Merge pull request #144 from jerrymannil/fail-fast

fail fast on connect error or non-zero return code
chaos · Jun 14, 2022 · 3bc95a8 · 3bc95a8
2 parents 139ef36 + c94c3ca
commit 3bc95a8
Show file tree

Hide file tree

Showing 4 changed files with 27 additions and 5 deletions.
diff --git a/doc/pdsh.1.in b/doc/pdsh.1.in
@@ -144,10 +144,13 @@ as if they had been given to \fB\-w\fR and preceeded with the minus `-'
 character.
 
 .SH "Standard pdsh options"
-.TP 
+.TP
 .I "-S"
 Return the largest of the remote command return values.
 .TP
+.I "-k"
+Fail fast on connect failure or non-zero return code.
+.TP
 .I "-h"
 Output usage menu and quit. A list of available rcmd modules
 will also be printed at the end of the usage message.

diff --git a/src/pdsh/dsh.c b/src/pdsh/dsh.c
@@ -753,7 +753,7 @@ static void *_rsh_thread(void *args)
         a->rc = rv;
 
     /* if a single qshell thread fails, terminate whole job */
-    if (a->kill_on_fail && a->state == DSH_FAILED) {
+    if (a->kill_on_fail && ((a->state == DSH_FAILED) || (a->rc > 0))) {
         _fwd_signal(SIGTERM);
         errx("%p: terminating all processes\n");
     }

diff --git a/src/pdsh/opt.c b/src/pdsh/opt.c
@@ -66,7 +66,8 @@
 
 #define OPT_USAGE_DSH "\
 Usage: pdsh [-options] command ...\n\
--S                return largest of remote command return values\n"
+-S                return largest of remote command return values\n\
+-k                fail fast on connect failure or non-zero return code\n"
 
 /* -s option only useful on AIX */
 #if	HAVE_MAGIC_RSHELL_CLEANUP
@@ -113,9 +114,9 @@ Usage: rpdcp [-options] src [src2...] dir\n\
 /* undocumented "-K" option -  keep domain name in output */
 
 #if	HAVE_MAGIC_RSHELL_CLEANUP
-#define DSH_ARGS	"sS"
+#define DSH_ARGS	"sSk"
 #else
-#define DSH_ARGS    "S"
+#define DSH_ARGS    "Sk"
 #endif
 #define PCP_ARGS	"pryzZe:"
 #define GEN_ARGS	"hLNKR:M:t:cqf:w:x:l:u:bI:dVT:Q"
@@ -691,6 +692,9 @@ void opt_args(opt_t * opt, int argc, char *argv[])
             else
                 goto test_module_option;
             break;
+        case 'k':
+            opt->kill_on_fail = true;
+            break;
         default: test_module_option:
             if (mod_process_opt(opt, c, optarg) < 0)
                _usage(opt);

diff --git a/tests/t2001-ssh.sh b/tests/t2001-ssh.sh
@@ -197,5 +197,20 @@ test_expect_success 'ssh works with pdsh -S and multiple targets' '
 	done
 '
 unset PDSH_SSH_ARGS
+test_expect_success 'ssh works with pdsh -k' '
+	TEST_EXIT_CODE=$(random 254) &&
+	export PDSH_SSH_ARGS="-n%n -i0 -e$TEST_EXIT_CODE"
+	test_expect_code "1" pdsh -Rssh -k -w foo0 command
+'
+unset PDSH_SSH_ARGS
+test_expect_success 'ssh works with pdsh -k and multiple targets' '
+	for n in $(seq 1 24); do
+		TEST_EXIT_CODE=$(random 254) &&
+		FAILING_RANK=$(random $n) &&
+		export PDSH_SSH_ARGS="-n%n -i$FAILING_RANK -e$TEST_EXIT_CODE"
+		test_expect_code "1" pdsh -Rssh -k -wfoo[0-$n] command
+	done
+'
+unset PDSH_SSH_ARGS
 
 test_done