[oe-commits] Richard Purdie : qemurunner: Ensure runqemu doesn't survive SIGKILL

[git at git.openembedded.org]

Module: openembedded-core.git
Branch: master
Commit: 99428eafb5352bd39bc4329bdba07c6d6f17b03f
URL:    http://git.openembedded.org/?p=openembedded-core.git&a=commit;h=99428eafb5352bd39bc4329bdba07c6d6f17b03f

Author: Richard Purdie <richard.purdie at linuxfoundation.org>
Date:   Tue Sep  8 23:35:04 2015 +0100

qemurunner: Ensure runqemu doesn't survive SIGKILL

Currently, we see runqemu and qemu-system-* processes left behind when
bitbake is killed by buildbot. This is due to the use of setpgrp() in
the runqemu subprocess call.

We need the setpgrp call so that all runqemu processes can easily be
killed (by killing their process group). This presents a problem if this
controlling process itself is killed however since those processes don't
notice the death of the parent and merrily continue on.

Rather than hack runqemu to deal with this, we add something to
qemurunner, at least for now to resolve the issue. Basically we fork off
another process which holds an open pipe to the parent and also is
setpgrp. If/when the pipe sees EOF from the parent dieing, it kills the
process group. This is like pctrl's PDEATHSIG but for a process group
rather than a single process.

Signed-off-by: Richard Purdie <richard.purdie at linuxfoundation.org>

---

 meta/lib/oeqa/utils/qemurunner.py | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/meta/lib/oeqa/utils/qemurunner.py b/meta/lib/oeqa/utils/qemurunner.py
index a137b11..3f3fd8b 100644
--- a/meta/lib/oeqa/utils/qemurunner.py
+++ b/meta/lib/oeqa/utils/qemurunner.py
@@ -125,6 +125,32 @@ class QemuRunner:
         self.runqemu = subprocess.Popen(launch_cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, stdin=subprocess.PIPE, preexec_fn=os.setpgrp)
         output = self.runqemu.stdout
 
+        #
+        # We need the preexec_fn above so that all runqemu processes can easily be killed 
+        # (by killing their process group). This presents a problem if this controlling
+        # process itself is killed however since those processes don't notice the death 
+        # of the parent and merrily continue on.
+        #
+        # Rather than hack runqemu to deal with this, we add something here instead. 
+        # Basically we fork off another process which holds an open pipe to the parent
+        # and also is setpgrp. If/when the pipe sees EOF from the parent dieing, it kills
+        # the process group. This is like pctrl's PDEATHSIG but for a process group
+        # rather than a single process.
+        #
+        r, w = os.pipe()
+        self.monitorpid = os.fork()
+        if self.monitorpid:
+            os.close(r)
+            self.monitorpipe = os.fdopen(w, "w")
+        else:
+            # child process
+            os.setpgrp()
+            os.close(w)
+            r = os.fdopen(r)
+            x = r.read()
+            os.killpg(os.getpgid(self.runqemu.pid), signal.SIGTERM)
+            sys.exit(0)
+
         logger.info("runqemu started, pid is %s" % self.runqemu.pid)
         logger.info("waiting at most %s seconds for qemu pid" % self.runqemutime)
         endtime = time.time() + self.runqemutime
@@ -235,6 +261,7 @@ class QemuRunner:
         self.stop_thread()
         if self.runqemu:
             signal.signal(signal.SIGCHLD, self.origchldhandler)
+            os.kill(self.monitorpid, signal.SIGKILL)
             logger.info("Sending SIGTERM to runqemu")
             try:
                 os.killpg(self.runqemu.pid, signal.SIGTERM)