diff -ruN pbs-2.3.11-stock/README.mpiexec pbs-2.3.11/README.mpiexec
--- pbs-2.3.11-stock/README.mpiexec Wed Dec 31 19:00:00 1969
+++ pbs-2.3.11/README.mpiexec Thu Jan 25 10:58:53 2001
@@ -0,0 +1,20 @@
+
+Documentation of changes applied as part of the mpiexec patch.
+http://www.osc.edu/~pw/mpiexec.html
+
+
+1. General bug fixes:
+
+src/lib/Libifl/rpp.c
+Incrcease rpp retry count for possibly congested networks.
+
+src/lib/Libifl/tm.c
+Do not call DIS_tcp_setup(-1) if connection to the local mom failed
+after 5 tries.
+
+2. Functionality additions:
+
+src/resmom/start_exec.c
+Added extension to get stdin/out/err files from the environment rather than
+always using /dev/null during tm_spawn.
+
diff -ruN pbs-2.3.11-stock/src/lib/Libifl/rpp.c pbs-2.3.11/src/lib/Libifl/rpp.c
--- pbs-2.3.11-stock/src/lib/Libifl/rpp.c Fri Sep 29 17:29:46 2000
+++ pbs-2.3.11/src/lib/Libifl/rpp.c Thu Dec 28 18:35:02 2000
@@ -186,7 +186,7 @@
/*
** Default number of sendto attempts on a *packet.
*/
-#define RPP_RETRY 10
+#define RPP_RETRY 30
/*
** Max allowed number of outstanding pkts
*/
diff -ruN pbs-2.3.11-stock/src/lib/Libifl/tm.c pbs-2.3.11/src/lib/Libifl/tm.c
--- pbs-2.3.11-stock/src/lib/Libifl/tm.c Thu Jan 25 10:10:54 2001
+++ pbs-2.3.11/src/lib/Libifl/tm.c Thu Jan 25 10:54:47 2001
@@ -451,7 +451,8 @@
}
}
- DIS_tcp_setup(local_conn);
+ if (local_conn >= 0)
+ DIS_tcp_setup(local_conn);
return (local_conn);
}
diff -ruN pbs-2.3.11-stock/src/resmom/start_exec.c pbs-2.3.11/src/resmom/start_exec.c
--- pbs-2.3.11-stock/src/resmom/start_exec.c Thu Jan 25 10:10:55 2001
+++ pbs-2.3.11/src/resmom/start_exec.c Thu Jan 25 10:01:54 2001
@@ -1389,6 +1389,42 @@
}
/*
+ * Look for a certain environment variable which has a port# which should
+ * be opened on the MS to establish communication for one of the 3 stdio
+ * streams. >=0 return is that valid fd, -1 means no env var found,
+ * -2 means malformed env value or failure to connect.
+ */
+static int
+search_env_and_open(const char *envname, u_long ipaddr)
+{
+ static char *id = "search_env_and_open";
+ int i, len = strlen(envname);
+
+ for (i=0; i<vtable.v_used; i++)
+ if (!strncmp(vtable.v_envp[i], envname, len)) {
+ const char *cp = vtable.v_envp[i] + len;
+ char *cq;
+ int fd, port;
+ if (*cp++ != '=') break; /* empty, ignore it */
+ port = strtol(cp, &cq, 10);
+ if (*cq) {
+ log_err(errno, id, "improper value for MPIEXEC_STD*_PORT");
+ return -2;
+ }
+#if 0 /* debugging */
+ log_err(0, "search_env_and_open attempting open", vtable.v_envp[i]);
+#endif
+ if ((fd = open_demux(ipaddr, port)) < 0) {
+ log_err(errno, id, "failed connect to mpiexec process on MS");
+ log_err(errno, id, vtable.v_envp[i]);
+ return -2;
+ }
+ return fd;
+ }
+ return -1; /* not found */
+}
+
+/*
** Start a process for a spawn request. This will be different from
** a job's initial shell task in that the environment will be specified
** and no interactive code need be included.
@@ -1407,7 +1443,7 @@
int pipes[2], kid_read, kid_write, parent_read, parent_write;
int pts;
int i, j;
- int fd;
+ int fd0, fd1, fd2;
u_long ipaddr;
struct array_strings *vstrs;
struct startjob_rtn sjr;
@@ -1631,30 +1667,40 @@
/*
** Set up stdin.
*/
- if ((fd = open("/dev/null", O_RDONLY)) == -1) {
+ /* look through env for a port# on MS we should use for stdin */
+ if ((fd0 = search_env_and_open("MPIEXEC_STDIN_PORT", ipaddr)) == -2)
+ starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
+ /* use /dev/null if no env var found */
+ if (fd0 < 0 && (fd0 = open("/dev/null", O_RDONLY)) == -1) {
log_err(errno, "newtask", "could not open devnull");
(void)close(0);
}
else {
- (void)dup2(fd, 0);
- if (fd > 0)
- (void)close(fd);
+ (void)dup2(fd0, 0);
+ if (fd0 > 0)
+ (void)close(fd0);
}
+ /* look through env for a port# on MS we should use for stdout/err */
+ if ((fd1 = search_env_and_open("MPIEXEC_STDOUT_PORT", ipaddr)) == -2)
+ starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
+ if ((fd2 = search_env_and_open("MPIEXEC_STDERR_PORT", ipaddr)) == -2)
+ starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
+
if (pjob->ji_numnodes > 1) {
/*
** Open sockets to demux proc for stdout and stderr.
*/
- if ((fd = open_demux(ipaddr, pjob->ji_stdout)) == -1)
+ if (fd1 < 0 && (fd1 = open_demux(ipaddr,pjob->ji_stdout)) == -1)
starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
- (void)dup2(fd, 1);
- if (fd > 1)
- (void)close(fd);
- if ((fd = open_demux(ipaddr, pjob->ji_stderr)) == -1)
+ (void)dup2(fd1, 1);
+ if (fd1 > 1)
+ (void)close(fd1);
+ if (fd2 < 0 && (fd2 = open_demux(ipaddr,pjob->ji_stderr)) == -1)
starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
- (void)dup2(fd, 2);
- if (fd > 2)
- (void)close(fd);
+ (void)dup2(fd2, 2);
+ if (fd2 > 2)
+ (void)close(fd2);
(void)write(1,pjob->ji_wattr[(int)JOB_ATR_Cookie].at_val.at_str,
strlen(pjob->ji_wattr[(int)JOB_ATR_Cookie].at_val.at_str));
@@ -1664,17 +1710,41 @@
(pjob->ji_wattr[(int)JOB_ATR_interactive].at_val.at_long > 0)) {
/* interactive job, single node, write to pty */
- if ((pts = open_pty(pjob)) < 0) {
+ pts = -1;
+ if (fd1 < 0 || fd2 < 0) {
+ if ((pts = open_pty(pjob)) < 0) {
log_err(errno, id,"cannot open slave");
starter_return(kid_write, kid_read,JOB_EXEC_FAIL1,&sjr);
+ }
+ if (fd1 < 0)
+ fd1 = pts;
+ if (fd2 < 0)
+ fd2 = pts;
}
- (void)dup2(pts, 1);
- (void)dup2(pts, 2);
-
+ (void)dup2(fd1, 1);
+ (void)dup2(fd2, 2);
+ if (fd1 != pts)
+ (void) close(fd1);
+ if (fd2 != pts)
+ (void) close(fd2);
} else {
/* normal batch job, single node, write straight to files */
- if (open_std_out_err(pjob) == -1) {
+ pts = -1;
+ if (fd1 < 0 || fd2 < 0) {
+ if (open_std_out_err(pjob) == -1)
starter_return(kid_write, kid_read,JOB_EXEC_FAIL1,&sjr);
+ }
+ if (fd1 >= 0) {
+ (void) close(1);
+ (void) dup2(fd1, 1);
+ if (fd1 > 1)
+ (void) close(fd1);
+ }
+ if (fd2 >= 0) {
+ (void) close(2);
+ (void) dup2(fd2, 2);
+ if (fd2 > 2)
+ (void) close(fd2);
}
}
syntax highlighted by Code2HTML, v. 0.9.1