Kobi Ianko has posted comments on this change.
Change subject: engine: HA VM Reservation feature
......................................................................
Patch Set 2:
(27 comments)
all the comments are fixed but form the one to filter ha vm in the DB and not
in the Java. Will consult with Gilad if it's the best solution.
....................................................
File
backend/manager/modules/bll/src/main/java/org/ovirt/engine/core/bll/scheduling/HaReservationHandling.java
Line 15: import org.ovirt.engine.core.utils.linq.Predicate;
Line 16:
Line 17: /**
Line 18: * A halper class for the scheduling mechanism for checking the HA
Reservation status of a Cluster
Line 19: * @author kianku
Done
Line 20: */
Line 21: public class HaReservationHandling {
Line 22:
Line 23: /**
Line 23: /**
Line 24: * @param cluster
Line 25: * - Cluster to check
Line 26: * @param failedHosts
Line 27: * - a list to return all the hosts that failed the
check, must be initialize outside this method
Done
Line 28: * @return true: Cluster is HaReservation safe. false: a fail over
in one of the Clusters VDSs could negatively
Line 29: * impacting performance.
Line 30: */
Line 31: public boolean checkHaReservationStatusForCluster(VDSGroup
cluster, List<VDS> failedHosts) {
Line 24: * @param cluster
Line 25: * - Cluster to check
Line 26: * @param failedHosts
Line 27: * - a list to return all the hosts that failed the
check, must be initialize outside this method
Line 28: * @return true: Cluster is HaReservation safe. false: a fail over
in one of the Clusters VDSs could negatively
Done
Line 29: * impacting performance.
Line 30: */
Line 31: public boolean checkHaReservationStatusForCluster(VDSGroup
cluster, List<VDS> failedHosts) {
Line 32: List<VDS> hosts =
DbFacade.getInstance().getVdsDao().getAllForVdsGroup(cluster.getId());
Line 31: public boolean checkHaReservationStatusForCluster(VDSGroup
cluster, List<VDS> failedHosts) {
Line 32: List<VDS> hosts =
DbFacade.getInstance().getVdsDao().getAllForVdsGroup(cluster.getId());
Line 33:
Line 34: // No hosts, return true
Line 35: if (hosts == null) {
Done
Line 36: return true;
Line 37: }
Line 38: // HA Reservation is not possible with less the 2 hosts
Line 39: if (hosts.size() < 2) {
Line 34: // No hosts, return true
Line 35: if (hosts == null) {
Line 36: return true;
Line 37: }
Line 38: // HA Reservation is not possible with less the 2 hosts
Done
Line 39: if (hosts.size() < 2) {
Line 40: return false;
Line 41: }
Line 42:
Line 42:
Line 43: // List of host id and cpu/ram free resources
Line 44: // for the outer Pair, first is host id second is a Pair of
cpu and ram
Line 45: // for the inner Pair, first is cpu second is ram
Line 46: List<Pair<Guid, Pair<Integer, Integer>>>
hostsUnutilizedResources = getUnutilizedResources(hosts);
Done
Line 47:
Line 48: List<VM> vmInCluster =
DbFacade.getInstance().getVmDao().getAllForVdsGroup(cluster.getId());
Line 49:
Line 50: if (vmInCluster == null) {
Line 55: @Override
Line 56: public boolean eval(VM v) {
Line 57: return v.isAutoStartup();
Line 58: }
Line 59: });
don't know if it has enough meat for SP, I'll talk to you
Line 60:
Line 61: Map<Guid, List<VM>> hostToHaVmsMapping =
mapVmToHost(vmInCluster);
Line 62:
Line 63: for (VDS host : hosts) {
Line 64: boolean isHaSafe =
Line 65: findReplacementForHost(cluster, host,
Line 66: hostToHaVmsMapping.get(host.getId()),
Line 67: hostsUnutilizedResources);
Line 68: if (!isHaSafe) {
I prefer not to put the function call in the if
Line 69: failedHosts.add(host);
Line 70: }
Line 71: }
Line 72:
Line 69: failedHosts.add(host);
Line 70: }
Line 71: }
Line 72:
Line 73: if (failedHosts.size() == 0) {
Done
used Gilad's suggestion
Line 74: return true;
Line 75: } else {
Line 76: return false;
Line 77: }
Line 69: failedHosts.add(host);
Line 70: }
Line 71: }
Line 72:
Line 73: if (failedHosts.size() == 0) {
Done
Line 74: return true;
Line 75: } else {
Line 76: return false;
Line 77: }
Line 84: Map<Guid, Pair<Integer, Integer>>
additionalHostsUtilizedResources =
Line 85: new HashMap<Guid, Pair<Integer, Integer>>();
Line 86:
Line 87: for (VM vm : vmList) {
Line 88: int curVmMemSize = (int) Math.round(vm.getMemSizeMb() *
(vm.getUsageMemPercent() / 100.0));
Done
Line 89: int curVmCpuPercent =
Line 90: vm.getUsageCpuPercent() * vm.getNumOfCpus()
Line 91: / SlaValidator.getEffectiveCpuCores(host,
cluster.getCountThreadsAsCores());
Line 92:
Line 116: // Found a place for current vm, add the RAM
and CPU size to additionalHostsUtilizedResources
Line 117: if
(!additionalHostsUtilizedResources.containsKey(hostData.getFirst())) {
Line 118: Pair<Integer, Long> cpuRamPair = new
Pair<Integer, Long>();
Line 119: cpuRamPair.setSecond(new
Long(curVmMemSize));
Line 120: cpuRamPair.setFirst(curVmCpuPercent);
Done
Line 121: } else {
Line 122: Pair<Integer, Integer> cpuRamPair =
Line 123:
additionalHostsUtilizedResources.get(hostData.getFirst());
Line 124:
cpuRamPair.setSecond(cpuRamPair.getSecond() + curVmMemSize);
Line 145: private Map<Guid, List<VM>> mapVmToHost(List<VM> vms) {
Line 146: Map<Guid, List<VM>> hostToHaVmsMapping = new HashMap<Guid,
List<VM>>();
Line 147:
Line 148: for (VM vm : vms) {
Line 149: if (vm.getRunOnVds() != null) {
Done
Line 150: if
(!hostToHaVmsMapping.containsKey(vm.getRunOnVds())) {
Line 151: List<VM> vmsOfHost = new ArrayList<VM>();
Line 152: vmsOfHost.add(vm);
Line 153: hostToHaVmsMapping.put(vm.getRunOnVds(),
vmsOfHost);
Line 160: }
Line 161:
Line 162: private List<Pair<Guid, Pair<Integer, Integer>>>
getUnutilizedResources(List<VDS> hosts) {
Line 163: List<Pair<Guid, Pair<Integer, Integer>>>
hostsUnutilizedResources =
Line 164: new ArrayList<Pair<Guid, Pair<Integer, Integer>>>();
Done
Line 165: for (VDS host : hosts) {
Line 166: Pair<Integer, Integer> innerUnutilizedCpuRamPair = new
Pair<Integer, Integer>();
Line 167: innerUnutilizedCpuRamPair.setFirst(100 -
host.getUsageCpuPercent());
Line 168:
....................................................
File
backend/manager/modules/bll/src/main/java/org/ovirt/engine/core/bll/scheduling/SchedulingManager.java
Line 749:
SchedulerUtilQuartzImpl.getInstance().scheduleAFixedDelayJob(instance,
Line 750: "performHaResevationCheck",
Line 751: new Class[] {},
Line 752: new Object[] {},
Line 753: Config.<Integer>
getValue(ConfigValues.VdsHaReservationIntervalInMinutes),
Done
Line 754: Config.<Integer>
getValue(ConfigValues.VdsHaReservationIntervalInMinutes),
Line 755: TimeUnit.MINUTES);
Line 756: log.info("Finished HA Reservation check");
Line 757: }
Line 767: for (VDSGroup cluster : clusters) {
Line 768: if (cluster.supportsHaReservation()) {
Line 769: List<VDS> returnedFailedHosts = new
ArrayList<VDS>();
Line 770: boolean status =
Line 771: new
HaReservationHandling().checkHaReservationStatusForCluster(cluster,
returnedFailedHosts);
Done
Line 772: if (!status) {
Line 773: // create Alert using returnedFailedHosts
Line 774: AuditLogableBase logable = new
AuditLogableBase();
Line 775: logable.setVdsGroupId(cluster.getId());
Line 783: if (commaCount > 0) {
Line 784: failedHostsStr.append(", ");
Line 785: commaCount--;
Line 786: }
Line 787: }
Done
Line 788: logable.addCustomValue("Hosts",
failedHostsStr.toString());
Line 789: AlertDirector.Alert(logable,
AuditLogType.CLUSTER_ALERT_HA_RESERVATION);
Line 790: }
Line 791: }
....................................................
File
backend/manager/modules/bll/src/main/java/org/ovirt/engine/core/bll/scheduling/policyunits/HaReservationBalancePolicyUnit.java
Line 20: import org.ovirt.engine.core.dal.dbbroker.DbFacade;
Line 21: import org.ovirt.engine.core.utils.linq.LinqUtils;
Line 22: import org.ovirt.engine.core.utils.linq.Predicate;
Line 23:
Line 24:
Done
Line 25: public class HaReservationBalancePolicyUnit extends PolicyUnitImpl {
Line 26:
Line 27: public HaReservationBalancePolicyUnit(PolicyUnit policyUnit) {
Line 28: super(policyUnit);
Line 52: int optimalHaDistribution = (int) Math.ceil(((double)
haVmsInCluster / hosts.size()));
Line 53:
Line 54: int overUtilizationParam = 100;
Line 55: if (parameters.get("OverUtilization") != null) {
Line 56: overUtilizationParam =
Integer.parseInt(parameters.get("OverUtilization"));
Done
Line 57: } else {
Line 58: overUtilizationParam = Config.<Integer>
getValue(ConfigValues.OverUtilizationForHaReservation);
Line 59: }
Line 60:
Line 60:
Line 61: int overUtilizationThreshold = (int)
Math.ceil(optimalHaDistribution * (overUtilizationParam / 100));
Line 62:
Line 63: List<VDS> overUtilizedHosts = getOverUtilizedHosts(hosts,
hostId2HaVmMapping, optimalHaDistribution);
Line 64: if (overUtilizedHosts.size() == 0) {
Done
Line 65: return null;
Line 66: }
Line 67:
Line 68: List<VDS> underUtilizedHosts = getUnderUtilizedHosts(hosts,
hostId2HaVmMapping, overUtilizationParam);
Line 90: return new Pair<List<Guid>, Guid>(underUtilizedHostsKeys,
vm.getId());
Line 91:
Line 92: }
Line 93:
Line 94: private int countHaVmsInCluster(Map<Guid, List<VM>>
hostId2HaVmMapping) {
when I get them, I don't know if there HA or not, I'm filtering the HA when I
map them to host.
I can count them when I map them to hosts, but then I will have 2 return value,
that is less elegant.
since there is no impact on performance, I'd like to leave it as is
Line 95: int result = 0;
Line 96: for (Entry<Guid, List<VM>> entry :
hostId2HaVmMapping.entrySet()) {
Line 97: result += entry.getValue().size();
Line 98: }
Line 119: }
Line 120:
Line 121: private List<VDS> getOverUtilizedHosts(List<VDS> hosts,
Line 122: Map<Guid, List<VM>> hostId2HaVmMapping,
Line 123: int overUtilizationThreshold) {
Done
Line 124: List<VDS> overUtilizedHosts = new ArrayList<VDS>();
Line 125:
Line 126: for (VDS host : hosts) {
Line 127: int count = 0;
Line 143: vms = LinqUtils.filter(vms, new Predicate<VM>() {
Line 144: @Override
Line 145: public boolean eval(VM v) {
Line 146: return v.getMigrationSupport() ==
MigrationSupport.MIGRATABLE
Line 147: && !hostId.equals(v.getDedicatedVmForVds());
Done
Line 148: }
Line 149: });
Line 150:
Line 151: return vms;
Line 150:
Line 151: return vms;
Line 152: }
Line 153:
Line 154: private Map<Guid, List<VM>> mapHaVmToHostByCluster(Guid
clusterId) {
Done
Line 155:
Line 156: Map<Guid, List<VM>> result = new HashMap<Guid, List<VM>>();
Line 157: List<VM> vms =
DbFacade.getInstance().getVmDao().getAllForVdsGroup(clusterId);
Line 158: if (vms != null) {
....................................................
File
backend/manager/modules/bll/src/main/java/org/ovirt/engine/core/bll/scheduling/policyunits/HaReservationWeightPolicyUnit.java
Line 84: return scores;
Line 85: }
Line 86:
Line 87:
Line 88: private Map<Guid, List<VM>> mapHaVmToHostByCluster(Guid clusterId)
{
Done
Line 89:
Line 90: Map<Guid, List<VM>> result = new HashMap<Guid, List<VM>>();
Line 91: List<VM> vms =
DbFacade.getInstance().getVmDao().getAllForVdsGroup(clusterId);
Line 92: if (vms != null) {
....................................................
File packaging/dbscripts/upgrade/pre_upgrade/0000_config.sql
Line 613: -- Cluster HA Reservation
Line 614: select
fn_db_add_config_value('OverUtilizationForHaReservation','200','3.4');
Line 615: select fn_db_add_config_value('ScaleDownForHaReservation','1','3.4');
Line 616: select fn_db_add_config_value('EnableVdsHaReservation','true','3.4');
Line 617: select
fn_db_add_config_value('VdsHaReservationIntervalInMinutes','5','3.4');
Done
Line 618:
Line 619:
------------------------------------------------------------------------------------
Line 620: -- Update with override section
Line 621:
------------------------------------------------------------------------------------
....................................................
File packaging/dbscripts/vds_groups_sp.sql
Line 45:
Line 46:
Line 47:
Line 48:
Line 49:
Done
Line 50: Create or replace FUNCTION UpdateVdsGroup(v_description VARCHAR(4000) ,
Line 51: v_free_text_comment text,
Line 52: v_name VARCHAR(40),
Line 53: v_vds_group_id UUID,
--
To view, visit http://gerrit.ovirt.org/22929
To unsubscribe, visit http://gerrit.ovirt.org/settings
Gerrit-MessageType: comment
Gerrit-Change-Id: I44e698e58ff5e4a0b74249d3fe3cf4acf042c324
Gerrit-PatchSet: 2
Gerrit-Project: ovirt-engine
Gerrit-Branch: master
Gerrit-Owner: Kobi Ianko <k...@redhat.com>
Gerrit-Reviewer: Arik Hadas <aha...@redhat.com>
Gerrit-Reviewer: Doron Fediuck <dfedi...@redhat.com>
Gerrit-Reviewer: Gilad Chaplik <gchap...@redhat.com>
Gerrit-Reviewer: Itamar Heim <ih...@redhat.com>
Gerrit-Reviewer: Kobi Ianko <k...@redhat.com>
Gerrit-Reviewer: Omer Frenkel <ofren...@redhat.com>
Gerrit-Reviewer: Oved Ourfali <oourf...@redhat.com>
Gerrit-Reviewer: Roy Golan <rgo...@redhat.com>
Gerrit-Reviewer: oVirt Jenkins CI Server
Gerrit-HasComments: Yes
_______________________________________________
Engine-patches mailing list
Engine-patches@ovirt.org
http://lists.ovirt.org/mailman/listinfo/engine-patches
