Search in sources :

Example 61 with Node

use of com.yahoo.vdslib.state.Node in project vespa by vespa-engine.

the class RpcServer method handleRpcRequests.

public boolean handleRpcRequests(ContentCluster cluster, ClusterState systemState, NodeStateOrHostInfoChangeHandler changeListener, NodeAddedOrRemovedListener addedListener) {
    boolean handledAnyRequests = false;
    if (!isConnected()) {
        long time = timer.getCurrentTimeInMillis();
        try {
            connect();
        } catch (ListenFailedException e) {
            if (!e.getMessage().equals(lastConnectError) || time - lastConnectErrorTime > 60 * 1000) {
                lastConnectError = e.getMessage();
                lastConnectErrorTime = time;
                log.log(LogLevel.WARNING, "Failed to bind RPC server to port " + port + ": " + e.getMessage());
            }
        } catch (Exception e) {
            if (!e.getMessage().equals(lastConnectError) || time - lastConnectErrorTime > 60 * 1000) {
                lastConnectError = e.getMessage();
                lastConnectErrorTime = time;
                log.log(LogLevel.WARNING, "Failed to initailize RPC server socket: " + e.getMessage());
            }
        }
    }
    for (int j = 0; j < 10; ++j) {
        // Max perform 10 RPC requests per cycle.
        Request req;
        synchronized (monitor) {
            if (rpcRequests.isEmpty())
                break;
            Iterator<Request> it = rpcRequests.iterator();
            req = it.next();
            it.remove();
            handledAnyRequests = true;
        }
        try {
            if (req.methodName().equals("getMaster")) {
                log.log(LogLevel.DEBUG, "Resolving RPC getMaster request");
                Integer master = masterHandler.getMaster();
                String masterReason = masterHandler.getMasterReason();
                req.returnValues().add(new Int32Value(master == null ? -1 : master));
                req.returnValues().add(new StringValue(masterReason == null ? "No reason given" : masterReason));
                req.returnRequest();
                continue;
            }
            if (!masterHandler.isMaster()) {
                throw new IllegalStateException("Refusing to answer RPC calls as we are not the master fleetcontroller.");
            }
            if (req.methodName().equals("getNodeList")) {
                log.log(LogLevel.DEBUG, "Resolving RPC getNodeList request");
                List<String> slobrok = new ArrayList<String>();
                List<String> rpc = new ArrayList<String>();
                for (NodeInfo node : cluster.getNodeInfo()) {
                    String s1 = node.getSlobrokAddress();
                    String s2 = node.getRpcAddress();
                    assert (s1 != null);
                    slobrok.add(s1);
                    rpc.add(s2 == null ? "" : s2);
                }
                req.returnValues().add(new StringArray(slobrok.toArray(new String[slobrok.size()])));
                req.returnValues().add(new StringArray(rpc.toArray(new String[rpc.size()])));
                req.returnRequest();
            } else if (req.methodName().equals("getSystemState")) {
                log.log(LogLevel.DEBUG, "Resolving RPC getSystemState request");
                req.returnValues().add(new StringValue(""));
                req.returnValues().add(new StringValue(systemState.toString(true)));
                req.returnRequest();
            } else if (req.methodName().equals("getNodeState")) {
                log.log(LogLevel.DEBUG, "Resolving RPC getNodeState request");
                NodeType nodeType = NodeType.get(req.parameters().get(0).asString());
                int nodeIndex = req.parameters().get(1).asInt32();
                Node node = new Node(nodeType, nodeIndex);
                // First parameter is current state in system state
                NodeState ns = systemState.getNodeState(node);
                req.returnValues().add(new StringValue(systemState.getNodeState(node).serialize()));
                // Second parameter is state node is reporting
                NodeInfo nodeInfo = cluster.getNodeInfo(node);
                if (nodeInfo == null)
                    throw new RuntimeException("No node " + node + " exists in cluster " + cluster.getName());
                NodeState fromNode = nodeInfo.getReportedState();
                req.returnValues().add(new StringValue(fromNode == null ? "unknown" : fromNode.serialize()));
                // Third parameter is state node has been requested to be in
                req.returnValues().add(new StringValue(nodeInfo.getWantedState().serialize()));
                // Fourth parameter is RPC address of node
                req.returnValues().add(new StringValue(nodeInfo.getRpcAddress() == null ? "" : nodeInfo.getRpcAddress()));
                req.returnRequest();
            } else if (req.methodName().equals("setNodeState")) {
                String slobrokAddress = req.parameters().get(0).asString();
                int lastSlash = slobrokAddress.lastIndexOf('/');
                int nextButLastSlash = slobrokAddress.lastIndexOf('/', lastSlash - 1);
                if (lastSlash == -1 || nextButLastSlash == -1) {
                    throw new IllegalStateException("Invalid slobrok address '" + slobrokAddress + "'.");
                }
                NodeType nodeType = NodeType.get(slobrokAddress.substring(nextButLastSlash + 1, lastSlash));
                Integer nodeIndex = Integer.valueOf(slobrokAddress.substring(lastSlash + 1));
                NodeInfo node = cluster.getNodeInfo(new Node(nodeType, nodeIndex));
                if (node == null)
                    throw new IllegalStateException("Cannot set wanted state of node " + new Node(nodeType, nodeIndex) + ". Index does not correspond to a configured node.");
                NodeState nodeState = NodeState.deserialize(nodeType, req.parameters().get(1).asString());
                if (nodeState.getDescription().equals("") && !nodeState.getState().equals(State.UP) && !nodeState.getState().equals(State.RETIRED)) {
                    nodeState.setDescription("Set by remote RPC client");
                }
                NodeState oldState = node.getUserWantedState();
                String message = (nodeState.getState().equals(State.UP) ? "Clearing wanted nodeState for node " + node : "New wantedstate '" + nodeState.toString() + "' stored for node " + node);
                if (!oldState.equals(nodeState) || !oldState.getDescription().equals(nodeState.getDescription())) {
                    if (!nodeState.getState().validWantedNodeState(nodeType)) {
                        throw new IllegalStateException("State " + nodeState.getState() + " can not be used as wanted state for node of type " + nodeType);
                    }
                    node.setWantedState(nodeState);
                    changeListener.handleNewWantedNodeState(node, nodeState);
                } else {
                    message = "Node " + node + " already had wanted state " + nodeState.toString();
                    log.log(LogLevel.DEBUG, message);
                }
                req.returnValues().add(new StringValue(message));
                req.returnRequest();
                if (nodeState.getState() == State.UP && node.getPrematureCrashCount() > 0) {
                    log.log(LogLevel.INFO, "Clearing premature crash count of " + node.getPrematureCrashCount() + " as wanted state was set to up");
                    node.setPrematureCrashCount(0);
                }
            }
        } catch (Exception e) {
            if (log.isLoggable(LogLevel.DEBUG)) {
                StringWriter sw = new StringWriter();
                e.printStackTrace(new PrintWriter(sw));
                log.log(LogLevel.DEBUG, "Failed RPC Request: " + sw);
            }
            String errorMsg = e.getMessage();
            if (errorMsg == null) {
                errorMsg = e.toString();
            }
            req.setError(ErrorCode.METHOD_FAILED, errorMsg);
            req.returnRequest();
        }
    }
    return handledAnyRequests;
}
Also used : NodeState(com.yahoo.vdslib.state.NodeState) Node(com.yahoo.vdslib.state.Node) Request(com.yahoo.jrt.Request) ArrayList(java.util.ArrayList) ListenFailedException(com.yahoo.jrt.ListenFailedException) UnknownHostException(java.net.UnknownHostException) ListenFailedException(com.yahoo.jrt.ListenFailedException) StringArray(com.yahoo.jrt.StringArray) StringWriter(java.io.StringWriter) NodeInfo(com.yahoo.vespa.clustercontroller.core.NodeInfo) NodeType(com.yahoo.vdslib.state.NodeType) Int32Value(com.yahoo.jrt.Int32Value) StringValue(com.yahoo.jrt.StringValue) PrintWriter(java.io.PrintWriter)

Example 62 with Node

use of com.yahoo.vdslib.state.Node in project vespa by vespa-engine.

the class SetNodeStateRequest method setDistributorWantedState.

/**
 * Set the wanted state on the distributor to something appropriate given the storage is being
 * set to (or is equal to) newStorageWantedState.
 */
private static void setDistributorWantedState(ContentCluster cluster, int index, NodeState newStorageWantedState, NodeStateOrHostInfoChangeHandler stateListener) {
    Node distributorNode = new Node(NodeType.DISTRIBUTOR, index);
    NodeInfo nodeInfo = cluster.getNodeInfo(distributorNode);
    if (nodeInfo == null) {
        throw new IllegalStateException("Missing distributor at index " + distributorNode.getIndex());
    }
    State newState;
    switch(newStorageWantedState.getState()) {
        case MAINTENANCE:
            newState = State.DOWN;
            break;
        case RETIRED:
            newState = State.UP;
            break;
        default:
            newState = newStorageWantedState.getState();
            if (!newState.validWantedNodeState(distributorNode.getType())) {
                throw new IllegalStateException("Distributor cannot be set to wanted state " + newState);
            }
    }
    NodeState newWantedState = new NodeState(distributorNode.getType(), newState);
    newWantedState.setDescription(newStorageWantedState.getDescription());
    NodeState currentWantedState = nodeInfo.getUserWantedState();
    if (newWantedState.getState() != currentWantedState.getState() || !Objects.equals(newWantedState.getDescription(), currentWantedState.getDescription())) {
        setNewWantedState(nodeInfo, newWantedState, stateListener);
    }
}
Also used : NodeState(com.yahoo.vdslib.state.NodeState) NodeInfo(com.yahoo.vespa.clustercontroller.core.NodeInfo) UnitState(com.yahoo.vespa.clustercontroller.utils.staterestapi.response.UnitState) ClusterState(com.yahoo.vdslib.state.ClusterState) NodeState(com.yahoo.vdslib.state.NodeState) State(com.yahoo.vdslib.state.State) Node(com.yahoo.vdslib.state.Node)

Example 63 with Node

use of com.yahoo.vdslib.state.Node in project vespa by vespa-engine.

the class NodeStateChangeChecker method contentNodesWithAvailableNodeState.

private int contentNodesWithAvailableNodeState(ClusterState clusterState) {
    final int nodeCount = clusterState.getNodeCount(NodeType.STORAGE);
    int upNodesCount = 0;
    for (int i = 0; i < nodeCount; ++i) {
        final Node node = new Node(NodeType.STORAGE, i);
        final State state = clusterState.getNodeState(node).getState();
        if (state == State.UP || state == State.RETIRED || state == State.INITIALIZING) {
            upNodesCount++;
        }
    }
    return upNodesCount;
}
Also used : ClusterState(com.yahoo.vdslib.state.ClusterState) NodeState(com.yahoo.vdslib.state.NodeState) State(com.yahoo.vdslib.state.State) StorageNode(com.yahoo.vespa.clustercontroller.core.hostinfo.StorageNode) Node(com.yahoo.vdslib.state.Node)

Example 64 with Node

use of com.yahoo.vdslib.state.Node in project vespa by vespa-engine.

the class ClusterStateGenerator method generatedStateFrom.

static AnnotatedClusterState generatedStateFrom(final Params params) {
    final ContentCluster cluster = params.cluster;
    final ClusterState workingState = ClusterState.emptyState();
    final Map<Node, NodeStateReason> nodeStateReasons = new HashMap<>();
    for (final NodeInfo nodeInfo : cluster.getNodeInfo()) {
        final NodeState nodeState = computeEffectiveNodeState(nodeInfo, params);
        workingState.setNodeState(nodeInfo.getNode(), nodeState);
    }
    takeDownGroupsWithTooLowAvailability(workingState, nodeStateReasons, params);
    final Optional<ClusterStateReason> reasonToBeDown = clusterDownReason(workingState, params);
    if (reasonToBeDown.isPresent()) {
        workingState.setClusterState(State.DOWN);
    }
    workingState.setDistributionBits(inferDistributionBitCount(cluster, workingState, params));
    return new AnnotatedClusterState(workingState, reasonToBeDown, nodeStateReasons);
}
Also used : ClusterState(com.yahoo.vdslib.state.ClusterState) NodeState(com.yahoo.vdslib.state.NodeState) HashMap(java.util.HashMap) Node(com.yahoo.vdslib.state.Node)

Example 65 with Node

use of com.yahoo.vdslib.state.Node in project vespa by vespa-engine.

the class ClusterStateGenerator method takeDownGroupsWithTooLowAvailability.

private static void takeDownGroupsWithTooLowAvailability(final ClusterState workingState, Map<Node, NodeStateReason> nodeStateReasons, final Params params) {
    final GroupAvailabilityCalculator calc = new GroupAvailabilityCalculator.Builder().withMinNodeRatioPerGroup(params.minNodeRatioPerGroup).withDistribution(params.cluster.getDistribution()).build();
    final Set<Integer> nodesToTakeDown = calc.nodesThatShouldBeDown(workingState);
    for (Integer idx : nodesToTakeDown) {
        final Node node = storageNode(idx);
        final NodeState newState = new NodeState(NodeType.STORAGE, State.DOWN);
        newState.setDescription("group node availability below configured threshold");
        workingState.setNodeState(node, newState);
        nodeStateReasons.put(node, NodeStateReason.GROUP_IS_DOWN);
    }
}
Also used : NodeState(com.yahoo.vdslib.state.NodeState) Node(com.yahoo.vdslib.state.Node)

Aggregations

Node (com.yahoo.vdslib.state.Node)65 Test (org.junit.Test)34 NodeState (com.yahoo.vdslib.state.NodeState)32 ConfiguredNode (com.yahoo.vdslib.distribution.ConfiguredNode)31 ClusterFixture.storageNode (com.yahoo.vespa.clustercontroller.core.ClusterFixture.storageNode)15 HasStateReasonForNode.hasStateReasonForNode (com.yahoo.vespa.clustercontroller.core.matchers.HasStateReasonForNode.hasStateReasonForNode)15 ClusterState (com.yahoo.vdslib.state.ClusterState)9 NodeInfo (com.yahoo.vespa.clustercontroller.core.NodeInfo)6 NodeType (com.yahoo.vdslib.state.NodeType)5 PrintWriter (java.io.PrintWriter)5 StringWriter (java.io.StringWriter)5 Request (com.yahoo.jrt.Request)4 State (com.yahoo.vdslib.state.State)4 IOException (java.io.IOException)4 Spec (com.yahoo.jrt.Spec)3 StringValue (com.yahoo.jrt.StringValue)3 Supervisor (com.yahoo.jrt.Supervisor)3 Target (com.yahoo.jrt.Target)3 Transport (com.yahoo.jrt.Transport)3 TreeMap (java.util.TreeMap)3