CP-51658: install script to allow stopping all in-progress GC operations

Signed-off-by: Mark Syms <[email protected]>

Makefile

@@ -202,6 +202,8 @@ install: precheck
 	install -m 755 scripts/usb_change $(SM_STAGING)$(LIBEXEC)
 	install -m 755 scripts/kickpipe $(SM_STAGING)$(LIBEXEC)
 	install -m 755 scripts/set-iscsi-initiator $(SM_STAGING)$(LIBEXEC)
+	mkdir -p $(SM_STAGING)/etc/xapi.d/xapi-pre-shutdown/
+	install -m 755 scripts/stop_all_gc $(SM_STAGING)/etc/xapi.d/xapi-pre-shutdown/
 	$(MAKE) -C dcopy install DESTDIR=$(SM_STAGING)
 	ln -sf $(SM_DEST)blktap2.py $(SM_STAGING)$(BIN_DEST)/blktap2
 	ln -sf $(SM_DEST)lcache.py $(SM_STAGING)$(BIN_DEST)tapdisk-cache-stats

drivers/cleanup.py

@@ -82,11 +82,22 @@
 
 NON_PERSISTENT_DIR = '/run/nonpersistent/sm'
 
+# Signal Handler
+SIGTERM = False
+
 
 class AbortException(util.SMException):
     pass
 
 
+def receiveSignal(signalNumber, frame):
+    global SIGTERM
+
+    util.SMlog("GC: recieved SIGTERM")
+    SIGTERM = True
+    return
+
+
 ################################################################################
 #
 #  Util
@@ -167,7 +178,7 @@ def runAbortable(func, ret, ns, abortTest, pollInterval, timeOut):
                     if resultFlag.test("failure"):
                         resultFlag.clear("failure")
                         raise util.SMException("Child process exited with error")
-                    if abortTest() or abortSignaled:
+                    if abortTest() or abortSignaled or SIGTERM:
                         os.killpg(pid, signal.SIGKILL)
                         raise AbortException("Aborting due to signal")
                     if timeOut and _time() - startTime > timeOut:
@@ -2966,6 +2977,10 @@ def _gcLoop(sr, dryRun=False, immediate=False):
             "Garbage collection for SR %s" % sr.uuid)
         _gcLoopPause(sr, dryRun, immediate=immediate)
         while True:
+            if SIGTERM:
+                Util.log("Term requested")
+                return
+
             if not sr.xapi.isPluggedHere():
                 Util.log("SR no longer attached, exiting")
                 break
@@ -3191,6 +3206,9 @@ def gc(session, srUuid, inBackground, dryRun=False):
     6. If there is something to coalesce, coalesce one pair, then goto 3
     """
     Util.log("=== SR %s: gc ===" % srUuid)
+
+    signal.signal(signal.SIGTERM, receiveSignal)
+
     if inBackground:
         if daemonize():
             # we are now running in the background. Catch & log any errors

mk/sm.spec.in

@@ -116,6 +116,7 @@ tests/run_python_unittests.sh
 /etc/xapi.d/plugins/testing-hooks
 /etc/xapi.d/plugins/intellicache-clean
 /etc/xapi.d/plugins/trim
+/etc/xapi.d/xapi-pre-shutdown/*
 /etc/xensource/master.d/02-vhdcleanup
 /opt/xensource/bin/blktap2
 /opt/xensource/bin/tapdisk-cache-stats

scripts/stop_all_gc

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+/usr/bin/systemctl list-units SMGC@* --all --no-legend | /usr/bin/cut -d ' ' -f1 | sed 's/\\/\\\\/g' | while read service;
+do
+    echo "Stopping $service"
+    /usr/bin/systemctl stop "$service"
+done

tests/test_cleanup.py

@@ -1,4 +1,5 @@
 import errno
+import signal
 import unittest
 import unittest.mock as mock
 
@@ -40,7 +41,8 @@ def acquireNoblock(self):
 
 
 class TestRelease(object):
-    pass
+    def acquireNoblock(self):
+        return True
 
 
 class IrrelevantLock(object):
@@ -72,6 +74,9 @@ def setUp(self):
 
         self.addCleanup(mock.patch.stopall)
 
+    def tearDown(self):
+        cleanup.SIGTERM = False
+
     def setup_abort_flag(self, ipc_mock, should_abort=False):
         flag = mock.Mock()
         flag.test = mock.Mock(return_value=should_abort)
@@ -89,6 +94,26 @@ def mock_cleanup_locks(self):
         cleanup.lockGCRunning = TestRelease()
         cleanup.lockGCRunning.release = mock.Mock(return_value=None)
 
+    def test_term_handler(self):
+        self.assertFalse(cleanup.SIGTERM)
+
+        cleanup.receiveSignal(signal.SIGTERM, None)
+
+        self.assertTrue(cleanup.SIGTERM)
+
+    @mock.patch('cleanup._create_init_file', autospec=True)
+    @mock.patch('cleanup.SR', autospec=True)
+    @mock.patch('cleanup._ensure_xapi_initialised', autospec=True)
+    def test_loop_exits_on_term(self, mock_init, mock_sr, mock_check_xapi):
+        # Set the term signel
+        cleanup.receiveSignal(signal.SIGTERM, None)
+        mock_session = mock.MagicMock(name='MockSession')
+        sr_uuid = str(uuid4())
+        self.mock_cleanup_locks()
+
+        # Trigger GC
+        cleanup.gc(mock_session, sr_uuid, inBackground=False)
+
     def test_lock_if_already_locked(self):
         """
         Given an already locked SR, a lock call