Skip to content

GitLab

Commit 5bd9ad1a authored by Leigh B Stoller's avatar Leigh B Stoller
Browse files

Add cancel support. The idea is that a DeleteSlice() with our internal 

cancel option, will stop a CreateSliver() in its tracks. We stop the
monitor, then cleanup the slice. I also added an optimization for tearing
down large numbers of VMs on shared nodes, previously we were doing them
one at a time. Note that only the Portal is going to use this option, since
it loosely depends on code in the XEN clientside (described in another
commit).
parent aacaf7b8
Hide whitespace changes
Inline Side-by-side
protogeni/lib/GeniAM.pm.in
View file @ 5bd9ad1a
......@@ -430,6 +430,11 @@ sub DeleteSliver()
exists($options->{'blocking'}) && $options->{'blocking'}) {
$delete_args->{'blocking'} = 1;
}
# And cancel
if (defined($options) &&
exists($options->{'cancel'}) && $options->{'cancel'}) {
$delete_args->{'cancel'} = 1;
}
my $response = GeniCMV2::DeleteSlice($delete_args);
if (!ref($response)) {
......
protogeni/lib/GeniAggregate.pm.in
View file @ 5bd9ad1a
......@@ -1076,6 +1076,7 @@ sub Action($$$;$)
goto bad;
}
$node->_reloaded(0);
$node->_rebooted(0);
# Backpointer used in WaitForNodes().
$node->_sliver($sliver);
$node->_image(undef);
......@@ -1549,6 +1550,9 @@ sub Action($$$;$)
if (keys(%reboots)) {
my @node_ids = keys(%reboots);
# Mark them as being rebooted. See below.
map { $_->_rebooted(1) } values(%reboots);
#
# Should waiting be an option?
#
......@@ -1773,10 +1777,18 @@ sub WaitForNodes($$@)
$canceled = $experiment->canceled();
if ($canceled) {
print STDERR "WaitForNodes canceled; terminating early!\n";
#
# Mark the remaining nodes as canceled (which really means failure).
#
foreach my $node (values(%nodes)) {
$node->_sliver()->SetStatus("canceled")
if (defined($node->_sliver()));
}
# Reset before return; do not want it left.
$slice->LockTables();
$experiment->SetCancelFlag(0);
# Do this first so others know we are reacting to the cancel.
$slice->ClearMonitorPid();
$experiment->SetCancelFlag(0);
$slice->UnLockTables();
return -1;
}
......@@ -1864,17 +1876,18 @@ sub WaitForNodes($$@)
($state eq TBDB_NODESTATE_BOOTING &&
time() - $node->_laststatestamp() > 400))) {
my $giveup = $node->_retried();
my $minutes = int($waittime / 60);
# physnode, give up right away.
$giveup = 1
if (!$node->isvirtnode());
if ($giveup) {
print STDERR
"$node_id still stuck in $state, giving up.\n";
print STDERR "$node_id still stuck in $state at $minutes, ".
"giving up.\n";
}
else {
print STDERR
"$node_id is stuck in $state, restarting it.\n";
print STDERR "$node_id is stuck in $state at $minutes, ".
"restarting it.\n";
$node->_retried(1);
system("$NODEREBOOT $node_id");
$giveup = 1
......@@ -1893,6 +1906,9 @@ sub WaitForNodes($$@)
if ($node->_laststate() ne $state) {
$node->_laststatestamp(time());
$node->_laststate($state);
my $minutes = int($waittime / 60);
print STDERR "$node switches to $state at ".
"$minutes minutes (" . time() . ")\n";
}
if (int($waittime / 60) > $minutes) {
# Changing minutes is why we get this print for just
......@@ -1978,11 +1994,11 @@ sub WaitForNodes($$@)
}
# Too late, but reset before return; do not want it left set.
$slice->LockTables();
# Do this first. See cancel in DeleteSlice().
$slice->ClearMonitorPid();
if ($experiment->canceled()) {
$experiment->SetCancelFlag(0);
}
# Do this last.
$slice->ClearMonitorPid();
$slice->UnLockTables();
return 0;
......@@ -2180,8 +2196,9 @@ sub UnProvision($;$)
# Might be an aggregate that includes link aggregates. Lets do those
# first to avoid work when tearing down the nodes.
#
my @links = ();
my @nodes = ();
my @links = ();
my @nodes = ();
my @shared = ();
foreach my $sliver (@slivers) {
if (ref($sliver) eq "GeniAggregate::Link" ||
......@@ -2193,7 +2210,22 @@ sub UnProvision($;$)
return -1;
}
elsif (ref($sliver) eq "GeniSliver::Node") {
push(@nodes, $sliver);
#
# We need an optimization here, to speed up teardown of
# vnodes on shared nodes, which will happen one at a time
# if we let them go through the sliver Unprovision() call.
#
my $node = Node->Lookup($sliver->resource_id());
if (!defined($node)) {
print STDERR "Could not map $sliver to its a node\n";
next;
}
if ($node->isvirtnode() && $node->sharing_mode()) {
push(@shared, $sliver);
}
else {
push(@nodes, $sliver);
}
}
}
foreach my $sliver (@links) {
......@@ -2210,6 +2242,34 @@ sub UnProvision($;$)
next;
}
}
# Now do the shared nodes in a group.
if (@shared) {
my $experiment = Experiment->Lookup($self->slice_uuid());
if (!defined($experiment)) {
print STDERR "Could not map $self to its experiment\n";
return 0;
}
my $pid = $experiment->pid();
my $eid = $experiment->eid();
my @ids = map { $_->resource_id() } @shared;
system("$VNODESETUP -j -q -m -k $pid $eid @ids");
if ($?) {
print STDERR "$VNODESETUP -k failed on @ids\n";
}
# Mark as stopped to avoid duplicate work later.
foreach my $sliver (@shared) {
$sliver->SetStatus("stopped");
#
# Now we can unprovision, which will be much faster now.
#
if ($sliver->UnProvision($nophysfree) != 0) {
print STDERR "Could not unprovision $sliver in $self\n";
$sliver->SetStatus("broken");
next;
}
}
}
return 0;
}
......
protogeni/lib/GeniCMV2.pm.in
View file @ 5bd9ad1a
......@@ -715,7 +715,9 @@ sub DeleteSlice($)
my $slice_urn = $argref->{'slice_urn'};
my $credentials = $argref->{'credentials'};
my $impotent = $argref->{'impotent'} || 0;
my $cancel = 0;
my $blocking = 0;
my $canceled = 0;
if (! (defined($credentials) && defined($slice_urn))) {
return GeniResponse->MalformedArgsResponse("Missing arguments");
......@@ -726,6 +728,9 @@ sub DeleteSlice($)
if (exists($argref->{'blocking'}) && $argref->{'blocking'}) {
$blocking = 1;
}
if (exists($argref->{'cancel'}) && $argref->{'cancel'}) {
$cancel = 1;
}
my ($credential,$speaksfor) = GeniStd::CheckCredentials($credentials);
return $credential
if (GeniResponse::IsResponse($credential));
......@@ -767,11 +772,26 @@ sub DeleteSlice($)
#
GeniCM::CheckMonitor($slice);
if ($slice->monitor_pid()) {
$slice->UnLock();
return GeniResponse->MonitorResponse();
if (!$cancel) {
$slice->UnLock();
return GeniResponse->MonitorResponse()
}
#
# But what we can do is set the cancel flag, which the monitor is
# checking each time through the loop. This will cause it to stop
# rebooting timed out nodes, and quit earlier. The caller will not
# have to retry as long.
#
my $slice_experiment = $slice->GetExperiment();
if (defined($slice_experiment)) {
$slice_experiment->SetCancelFlag(1);
}
print STDERR "Canceling the monitor (" . $slice->monitor_pid() . ")\n";
$canceled = 1;
}
# If any slivers are imaging, then we are busy as well.
if (defined($aggregate) &&
elsif (defined($aggregate) &&
$aggregate->CheckSliverStates("imaging")) {
$slice->UnLock();
return GeniResponse->BusyResponse();
......@@ -787,6 +807,16 @@ sub DeleteSlice($)
return GeniResponse->Create(GENIRESPONSE_SUCCESS);
}
}
#
# If we were canceled, we wait for the monitor to stop, instead of
#
if ($canceled) {
while ($slice->monitor_pid()) {
sleep(10);
GeniCM::CheckMonitor($slice);
print STDERR "Checking to see if monitor has stopped ...\n";
}
}
my $retval = GeniCM::CleanupDeadSlice($slice, 1);
if ($retval) {
return -1
......
protogeni/lib/GeniSliver.pm.in
View file @ 5bd9ad1a
......@@ -1004,18 +1004,24 @@ sub UnProvision($;$)
my $pid = $experiment->pid();
my $eid = $experiment->eid();
if ($node->isremotenode() && $node->isvirtnode()) {
system("$VNODESETUP -p -q -m -k $pid $eid $node_id");
if ($?) {
print STDERR "$VNODESETUP -k failed on $node_id\n";
return -1;
#
# Shared virtnode only for this part, we want to avoid duplicated
# or unnecessary work (no need to tear down on dedicated hosts).
#
if ($self->status() ne "stopped") {
if ($node->isremotenode() && $node->isvirtnode()) {
system("$VNODESETUP -p -q -m -k $pid $eid $node_id");
if ($?) {
print STDERR "$VNODESETUP -k failed on $node_id\n";
return -1;
}
}
}
elsif ($node->sharing_mode()) {
system("$VNODESETUP -j -q -m -k $pid $eid $node_id");
if ($?) {
print STDERR "$VNODESETUP -k failed on $node_id\n";
return -1;
elsif ($node->sharing_mode()) {
system("$VNODESETUP -j -q -m -k $pid $eid $node_id");
if ($?) {
print STDERR "$VNODESETUP -k failed on $node_id\n";
return -1;
}
}
}
......@@ -1311,7 +1317,7 @@ sub ComputeStatus($$)
# If the sliver is "broken" then call it failed. It might be failed,
# but if the node actually came up okay later, we set it back to okay.
#
if ($self->status() eq "broken") {
if ($self->status() eq "broken" || $self->status() eq "canceled") {
$$pref = "failed";
return 0;
}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment