Search in sources :

Example 1 with InterProcessLock

use of org.apache.curator.framework.recipes.locks.InterProcessLock in project coprhd-controller by CoprHD.

the class DistributedOwnerLockServiceImpl method acquireLock.

/*
     * (non-Javadoc)
     * 
     * @see com.emc.storageos.volumecontroller.impl.DistributedOwnerLock#acquireLock(java.lang.String, java.lang.String, long)
     */
@Override
public boolean acquireLock(String lockKey, String owner, long lockingStartedTimeSeconds, long maxWaitSeconds) {
    boolean acquired = false;
    long waitTime = 0;
    InterProcessLock lock = null;
    boolean reportedLongLock = false;
    boolean reportedBlocking = false;
    do {
        long currentTime = System.currentTimeMillis();
        try {
            // Get semaphore
            lock = lockIPL(lockKey);
            if (lock != null) {
                // Get the lock data.
                DistributedOwnerLockData data = loadLockData(lockKey);
                // If no data, then we got the lock
                if (data == null) {
                    data = new DistributedOwnerLockData(owner, currentTime);
                    persistLockData(lockKey, data);
                    acquired = true;
                } else {
                    // If we're already the owner, that's fine.
                    if (data.owner.equals(owner)) {
                        acquired = true;
                    } else if (!reportedLongLock && currentTime / 1000 > data.timeAcquired + 3600) {
                        reportedLongLock = true;
                        log.info("Lock held more than 1 hour: " + lockKey + " owner: " + data.owner);
                    }
                }
            }
        } finally {
            unlockIPL(lock);
        }
        // Report the time to acquire the lock if acquired.
        if (acquired) {
            log.info(String.format("Lock %s owner %s acquired after %d seconds", lockKey, owner, (currentTime / 1000) - lockingStartedTimeSeconds));
        } else // Sleep if we did not acquire the lock and want to block
        if (maxWaitSeconds > 0) {
            try {
                if (!reportedBlocking) {
                    reportedBlocking = true;
                    log.info(String.format("Owner %s blocking to wait for lock %s maxWaitSeconds %d", owner, lockKey, maxWaitSeconds));
                }
                Thread.sleep(SLEEP_MS_BETWEEN_ACQUIRE_ATTEMPTS);
            } catch (Exception ex) {
                log.error(ex.getMessage(), ex);
            }
        }
        waitTime = (System.currentTimeMillis() / 1000) - lockingStartedTimeSeconds;
    } while (!acquired && waitTime < maxWaitSeconds);
    if (!acquired && maxWaitSeconds > 0 && waitTime >= maxWaitSeconds) {
        log.info("Timeout waiting on lock: " + lockKey + " owner: " + owner);
    }
    return acquired;
}
Also used : InterProcessLock(org.apache.curator.framework.recipes.locks.InterProcessLock) DeviceControllerException(com.emc.storageos.exceptions.DeviceControllerException)

Example 2 with InterProcessLock

use of org.apache.curator.framework.recipes.locks.InterProcessLock in project coprhd-controller by CoprHD.

the class DistributedOwnerLockServiceImpl method lockIPL.

/**
 * Locks an InterProcessLock using ZK
 *
 * @SlockName
 * @return true if lock acquired, null if not
 */
private InterProcessLock lockIPL(String lockKey) {
    boolean acquired = false;
    InterProcessLock lock = getIPLock(lockKey);
    if (lock == null) {
        return null;
    }
    try {
        acquired = lock.acquire(60, TimeUnit.MINUTES);
        if (acquired) {
            return lock;
        }
    } catch (Exception ex) {
        log.error("Exception locking IPL: " + lockKey, ex);
    }
    log.error("Unable to acquire IPL: " + lockKey);
    return null;
}
Also used : InterProcessLock(org.apache.curator.framework.recipes.locks.InterProcessLock) DeviceControllerException(com.emc.storageos.exceptions.DeviceControllerException)

Example 3 with InterProcessLock

use of org.apache.curator.framework.recipes.locks.InterProcessLock in project coprhd-controller by CoprHD.

the class NetworkDeviceController method removeSanZones.

@Override
public void removeSanZones(URI uri, String fabricId, String fabricWwn, List<Zone> zones, boolean activateZones, String taskId) throws ControllerException {
    NetworkSystem networkSytem = getNetworkSystemObject(uri);
    // Lock to prevent concurrent operations on the same VSAN / FABRIC.
    InterProcessLock fabricLock = NetworkFabricLocker.lockFabric(fabricId, _coordinator);
    try {
        // Get the network system reference for the type of network system managed
        // by the controller.
        NetworkSystemDevice networkDevice = getDevice(networkSytem.getSystemType());
        if (networkDevice == null) {
            throw NetworkDeviceControllerException.exceptions.removeSanZonesFailedNull(networkSytem.getSystemType());
        }
        BiosCommandResult result = networkDevice.removeZones(networkSytem, zones, fabricId, fabricWwn, activateZones);
        setStatus(NetworkSystem.class, networkSytem.getId(), taskId, result.isCommandSuccess(), result.getServiceCoded());
        _auditMgr.recordAuditLog(null, null, EVENT_SERVICE_TYPE, OperationTypeEnum.REMOVE_SAN_ZONE, System.currentTimeMillis(), AuditLogManager.AUDITLOG_SUCCESS, AuditLogManager.AUDITOP_END, networkSytem.getId().toString(), networkSytem.getLabel(), networkSytem.getPortNumber(), networkSytem.getUsername(), networkSytem.getSmisProviderIP(), networkSytem.getSmisPortNumber(), networkSytem.getSmisUserName(), networkSytem.getSmisUseSSL());
    } catch (Exception ex) {
        ServiceError serviceError = NetworkDeviceControllerException.errors.removeSanZonesFailedExc(networkSytem.getSystemType(), ex);
        _dbClient.error(NetworkSystem.class, networkSytem.getId(), taskId, serviceError);
    } finally {
        NetworkFabricLocker.unlockFabric(fabricId, fabricLock);
    }
}
Also used : ServiceError(com.emc.storageos.svcs.errorhandling.model.ServiceError) BiosCommandResult(com.emc.storageos.volumecontroller.impl.BiosCommandResult) NetworkSystem(com.emc.storageos.db.client.model.NetworkSystem) InterProcessLock(org.apache.curator.framework.recipes.locks.InterProcessLock) DatabaseException(com.emc.storageos.db.exceptions.DatabaseException) DeviceControllerException(com.emc.storageos.exceptions.DeviceControllerException) ControllerException(com.emc.storageos.volumecontroller.ControllerException) NetworkDeviceControllerException(com.emc.storageos.networkcontroller.exceptions.NetworkDeviceControllerException)

Example 4 with InterProcessLock

use of org.apache.curator.framework.recipes.locks.InterProcessLock in project coprhd-controller by CoprHD.

the class NetworkDeviceController method addRemoveZones.

/**
 * Add/remove a group of zones as given by their NetworkFabricInfo structures.
 * ALL fabricInfos must be using the same NetworkDevice, and the same fabricId. There is a higher level
 * subroutine to split complex requests into sets of requests with the same NetworkDevice and fabricId.
 *
 * @param networkSystem NetworkDevice
 * @param fabricId String
 * @param exportGroupUri The ExportGroup URI. Used for reference counting.
 * @param fabricInfos - Describe each zone.
 * @param activateZones - activate active zoneset after zones change
 * @param retryAltNetworkDevice - a boolean to indicate if re-try to be done.
 *            This is to stop this function from running again after the alternate
 *            system is retried once.
 * @return BiosCommandResult
 * @throws ControllerException
 */
private BiosCommandResult addRemoveZones(NetworkSystem networkSystem, String fabricId, String fabricWwn, URI exportGroupUri, List<NetworkFCZoneInfo> fabricInfos, boolean doRemove, boolean retryAltNetworkDevice) throws ControllerException {
    BiosCommandResult result = null;
    String taskId = UUID.randomUUID().toString();
    List<Zone> zones = new ArrayList<Zone>();
    // Make the zone operations. Don't make the same zone more than once,
    // as determined by its key. The same zone shows up multiple times because it
    // must be recorded for each volume in the FCZoneReference table.
    HashSet<String> keySet = new HashSet<String>();
    for (NetworkFCZoneInfo fabricInfo : fabricInfos) {
        String key = fabricInfo.makeEndpointsKey();
        if (false == keySet.contains(key)) {
            keySet.add(key);
            // neither create nor delete zones found on the switch
            if (fabricInfo.isExistingZone()) {
                _log.info("Zone {} will not be created or removed on {}, as it is not vipr created. ", fabricInfo.getZoneName(), fabricInfo.toString());
                // neither create nor delete zones found on the switch
                continue;
            }
            // Don't actually remove the zone if it's not the last reference
            if (doRemove && !fabricInfo._isLastReference) {
                _log.info("Zone {} will not be removed on {}, as still the zone is used to expose other volumes in export groups ", fabricInfo.getZoneName(), fabricInfo.toString());
                continue;
            }
            Zone zone = new Zone(fabricInfo.getZoneName());
            for (String address : fabricInfo.getEndPoints()) {
                ZoneMember member = new ZoneMember(address, ConnectivityMemberType.WWPN);
                zone.getMembers().add(member);
            }
            zones.add(zone);
        }
    }
    // Get the network device reference for the type of network device managed
    // by the controller.
    NetworkSystemDevice networkDevice = getDevice(networkSystem.getSystemType());
    if (networkDevice == null) {
        throw NetworkDeviceControllerException.exceptions.addRemoveZonesFailedNull(networkSystem.getSystemType());
    }
    // Lock to prevent concurrent operations on the same VSAN / FABRIC.
    InterProcessLock fabricLock = NetworkFabricLocker.lockFabric(fabricId, _coordinator);
    try {
        if (doRemove) {
            /* Removing zones */
            result = networkDevice.removeZones(networkSystem, zones, fabricId, fabricWwn, true);
            if (result.isCommandSuccess()) {
                for (NetworkFCZoneInfo fabricInfo : fabricInfos) {
                    String refKey = fabricInfo.getZoneName() + " " + fabricInfo.getFcZoneReferenceId().toString();
                    try {
                        FCZoneReference ref = deleteFCZoneReference(fabricInfo);
                        if (ref != null && !zones.isEmpty()) {
                            recordZoneEvent(ref, OperationTypeEnum.REMOVE_SAN_ZONE.name(), OperationTypeEnum.REMOVE_SAN_ZONE.getDescription());
                        }
                    } catch (DatabaseException ex) {
                        _log.error("Could not delete FCZoneReference: " + refKey);
                    }
                }
            }
        } else {
            /* Adding zones */
            _log.debug("Adding zones on network system {} ", networkSystem.getNativeGuid());
            result = networkDevice.addZones(networkSystem, zones, fabricId, fabricWwn, true);
            if (result.isCommandSuccess()) {
                for (NetworkFCZoneInfo fabricInfo : fabricInfos) {
                    String refKey = fabricInfo.getZoneName() + " " + fabricInfo.getVolumeId().toString();
                    try {
                        String[] newOrExisting = new String[1];
                        FCZoneReference ref = addZoneReference(exportGroupUri, fabricInfo, newOrExisting);
                        // this is needed for rollback
                        fabricInfo.setFcZoneReferenceId(ref.getId());
                        _log.info(String.format("%s FCZoneReference key: %s volume %s group %s", newOrExisting[0], ref.getPwwnKey(), ref.getVolumeUri(), exportGroupUri));
                        if (!zones.isEmpty()) {
                            recordZoneEvent(ref, OperationTypeEnum.ADD_SAN_ZONE.name(), OperationTypeEnum.ADD_SAN_ZONE.getDescription());
                        }
                    } catch (DatabaseException ex) {
                        _log.error("Could not persist FCZoneReference: " + refKey);
                    }
                }
            }
        }
        // Update the FCZoneInfo structures if we changed device state for rollback.
        Map<String, String> map = (Map<String, String>) result.getObjectList().get(0);
        for (NetworkFCZoneInfo info : fabricInfos) {
            if (NetworkSystemDevice.SUCCESS.equals(map.get(info.getZoneName()))) {
                info.setCanBeRolledBack(true);
            } else {
                info.setCanBeRolledBack(false);
            }
        }
        if (!result.isCommandSuccess()) {
            ServiceError serviceError = NetworkDeviceControllerException.errors.addRemoveZonesFailed(networkSystem.getSystemType());
            setStatus(ExportGroup.class, exportGroupUri, taskId, false, serviceError);
        } else {
            setStatus(ExportGroup.class, exportGroupUri, taskId, true, null);
        }
        return result;
    } catch (ControllerException ex) {
        String operation = doRemove ? "Remove Zones" : "Add Zones";
        _log.info(String.format("waiting for 2 min before retrying %s with alternate device", operation));
        try {
            Thread.sleep(1000 * 120);
        } catch (InterruptedException e) {
            _log.warn("Thread sleep interrupted.  Allowing to continue without sleep");
        }
        NetworkFCZoneInfo fabricInfo = fabricInfos.get(0);
        URI primaryUri = fabricInfo.getNetworkDeviceId();
        URI altUri = fabricInfo.getAltNetworkDeviceId();
        // If we took an error, attempt a retry with an alternate device if possible.
        if (altUri != null && retryAltNetworkDevice) {
            NetworkFabricLocker.unlockFabric(fabricId, fabricLock);
            fabricLock = null;
            _log.error("Zone operation failed using device: " + primaryUri + " retrying with alternate device: " + altUri);
            fabricInfo.setNetworkDeviceId(altUri);
            networkSystem = getNetworkSystemObject(altUri);
            return addRemoveZones(networkSystem, fabricId, fabricWwn, exportGroupUri, fabricInfos, doRemove, false);
        } else {
            if (result != null) {
                if (!result.isCommandSuccess()) {
                    ServiceError serviceError = NetworkDeviceControllerException.errors.addRemoveZonesFailed(networkSystem.getSystemType());
                    setStatus(ExportGroup.class, exportGroupUri, taskId, false, serviceError);
                } else {
                    setStatus(ExportGroup.class, exportGroupUri, taskId, true, null);
                }
            }
            throw ex;
        }
    } finally {
        NetworkFabricLocker.unlockFabric(fabricId, fabricLock);
    }
}
Also used : ServiceError(com.emc.storageos.svcs.errorhandling.model.ServiceError) DeviceControllerException(com.emc.storageos.exceptions.DeviceControllerException) ControllerException(com.emc.storageos.volumecontroller.ControllerException) NetworkDeviceControllerException(com.emc.storageos.networkcontroller.exceptions.NetworkDeviceControllerException) Zone(com.emc.storageos.networkcontroller.impl.mds.Zone) ArrayList(java.util.ArrayList) URI(java.net.URI) FCZoneReference(com.emc.storageos.db.client.model.FCZoneReference) ExportGroup(com.emc.storageos.db.client.model.ExportGroup) NetworkFCZoneInfo(com.emc.storageos.networkcontroller.NetworkFCZoneInfo) BiosCommandResult(com.emc.storageos.volumecontroller.impl.BiosCommandResult) ZoneMember(com.emc.storageos.networkcontroller.impl.mds.ZoneMember) InterProcessLock(org.apache.curator.framework.recipes.locks.InterProcessLock) DatabaseException(com.emc.storageos.db.exceptions.DatabaseException) ZoneInfoMap(com.emc.storageos.db.client.model.ZoneInfoMap) Map(java.util.Map) HashMap(java.util.HashMap) StringSetMap(com.emc.storageos.db.client.model.StringSetMap) StringMap(com.emc.storageos.db.client.model.StringMap) HashSet(java.util.HashSet)

Example 5 with InterProcessLock

use of org.apache.curator.framework.recipes.locks.InterProcessLock in project coprhd-controller by CoprHD.

the class NetworkDeviceController method removeZone.

/**
 * Remove a zone.
 *
 * @param volUri URI of the Volume
 * @param fabricInfo NetworkFabricInfo generated by NetworkScheduler
 * @return BiosCommandResult
 */
public BiosCommandResult removeZone(URI volUri, NetworkFCZoneInfo fabricInfo, boolean activateZones) throws ControllerException {
    ServiceError serviceError = NetworkDeviceControllerException.errors.zoningFailedArgs(volUri.toString());
    BiosCommandResult result = BiosCommandResult.createErrorResult(serviceError);
    List<Zone> zones = new ArrayList<Zone>();
    Zone zone = new Zone(fabricInfo.getZoneName());
    zones.add(zone);
    String taskId = UUID.randomUUID().toString();
    for (String address : fabricInfo.getEndPoints()) {
        ZoneMember member = new ZoneMember(address, ConnectivityMemberType.WWPN);
        zone.getMembers().add(member);
    }
    // Lock to prevent concurrent operations on the same VSAN / FABRIC.
    InterProcessLock fabricLock = NetworkFabricLocker.lockFabric(fabricInfo.getFabricId(), _coordinator);
    try {
        NetworkSystem device = getNetworkSystemObject(fabricInfo.getNetworkDeviceId());
        // Get the file device reference for the type of file device managed
        // by the controller.
        NetworkSystemDevice networkDevice = getDevice(device.getSystemType());
        if (networkDevice == null) {
            throw NetworkDeviceControllerException.exceptions.removeZoneFailedNull(device.getSystemType());
        }
        if (fabricInfo.isLastReference() == true && !fabricInfo.isExistingZone()) {
            result = networkDevice.removeZones(device, zones, fabricInfo.getFabricId(), fabricInfo.getFabricWwn(), activateZones);
        } else {
            // This is not the last reference, just mark our FCZoneReference for deletion
            result = BiosCommandResult.createSuccessfulResult();
        }
        if (result.isCommandSuccess()) {
            if (fabricInfo.getFcZoneReferenceId() != null) {
                try {
                    // Mark our FcZoneReference object for removal
                    FCZoneReference reference = _dbClient.queryObject(FCZoneReference.class, fabricInfo.getFcZoneReferenceId());
                    if (reference != null) {
                        _dbClient.markForDeletion(reference);
                        recordZoneEvent(reference, OperationTypeEnum.REMOVE_SAN_ZONE.name(), OperationTypeEnum.REMOVE_SAN_ZONE.getDescription());
                    }
                } catch (Exception ex) {
                    _log.error("Can't mark object for removal: " + fabricInfo.getFcZoneReferenceId());
                }
            }
        }
        if (!result.isCommandSuccess()) {
            ServiceError svcError = NetworkDeviceControllerException.errors.removeZoneFailed(volUri.toString(), device.getSystemType());
            setStatus(Volume.class, volUri, taskId, false, svcError);
        } else {
            setStatus(Volume.class, volUri, taskId, true, null);
        }
    } catch (ControllerException ex) {
        _log.info("waiting for 2 min before retrying removeZone with alternate device");
        try {
            Thread.sleep(1000 * 120);
        } catch (InterruptedException e) {
            _log.warn("Thread sleep interrupted.  Allowing to continue without sleep");
        }
        URI primaryUri = fabricInfo.getNetworkDeviceId();
        URI altUri = fabricInfo.getAltNetworkDeviceId();
        if (altUri != null && altUri != primaryUri) {
            NetworkFabricLocker.unlockFabric(fabricInfo.getFabricId(), fabricLock);
            fabricLock = null;
            _log.error("Remove Zone failed using device: " + primaryUri + " retrying with alternate device: " + altUri);
            fabricInfo.setNetworkDeviceId(altUri);
            return removeZone(volUri, fabricInfo, activateZones);
        } else {
            ServiceError svcError = NetworkDeviceControllerException.errors.removeZoneFailedExc(volUri.toString());
            setStatus(Volume.class, volUri, taskId, false, svcError);
            throw ex;
        }
    } finally {
        NetworkFabricLocker.unlockFabric(fabricInfo.getFabricId(), fabricLock);
    }
    return result;
}
Also used : ServiceError(com.emc.storageos.svcs.errorhandling.model.ServiceError) DeviceControllerException(com.emc.storageos.exceptions.DeviceControllerException) ControllerException(com.emc.storageos.volumecontroller.ControllerException) NetworkDeviceControllerException(com.emc.storageos.networkcontroller.exceptions.NetworkDeviceControllerException) Zone(com.emc.storageos.networkcontroller.impl.mds.Zone) ArrayList(java.util.ArrayList) NetworkSystem(com.emc.storageos.db.client.model.NetworkSystem) URI(java.net.URI) DatabaseException(com.emc.storageos.db.exceptions.DatabaseException) DeviceControllerException(com.emc.storageos.exceptions.DeviceControllerException) ControllerException(com.emc.storageos.volumecontroller.ControllerException) NetworkDeviceControllerException(com.emc.storageos.networkcontroller.exceptions.NetworkDeviceControllerException) FCZoneReference(com.emc.storageos.db.client.model.FCZoneReference) Volume(com.emc.storageos.db.client.model.Volume) BiosCommandResult(com.emc.storageos.volumecontroller.impl.BiosCommandResult) ZoneMember(com.emc.storageos.networkcontroller.impl.mds.ZoneMember) InterProcessLock(org.apache.curator.framework.recipes.locks.InterProcessLock)

Aggregations

InterProcessLock (org.apache.curator.framework.recipes.locks.InterProcessLock)98 APIException (com.emc.storageos.svcs.errorhandling.resources.APIException)25 DatabaseException (com.emc.storageos.db.exceptions.DatabaseException)21 DeviceControllerException (com.emc.storageos.exceptions.DeviceControllerException)15 IOException (java.io.IOException)15 ControllerException (com.emc.storageos.volumecontroller.ControllerException)14 Configuration (com.emc.storageos.coordinator.common.Configuration)12 CoordinatorException (com.emc.storageos.coordinator.exceptions.CoordinatorException)12 UnknownHostException (java.net.UnknownHostException)12 Site (com.emc.storageos.coordinator.client.model.Site)11 RetryableCoordinatorException (com.emc.storageos.coordinator.exceptions.RetryableCoordinatorException)11 NetworkDeviceControllerException (com.emc.storageos.networkcontroller.exceptions.NetworkDeviceControllerException)10 CheckPermission (com.emc.storageos.security.authorization.CheckPermission)9 ServiceError (com.emc.storageos.svcs.errorhandling.model.ServiceError)9 BiosCommandResult (com.emc.storageos.volumecontroller.impl.BiosCommandResult)9 ArrayList (java.util.ArrayList)9 POST (javax.ws.rs.POST)9 NetworkSystem (com.emc.storageos.db.client.model.NetworkSystem)8 Path (javax.ws.rs.Path)8 ConfigurationImpl (com.emc.storageos.coordinator.common.impl.ConfigurationImpl)6