Merge pull request #1385 from grondo/issue#1379

flux-wreck cancel: fall back to kill -9 if job is not pending

src/cmd/flux-wreck

@@ -118,15 +118,24 @@ end
 prog:SubCommand {
  name = "cancel",
  description = "Cancel a pending job",
- usage = "JOBID",
+ usage = "[-f|--force] JOBID",
+ options = {
+    { name = "force", char = "f",
+      usage = "Force cancel even if scheduler is not loaded"
+    },
+ },
  handler = function (self, arg)
     local id = check_jobid_arg (self, arg[1])
     local resp, err = f:rpc ("sched.cancel", { jobid = tonumber (id) })
     if not resp then
-	if err == "Function not implemented" then
+        if err == "Function not implemented" and not self.opt.f then
             prog:die ("job cancel not supported when scheduler not loaded")
-	else
-	    prog:die ("Unable to cancel %d: %s\n", id, err)
+        elseif self.opt.f or err == "Invalid argument" then
+            prog:log ("Sending SIGKILL to %d\n", id)
+            local rc, err = f:sendevent ({signal = 9}, "wreck.%d.kill", id)
+            if not rc then self:die ("signal: %s\n", err) end
+        else
+            prog:die ("Unable to cancel %d: %s\n", id, err)
         end
     end
  end

t/t2000-wreck.t

@@ -393,6 +393,20 @@ test_expect_success NO_SCHED 'flux-wreck cancel: fails when sched not loaded' '
 	EOF
 	test_cmp expected.cancel err.cancel
 '
+test_expect_success 'flux-wreck cancel: falls back to SIGKILL with -f' '
+	run_timeout 1 flux wreckrun --detach sleep 100 &&
+	id=$(last_job_id) &&
+	LWJ=$(last_job_path) &&
+	${SHARNESS_TEST_SRCDIR}/scripts/kvs-watch-until.lua -vt 1 $LWJ.state "v == \"running\"" &&
+	flux wreck cancel -f $id &&
+	${SHARNESS_TEST_SRCDIR}/scripts/kvs-watch-until.lua -vt 1 $LWJ.state "v == \"complete\"" &&
+	test_expect_code 137 flux wreck status $id >output.cancel-f &&
+	cat >expected.cancel-f <<-EOF &&
+	Job $id status: complete
+	task0: exited with signal 9
+	EOF
+	test_cmp expected.cancel-f output.cancel-f
+'
 
 check_complete_link() {
     for i in `seq 0 5`; do