Revision 39eb39dc
| b/lib/cmdlib.py | ||
|---|---|---|
| 2610 | 2610 |
msg = "cannot reach the master IP" |
| 2611 | 2611 |
_ErrorIf(True, constants.CV_ENODENET, node, msg) |
| 2612 | 2612 |
|
| 2613 |
def _VerifyInstance(self, instance, instanceconfig, node_image,
|
|
| 2613 |
def _VerifyInstance(self, instance, inst_config, node_image,
|
|
| 2614 | 2614 |
diskstatus): |
| 2615 | 2615 |
"""Verify an instance. |
| 2616 | 2616 |
|
| 2617 | 2617 |
This function checks to see if the required block devices are |
| 2618 |
available on the instance's node. |
|
| 2618 |
available on the instance's node, and that the nodes are in the correct |
|
| 2619 |
state. |
|
| 2619 | 2620 |
|
| 2620 | 2621 |
""" |
| 2621 | 2622 |
_ErrorIf = self._ErrorIf # pylint: disable=C0103 |
| 2622 |
node_current = instanceconfig.primary_node |
|
| 2623 |
pnode = inst_config.primary_node |
|
| 2624 |
pnode_img = node_image[pnode] |
|
| 2625 |
groupinfo = self.cfg.GetAllNodeGroupsInfo() |
|
| 2623 | 2626 |
|
| 2624 | 2627 |
node_vol_should = {}
|
| 2625 |
instanceconfig.MapLVsByNode(node_vol_should)
|
|
| 2628 |
inst_config.MapLVsByNode(node_vol_should)
|
|
| 2626 | 2629 |
|
| 2627 | 2630 |
cluster = self.cfg.GetClusterInfo() |
| 2628 | 2631 |
ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster, |
| 2629 | 2632 |
self.group_info) |
| 2630 |
err = _ComputeIPolicyInstanceViolation(ipolicy, instanceconfig)
|
|
| 2633 |
err = _ComputeIPolicyInstanceViolation(ipolicy, inst_config)
|
|
| 2631 | 2634 |
_ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, utils.CommaJoin(err), |
| 2632 | 2635 |
code=self.ETYPE_WARNING) |
| 2633 | 2636 |
|
| ... | ... | |
| 2641 | 2644 |
_ErrorIf(test, constants.CV_EINSTANCEMISSINGDISK, instance, |
| 2642 | 2645 |
"volume %s missing on node %s", volume, node) |
| 2643 | 2646 |
|
| 2644 |
if instanceconfig.admin_state == constants.ADMINST_UP: |
|
| 2645 |
pri_img = node_image[node_current] |
|
| 2646 |
test = instance not in pri_img.instances and not pri_img.offline |
|
| 2647 |
if inst_config.admin_state == constants.ADMINST_UP: |
|
| 2648 |
test = instance not in pnode_img.instances and not pnode_img.offline |
|
| 2647 | 2649 |
_ErrorIf(test, constants.CV_EINSTANCEDOWN, instance, |
| 2648 | 2650 |
"instance not running on its primary node %s", |
| 2649 |
node_current) |
|
| 2651 |
pnode) |
|
| 2652 |
_ErrorIf(pnode_img.offline, constants.CV_EINSTANCEBADNODE, instance, |
|
| 2653 |
"instance is marked as running and lives on offline node %s", |
|
| 2654 |
pnode) |
|
| 2650 | 2655 |
|
| 2651 | 2656 |
diskdata = [(nname, success, status, idx) |
| 2652 | 2657 |
for (nname, disks) in diskstatus.items() |
| ... | ... | |
| 2657 | 2662 |
# node here |
| 2658 | 2663 |
snode = node_image[nname] |
| 2659 | 2664 |
bad_snode = snode.ghost or snode.offline |
| 2660 |
_ErrorIf(instanceconfig.admin_state == constants.ADMINST_UP and
|
|
| 2665 |
_ErrorIf(inst_config.admin_state == constants.ADMINST_UP and
|
|
| 2661 | 2666 |
not success and not bad_snode, |
| 2662 | 2667 |
constants.CV_EINSTANCEFAULTYDISK, instance, |
| 2663 | 2668 |
"couldn't retrieve status for disk/%s on %s: %s", |
| 2664 | 2669 |
idx, nname, bdev_status) |
| 2665 |
_ErrorIf((instanceconfig.admin_state == constants.ADMINST_UP and
|
|
| 2670 |
_ErrorIf((inst_config.admin_state == constants.ADMINST_UP and
|
|
| 2666 | 2671 |
success and bdev_status.ldisk_status == constants.LDS_FAULTY), |
| 2667 | 2672 |
constants.CV_EINSTANCEFAULTYDISK, instance, |
| 2668 | 2673 |
"disk/%s on %s is faulty", idx, nname) |
| 2669 | 2674 |
|
| 2675 |
_ErrorIf(pnode_img.rpc_fail and not pnode_img.offline, |
|
| 2676 |
constants.CV_ENODERPC, pnode, "instance %s, connection to" |
|
| 2677 |
" primary node failed", instance) |
|
| 2678 |
|
|
| 2679 |
_ErrorIf(len(inst_config.secondary_nodes) > 1, |
|
| 2680 |
constants.CV_EINSTANCELAYOUT, |
|
| 2681 |
instance, "instance has multiple secondary nodes: %s", |
|
| 2682 |
utils.CommaJoin(inst_config.secondary_nodes), |
|
| 2683 |
code=self.ETYPE_WARNING) |
|
| 2684 |
|
|
| 2685 |
if inst_config.disk_template in constants.DTS_INT_MIRROR: |
|
| 2686 |
instance_nodes = utils.NiceSort(inst_config.all_nodes) |
|
| 2687 |
instance_groups = {}
|
|
| 2688 |
|
|
| 2689 |
for node in instance_nodes: |
|
| 2690 |
instance_groups.setdefault(self.all_node_info[node].group, |
|
| 2691 |
[]).append(node) |
|
| 2692 |
|
|
| 2693 |
pretty_list = [ |
|
| 2694 |
"%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name) |
|
| 2695 |
# Sort so that we always list the primary node first. |
|
| 2696 |
for group, nodes in sorted(instance_groups.items(), |
|
| 2697 |
key=lambda (_, nodes): pnode in nodes, |
|
| 2698 |
reverse=True)] |
|
| 2699 |
|
|
| 2700 |
self._ErrorIf(len(instance_groups) > 1, |
|
| 2701 |
constants.CV_EINSTANCESPLITGROUPS, |
|
| 2702 |
instance, "instance has primary and secondary nodes in" |
|
| 2703 |
" different groups: %s", utils.CommaJoin(pretty_list), |
|
| 2704 |
code=self.ETYPE_WARNING) |
|
| 2705 |
|
|
| 2706 |
inst_nodes_offline = [] |
|
| 2707 |
for snode in inst_config.secondary_nodes: |
|
| 2708 |
s_img = node_image[snode] |
|
| 2709 |
_ErrorIf(s_img.rpc_fail and not s_img.offline, constants.CV_ENODERPC, |
|
| 2710 |
snode, "instance %s, connection to secondary node failed", |
|
| 2711 |
instance) |
|
| 2712 |
|
|
| 2713 |
if s_img.offline: |
|
| 2714 |
inst_nodes_offline.append(snode) |
|
| 2715 |
|
|
| 2716 |
# warn that the instance lives on offline nodes |
|
| 2717 |
_ErrorIf(inst_nodes_offline, constants.CV_EINSTANCEBADNODE, instance, |
|
| 2718 |
"instance has offline secondary node(s) %s", |
|
| 2719 |
utils.CommaJoin(inst_nodes_offline)) |
|
| 2720 |
# ... or ghost/non-vm_capable nodes |
|
| 2721 |
for node in inst_config.all_nodes: |
|
| 2722 |
_ErrorIf(node_image[node].ghost, constants.CV_EINSTANCEBADNODE, |
|
| 2723 |
instance, "instance lives on ghost node %s", node) |
|
| 2724 |
_ErrorIf(not node_image[node].vm_capable, constants.CV_EINSTANCEBADNODE, |
|
| 2725 |
instance, "instance lives on non-vm_capable node %s", node) |
|
| 2726 |
|
|
| 2670 | 2727 |
def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved): |
| 2671 | 2728 |
"""Verify if there are any unknown volumes in the cluster. |
| 2672 | 2729 |
|
| ... | ... | |
| 3305 | 3362 |
vg_name = self.cfg.GetVGName() |
| 3306 | 3363 |
drbd_helper = self.cfg.GetDRBDHelper() |
| 3307 | 3364 |
cluster = self.cfg.GetClusterInfo() |
| 3308 |
groupinfo = self.cfg.GetAllNodeGroupsInfo() |
|
| 3309 | 3365 |
hypervisors = cluster.enabled_hypervisors |
| 3310 | 3366 |
node_data_list = [self.my_node_info[name] for name in self.my_node_names] |
| 3311 | 3367 |
|
| ... | ... | |
| 3584 | 3640 |
inst_config = self.my_inst_info[instance] |
| 3585 | 3641 |
self._VerifyInstance(instance, inst_config, node_image, |
| 3586 | 3642 |
instdisk[instance]) |
| 3587 |
inst_nodes_offline = [] |
|
| 3588 |
|
|
| 3589 |
pnode = inst_config.primary_node |
|
| 3590 |
pnode_img = node_image[pnode] |
|
| 3591 |
_ErrorIf(pnode_img.rpc_fail and not pnode_img.offline, |
|
| 3592 |
constants.CV_ENODERPC, pnode, "instance %s, connection to" |
|
| 3593 |
" primary node failed", instance) |
|
| 3594 |
|
|
| 3595 |
_ErrorIf(inst_config.admin_state == constants.ADMINST_UP and |
|
| 3596 |
pnode_img.offline, |
|
| 3597 |
constants.CV_EINSTANCEBADNODE, instance, |
|
| 3598 |
"instance is marked as running and lives on offline node %s", |
|
| 3599 |
inst_config.primary_node) |
|
| 3600 | 3643 |
|
| 3601 | 3644 |
# If the instance is non-redundant we cannot survive losing its primary |
| 3602 | 3645 |
# node, so we are not N+1 compliant. |
| 3603 | 3646 |
if inst_config.disk_template not in constants.DTS_MIRRORED: |
| 3604 | 3647 |
i_non_redundant.append(instance) |
| 3605 | 3648 |
|
| 3606 |
_ErrorIf(len(inst_config.secondary_nodes) > 1, |
|
| 3607 |
constants.CV_EINSTANCELAYOUT, |
|
| 3608 |
instance, "instance has multiple secondary nodes: %s", |
|
| 3609 |
utils.CommaJoin(inst_config.secondary_nodes), |
|
| 3610 |
code=self.ETYPE_WARNING) |
|
| 3611 |
|
|
| 3612 |
if inst_config.disk_template in constants.DTS_INT_MIRROR: |
|
| 3613 |
pnode = inst_config.primary_node |
|
| 3614 |
instance_nodes = utils.NiceSort(inst_config.all_nodes) |
|
| 3615 |
instance_groups = {}
|
|
| 3616 |
|
|
| 3617 |
for node in instance_nodes: |
|
| 3618 |
instance_groups.setdefault(self.all_node_info[node].group, |
|
| 3619 |
[]).append(node) |
|
| 3620 |
|
|
| 3621 |
pretty_list = [ |
|
| 3622 |
"%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name) |
|
| 3623 |
# Sort so that we always list the primary node first. |
|
| 3624 |
for group, nodes in sorted(instance_groups.items(), |
|
| 3625 |
key=lambda (_, nodes): pnode in nodes, |
|
| 3626 |
reverse=True)] |
|
| 3627 |
|
|
| 3628 |
self._ErrorIf(len(instance_groups) > 1, |
|
| 3629 |
constants.CV_EINSTANCESPLITGROUPS, |
|
| 3630 |
instance, "instance has primary and secondary nodes in" |
|
| 3631 |
" different groups: %s", utils.CommaJoin(pretty_list), |
|
| 3632 |
code=self.ETYPE_WARNING) |
|
| 3633 |
|
|
| 3634 | 3649 |
if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]: |
| 3635 | 3650 |
i_non_a_balanced.append(instance) |
| 3636 | 3651 |
|
| 3637 |
for snode in inst_config.secondary_nodes: |
|
| 3638 |
s_img = node_image[snode] |
|
| 3639 |
_ErrorIf(s_img.rpc_fail and not s_img.offline, constants.CV_ENODERPC, |
|
| 3640 |
snode, "instance %s, connection to secondary node failed", |
|
| 3641 |
instance) |
|
| 3642 |
|
|
| 3643 |
if s_img.offline: |
|
| 3644 |
inst_nodes_offline.append(snode) |
|
| 3645 |
|
|
| 3646 |
# warn that the instance lives on offline nodes |
|
| 3647 |
_ErrorIf(inst_nodes_offline, constants.CV_EINSTANCEBADNODE, instance, |
|
| 3648 |
"instance has offline secondary node(s) %s", |
|
| 3649 |
utils.CommaJoin(inst_nodes_offline)) |
|
| 3650 |
# ... or ghost/non-vm_capable nodes |
|
| 3651 |
for node in inst_config.all_nodes: |
|
| 3652 |
_ErrorIf(node_image[node].ghost, constants.CV_EINSTANCEBADNODE, |
|
| 3653 |
instance, "instance lives on ghost node %s", node) |
|
| 3654 |
_ErrorIf(not node_image[node].vm_capable, constants.CV_EINSTANCEBADNODE, |
|
| 3655 |
instance, "instance lives on non-vm_capable node %s", node) |
|
| 3656 |
|
|
| 3657 | 3652 |
feedback_fn("* Verifying orphan volumes")
|
| 3658 | 3653 |
reserved = utils.FieldSet(*cluster.reserved_lvs) |
| 3659 | 3654 |
|
Also available in: Unified diff