINFO, "delete_in_dir(%s)\n", fbuf);
if (allowed_lull)
maybe_send_keepalive();
maybe_send_keepalive(time(NULL), True);
if (io_error && !ignore_errors) {
if (already_warned)
......@@ -1929,7 +1928,7 @@ static void touch_up_dirs(struct file_list *flist, int ndx)
}
if (counter >= loopchk_limit) {
if (allowed_lull)
maybe_send_keepalive();
maybe_send_keepalive(time(NULL), True);
else
maybe_flush_socket(0);
counter = 0;
......@@ -2076,12 +2075,6 @@ void generate_files(int f_out, const char *local_name)
: "enabled");
}
/* Since we often fill up the outgoing socket and then just sit around
* waiting for the other 2 processes to do their thing, we don't want
* to exit on a timeout. If the data stops flowing, the receiver will
* notice that and let us know via the redo pipe (or its closing). */
ignore_timeout = 1;
dflt_perms = (ACCESSPERMS & ~orig_umask);
do {
......@@ -2135,7 +2128,7 @@ void generate_files(int f_out, const char *local_name)
if (i + cur_flist->ndx_start >= next_loopchk) {
if (allowed_lull)
maybe_send_keepalive();
maybe_send_keepalive(time(NULL), True);
else
maybe_flush_socket(0);
next_loopchk += loopchk_limit;
......
......@@ -39,6 +39,7 @@ extern size_t bwlimit_writemax;
extern int io_timeout;
extern int am_server;
extern int am_sender;
extern int am_receiver;
extern int am_generator;
extern int msgs2stderr;
extern int inc_recurse;
......@@ -64,11 +65,11 @@ extern iconv_t ic_send, ic_recv;
int csum_length = SHORT_SUM_LENGTH; /* initial value */
int allowed_lull = 0;
int ignore_timeout = 0;
int batch_fd = -1;
int msgdone_cnt = 0;
int forward_flist_data = 0;
BOOL flist_receiving_enabled = False;
BOOL we_send_keepalive_messages = False;
/* Ignore an EOF error if non-zero. See whine_about_eof(). */
int kluge_around_eof = 0;
......@@ -152,25 +153,35 @@ static void read_a_msg(void);
static void drain_multiplex_messages(void);
static void sleep_for_bwlimit(int bytes_written);
static void check_timeout(void)
static void check_timeout(BOOL allow_keepalive)
{
time_t t;
time_t t, chk;
if (!io_timeout || ignore_timeout)
/* On the receiving side, the generator is now handling timeouts, so
* the receiver ignores them. Note that the am_receiver flag is not
* set until the receiver forks from the generator, so timeouts will be
* based on receiving data on the receiving side until that event. */
if (!io_timeout || am_receiver)
return;
if (!last_io_in) {
last_io_in = time(NULL);
return;
t = time(NULL);
if (allow_keepalive && we_send_keepalive_messages) {
/* This may put data into iobuf.msg w/o flushing. */
maybe_send_keepalive(t, False);
}
t = time(NULL);
if (!last_io_in)
last_io_in = t;
if (!last_io_out)
last_io_out = t;
if (t - last_io_in >= io_timeout) {
chk = MAX(last_io_out, last_io_in);
if (t - chk >= io_timeout) {
if (am_server)
msgs2stderr = 1;
rprintf(FERROR, "[%s] io timeout after %d seconds -- exiting\n",
who_am_i(), (int)(t-last_io_in));
who_am_i(), (int)(t-chk));
exit_cleanup(RERR_TIMEOUT);
}
}
......@@ -252,7 +263,8 @@ static size_t safe_read(int fd, char *buf, size_t len)
who_am_i());
exit_cleanup(RERR_FILEIO);
}
check_timeout();
if (we_send_keepalive_messages)
maybe_send_keepalive(time(NULL), True);
continue;
}
......@@ -336,7 +348,8 @@ static void safe_write(int fd, const char *buf, size_t len)
what_fd_is(fd), who_am_i());
exit_cleanup(RERR_FILEIO);
}
check_timeout();
if (we_send_keepalive_messages)
maybe_send_keepalive(time(NULL), True);
continue;
}
......@@ -733,7 +746,7 @@ static char *perform_io(size_t needed, int flags)
send_extra_file_list(sock_f_out, -1);
extra_flist_sending_enabled = !flist_eof;
} else
check_timeout();
check_timeout((flags & PIO_NEED_INPUT) != 0);
FD_ZERO(&r_fds); /* Just in case... */
FD_ZERO(&w_fds);
}
......@@ -1315,18 +1328,21 @@ void maybe_flush_socket(int important)
io_flush(NORMAL_FLUSH);
}
/* This never adds new non-msg-buffer data, since we don't know the state
* of the raw-data buffer. */
void maybe_send_keepalive(void)
{
if (time(NULL) - last_io_out >= allowed_lull) {
if (!iobuf.msg.len && iobuf.out.len == iobuf.out_empty_len) {
if (protocol_version >= 30)
send_msg(MSG_NOOP, "", 0, 0);
else
send_msg(MSG_DATA, "", 0, 0);
}
if (iobuf.msg.len)
/* Older rsync versions used to send either a MSG_NOOP (protocol 30) or a
* raw-data-based keep-alive (protocol 29), both of which implied forwarding of
* the message through the sender. Since the new timeout method does not need
* any forwarding, we just send an empty MSG_DATA message, which works with all
* rsync versions. This avoids any message forwarding, and leaves the raw-data
* stream alone (since we can never be quite sure if that stream is in the
* right state for a keep-alive message). */
void maybe_send_keepalive(time_t now, BOOL allow_flush)
{
if (now - last_io_out >= allowed_lull) {
if (!iobuf.msg.len && iobuf.out.len == iobuf.out_empty_len)
send_msg(MSG_DATA, "", 0, 0);
if (!allow_flush) {
/* Let the caller worry about writing out the data. */
} else if (iobuf.msg.len)
perform_io(iobuf.msg.size - iobuf.msg.len + 1, PIO_NEED_MSGROOM);
else if (iobuf.out.len > iobuf.out_empty_len)
io_flush(NORMAL_FLUSH);
......@@ -1411,11 +1427,12 @@ static void read_a_msg(void)
}
break;
case MSG_NOOP:
/* Support protocol-30 keep-alive method. */
if (msg_bytes != 0)
goto invalid_msg;
iobuf.in_multiplexed = 1;
if (am_sender)
maybe_send_keepalive();
maybe_send_keepalive(time(NULL), True);
break;
case MSG_DELETED:
if (msg_bytes >= sizeof data)
......
......@@ -73,6 +73,7 @@ extern int send_msgs_to_gen;
extern pid_t cleanup_child_pid;
extern size_t bwlimit_writemax;
extern unsigned int module_dirlen;
extern BOOL we_send_keepalive_messages;
extern BOOL flist_receiving_enabled;
extern BOOL shutting_down;
extern struct stats stats;
......@@ -764,6 +765,8 @@ static void do_server_sender(int f_in, int f_out, int argc, char *argv[])
struct file_list *flist;
char *dir = argv[0];
we_send_keepalive_messages = io_timeout != 0 && protocol_version < 31;
if (DEBUG_GTE(SEND, 1)) {
rprintf(FINFO, "server_sender starting pid=%ld\n",
(long)getpid());
......@@ -904,6 +907,7 @@ static int do_recv(int f_in, int f_out, char *local_name)
}
am_generator = 1;
we_send_keepalive_messages = io_timeout != 0;
flist_receiving_enabled = True;
io_end_multiplex_in(MPLX_SWITCHING);
......@@ -1108,6 +1112,7 @@ int client_run(int f_in, int f_out, pid_t pid, int argc, char *argv[])
if (am_sender) {
keep_dirlinks = 0; /* Must be disabled on the sender. */
we_send_keepalive_messages = io_timeout != 0 && protocol_version < 31;
if (always_checksum
&& (log_format_has(stdout_format, 'C')
......
......@@ -365,11 +365,10 @@ int read_ndx_and_attrs(int f_in, int f_out, int *iflag_ptr, uchar *type_ptr,
iflags = protocol_version >= 29 ? read_shortint(f_in)
: ITEM_TRANSFER | ITEM_MISSING_DATA;
/* Honor the old-style keep-alive indicator. */
if (protocol_version < 30
&& ndx == cur_flist->used && iflags == ITEM_IS_NEW) {
/* Support the protocol-29 keep-alive style. */
if (protocol_version < 30 && ndx == cur_flist->used && iflags == ITEM_IS_NEW) {
if (am_sender)
maybe_send_keepalive();
maybe_send_keepalive(time(NULL), True);
goto read_loop;
}
......
......@@ -234,7 +234,7 @@ enum msgcode {
MSG_STATS=10, /* message has stats data for generator */
MSG_IO_ERROR=22,/* the sending side had an I/O error */
MSG_IO_TIMEOUT=33,/* tell client about a daemon's timeout value */
MSG_NOOP=42, /* a do-nothing message */
MSG_NOOP=42, /* a do-nothing message (legacy protocol-30 only) */
MSG_ERROR_EXIT=86, /* used by siblings and by protocol-31 */
MSG_SUCCESS=100,/* successfully updated indicated flist index */
MSG_DELETED=101,/* successfully deleted a file on receiving side */
......
......@@ -43,6 +43,7 @@ extern int inplace;
extern int batch_fd;
extern int write_batch;
extern int file_old_total;
extern BOOL we_send_keepalive_messages;
extern struct stats stats;
extern struct file_list *cur_flist, *first_flist, *dir_flist;
......@@ -104,8 +105,8 @@ static struct sum_struct *receive_sums(int f)
s->sums[i].len = s->blength;
offset += s->sums[i].len;
if (allowed_lull && !(i % lull_mod))
maybe_send_keepalive();
if (we_send_keepalive_messages && !(i % lull_mod))
maybe_send_keepalive(time(NULL), True);
if (DEBUG_GTE(DELTASUM, 3)) {
rprintf(FINFO,
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment