[distcc] [patch] keep trying servers instead of falling back locally
Dan Kegel
dank at kegel.com
Sat Aug 13 07:45:02 GMT 2005
Dan Kegel wrote:
> I should mention that the scheme is identical to having the hostnames
> distcc1
> distcc2
> distcc3
> ...
> in the hosts list, except that it discards missing servers
> without forcing any local compilations, and ignores
> servers which are on a different network than distcc1.
Here's a draft of the first part of the change, kindly written by
Tommy Kho of UCLA.
If the first server distcc tries is down,
instead of immediately falling back
to localhost, it tries more servers.
That way, you could use a huge constant host list (say,
distcc1 .. distcc64) even if you only have 16 real servers;
distcc will do a better job ignoring the bogus ones this way.
That's one tiny step towards better autodiscovery of distcc servers, I think,
without radically changing anything.
I haven't tried it myself yet, and I think it needs a limit on
the number of other servers it tries, but I'm putting it out
for comment. Oh, ok, just to see if anybody's reading the list :-)
- Dan
--
Trying to get a job as a c++ developer? See http://kegel.com/academy/getting-hired.html
-------------- next part --------------
A patch to make distcc keep trying servers if the first one can't be
connected to. This may be useful if most of the servers are down,
and there is not yet any cached information about which servers are good.
e.g. if you only have 16 servers distcc1..distcc16,
but the host list has 64 hosts distcc1..distcc64 in it.
TODO: add limit on number of servers tried (old code implicitly
limited it to one server).
Copyright 2005 Google, Inc.
Licensed under the same terms as distcc
T. Kho (and D. Kegel)
diff -ur distcc-2.18.3-orig/src/compile.c distcc-2.18.3/src/compile.c
--- distcc-2.18.3-orig/src/compile.c 2004-10-01 17:47:07.000000000 -0700
+++ distcc-2.18.3/src/compile.c 2005-08-12 15:17:00.858701000 -0700
@@ -40,10 +40,10 @@
#include "bulk.h"
#include "implicit.h"
#include "exec.h"
+#include "compile.h"
#include "where.h"
#include "lock.h"
#include "timeval.h"
-#include "compile.h"
/**
@@ -118,6 +118,7 @@
int cpu_lock_fd;
int ret;
struct dcc_hostdef *host = NULL;
+ struct dcc_conndef conn;
if (sg_level)
goto run_local;
@@ -135,7 +136,7 @@
dcc_note_state(DCC_PHASE_STARTUP, input_fname, NULL);
#endif
- if ((ret = dcc_pick_host_from_list(&host, &cpu_lock_fd)) != 0) {
+ if ((ret = dcc_connect_host_from_list(&host, &conn, &cpu_lock_fd)) != 0) {
/* Doesn't happen at the moment: all failures are masked by
returning localhost. */
goto fallback;
@@ -157,8 +158,8 @@
if ((ret = dcc_compile_remote(argv_stripped,
input_fname,
cpp_fname,
- output_fname,
- cpp_pid, host, status)) != 0) {
+ output_fname, cpp_pid,
+ host, &conn, status)) != 0) {
/* Returns zero if we successfully ran the compiler, even if
* the compiler itself bombed out. */
goto fallback;
diff -ur distcc-2.18.3-orig/src/compile.h distcc-2.18.3/src/compile.h
--- distcc-2.18.3-orig/src/compile.h 2004-07-29 18:12:13.000000000 -0700
+++ distcc-2.18.3/src/compile.h 2005-08-12 14:16:47.554269000 -0700
@@ -20,6 +20,15 @@
* USA
*/
+/**
+ * A simple structure to hold connection info
+ **/
+struct dcc_conndef {
+ int to_net_fd;
+ int from_net_fd;
+ pid_t ssh_pid;
+};
+
/* remote.c */
int dcc_compile_remote(char **argv,
char *input_fname,
@@ -27,8 +36,15 @@
char *output_fname,
pid_t cpp_pid,
struct dcc_hostdef *host,
+ struct dcc_conndef *conn,
int *status);
int dcc_build_somewhere_timed(char *argv[],
int sg_level,
int *status);
+
+int dcc_remote_connect(struct dcc_hostdef *host,
+ int *to_net_fd,
+ int *from_net_fd,
+ pid_t *ssh_pid);
+
diff -ur distcc-2.18.3-orig/src/remote.c distcc-2.18.3/src/remote.c
--- distcc-2.18.3-orig/src/remote.c 2004-10-23 22:05:49.000000000 -0700
+++ distcc-2.18.3/src/remote.c 2005-08-12 15:19:09.009217000 -0700
@@ -63,7 +63,7 @@
* Open a connection using either a TCP socket or SSH. Return input
* and output file descriptors (which may or may not be different.)
**/
-static int dcc_remote_connect(struct dcc_hostdef *host,
+int dcc_remote_connect(struct dcc_hostdef *host,
int *to_net_fd,
int *from_net_fd,
pid_t *ssh_pid)
@@ -175,11 +175,10 @@
char *output_fname,
pid_t cpp_pid,
struct dcc_hostdef *host,
+ struct dcc_conndef *conn,
int *status)
{
- int to_net_fd, from_net_fd;
int ret;
- pid_t ssh_pid = 0;
int ssh_status;
off_t doti_size;
struct timeval before, after;
@@ -188,31 +187,24 @@
rs_log_warning("gettimeofday failed");
dcc_note_execution(host, argv);
- dcc_note_state(DCC_PHASE_CONNECT, input_fname, host->hostname);
-
- /* For ssh support, we need to allow for separate fds writing to and
- * reading from the network, because our connection to the ssh client may
- * be over pipes, which are one-way connections. */
*status = 0;
- if ((ret = dcc_remote_connect(host, &to_net_fd, &from_net_fd, &ssh_pid)))
- goto out;
dcc_note_state(DCC_PHASE_SEND, NULL, NULL);
/* This waits for cpp and puts its status in *status. If cpp failed, then
* the connection will have been dropped and we need not bother trying to
* get any response from the server. */
- ret = dcc_send_header(to_net_fd, argv, host);
+ ret = dcc_send_header(conn->to_net_fd, argv, host);
if ((ret = dcc_wait_for_cpp(cpp_pid, status, input_fname)))
goto out;
- if ((ret = dcc_x_file(to_net_fd, cpp_fname, "DOTI", host->compr, &doti_size)))
+ if ((ret = dcc_x_file(conn->to_net_fd, cpp_fname, "DOTI", host->compr, &doti_size)))
goto out;
rs_trace("client finished sending request to server");
- tcp_cork_sock(to_net_fd, 0);
+ tcp_cork_sock(conn->to_net_fd, 0);
/* but it might not have been read in by the server yet; there's
* 100kB or more of buffers in the two kernels. */
@@ -221,21 +213,21 @@
* start compiling it. */
dcc_note_state(DCC_PHASE_COMPILE, NULL, host->hostname);
- if (to_net_fd != from_net_fd) {
+ if (conn->to_net_fd != conn->from_net_fd) {
/* in ssh mode, we can start closing down early */
- dcc_close(to_net_fd);
+ dcc_close(conn->to_net_fd);
}
/* If cpp failed, just abandon the connection, without trying to
* receive results. */
if (ret == 0 && *status == 0) {
- ret = dcc_retrieve_results(from_net_fd, status, output_fname,
+ ret = dcc_retrieve_results(conn->from_net_fd, status, output_fname,
host);
}
/* Close socket so that the server can terminate, rather than
* making it wait until we've finished our work. */
- dcc_close(from_net_fd);
+ dcc_close(conn->from_net_fd);
if (gettimeofday(&after, NULL)) {
rs_log_warning("gettimeofday failed");
@@ -255,8 +247,8 @@
* background. But it helps make sure that we don't assume we succeeded
* when something possibly went wrong, and it allows us to account for the
* cost of the ssh child. */
- if (ssh_pid) {
- dcc_collect_child("ssh", ssh_pid, &ssh_status); /* ignore failure */
+ if (conn->ssh_pid) {
+ dcc_collect_child("ssh", conn->ssh_pid, &ssh_status); /* ignore failure */
}
return ret;
diff -ur distcc-2.18.3-orig/src/where.c distcc-2.18.3/src/where.c
--- distcc-2.18.3-orig/src/where.c 2004-10-23 22:05:47.000000000 -0700
+++ distcc-2.18.3/src/where.c 2005-08-12 15:28:43.783513000 -0700
@@ -73,6 +73,8 @@
#include "util.h"
#include "hosts.h"
#include "lock.h"
+#include "exec.h"
+#include "compile.h"
#include "where.h"
#include "exitcode.h"
@@ -81,6 +83,57 @@
struct dcc_hostdef **buildhost,
int *cpu_lock_fd);
+int dcc_pick_host_from_list(struct dcc_hostdef **buildhost,
+ int *cpu_lock_fd);
+
+
+/**
+ * Loops through available hosts and establishes a connection
+ **/
+int dcc_connect_host_from_list(struct dcc_hostdef **buildhost,
+ struct dcc_conndef *conn,
+ int *cpu_lock_fd)
+{
+ int ret;
+ int ssh_status;
+
+ /* loop until hostlist is empty, or we have a server to connect to */
+ while (1) {
+ conn->ssh_pid = 0;
+
+ ret = dcc_pick_host_from_list(buildhost, cpu_lock_fd);
+ if (ret != 0) {
+ /* pass through error */
+ return ret;
+ }
+
+ if ((*buildhost)->mode == DCC_MODE_LOCAL) {
+ /* don't need to initiate connection for local builds */
+ return 0;
+ }
+
+ dcc_note_state(DCC_PHASE_CONNECT, NULL, (*buildhost)->hostname);
+
+ /* For ssh support, we need to allow for separate fds writing to and
+ * reading from the network, because our connection to the ssh
+ * client may be over pipes, which are one-way connections. */
+
+ if (!dcc_remote_connect(*buildhost, &conn->to_net_fd, &conn->from_net_fd, &conn->ssh_pid)) {
+ /* successfully connected to a host */
+ return 0;
+ }
+
+ dcc_disliked_host(*buildhost);
+
+ /* cleanup connection, if necessary */
+ if (conn->ssh_pid) {
+ dcc_collect_child("ssh", conn->ssh_pid, &ssh_status); /* ignore failure */
+ }
+ }
+
+ return 0;
+}
+
int dcc_pick_host_from_list(struct dcc_hostdef **buildhost,
int *cpu_lock_fd)
diff -ur distcc-2.18.3-orig/src/where.h distcc-2.18.3/src/where.h
--- distcc-2.18.3-orig/src/where.h 2004-07-29 18:12:14.000000000 -0700
+++ distcc-2.18.3/src/where.h 2005-08-12 13:12:38.493043000 -0700
@@ -21,7 +21,8 @@
*/
/* where.c */
-int dcc_pick_host_from_list(struct dcc_hostdef **,
- int *cpu_lock_fd);
+int dcc_connect_host_from_list(struct dcc_hostdef **,
+ struct dcc_conndef *,
+ int *cpu_lock_fd);
int dcc_lock_local(int *cpu_lock_fd);
More information about the distcc
mailing list