Examples with MemberHeartbeatData com.torodb.mongodb.commands.pojos.MemberHeartbeatData used on opensource projects

Example 1 with MemberHeartbeatData

use of com.torodb.mongodb.commands.pojos.MemberHeartbeatData in project torodb by torodb.

the class TopologyCoordinator method lookForSyncSource.

/**
   * Looks for an optimal sync source to replicate from.
   *
   * The first attempt, we ignore those nodes with slave delay higher than our own, hidden nodes,
   * and nodes that are excessively lagged. The second attempt includes such nodes, in case those
   * are the only ones we can reach. This loop attempts to set 'closestIndex'.
   *
   * @param now              the current time
   * @param lastOpAppliedOp  the last OpTime this node has apply
   * @param onlyOptimal      if true, slaves with more delay than ourselve, hidden nodes or
   *                         excessively lagged nodes are ignored
   * @param oldestSyncOpTime the oldest optime considered not excessively lagged. Only used if
   *                         onlyOptimal is true.
   * @return the new optimal sync source, which is not {@link Optional#isPresent() present} if no
   *         one can be chosen
   */
private Optional<MemberConfig> lookForSyncSource(Instant now, Optional<OpTime> lastOpAppliedOp, boolean onlyOptimal, OpTime oldestSyncOpTime) {
    OpTime lastOpApplied = lastOpAppliedOp.orElse(OpTime.EPOCH);
    Stream<MemberHeartbeatData> hbCandidateStream = _hbdata.stream().filter(MemberHeartbeatData::isUp).filter(hbData -> hbData.getState().isReadable()).filter(hbData -> hbData.getOpTime().isAfter(lastOpApplied));
    if (onlyOptimal) {
        hbCandidateStream = hbCandidateStream.filter(hbData -> hbData.getOpTime().isEqualOrAfter(oldestSyncOpTime));
    }
    Stream<MemberConfig> mcCandidateStream = hbCandidateStream.map(this::getMemberConfig).filter(mc -> !isBlacklistedMember(mc, now));
    if (onlyOptimal) {
        mcCandidateStream = mcCandidateStream.filter(mc -> !mc.isHidden()).filter(mc -> mc.getSlaveDelay() < slaveDelaySecs);
    }
    //If there are several candidates, the one whose ping is lower is returned
    return mcCandidateStream.reduce((MemberConfig cand1, MemberConfig cand2) -> {
        long ping1 = getPing(cand1.getHostAndPort());
        long ping2 = getPing(cand2.getHostAndPort());
        if (ping1 < ping2) {
            return cand1;
        }
        return cand2;
    });
}

Example 2 with MemberHeartbeatData

use of com.torodb.mongodb.commands.pojos.MemberHeartbeatData in project torodb by torodb.

the class TopologyCoordinator method processHeartbeatResponse.

/**
   * Processes a heartbeat response from "target" that arrived around "now", having spent
   * "networkRoundTripTime" millis on the network.
   * <p>
   * Updates internal topology coordinator state, and returns instructions about what action to take
   * next.
   * <p>
   * If the next action is {@link HeartbeatResponseAction#makeNoAction() "NoAction"} then nothing
   * has to be done.
   * <p>
   * If the next action indicates {@link HeartbeatResponseAction#makeReconfigAction() "Reconfig"},
   * the caller should verify the configuration in hbResponse is acceptable, perform any other
   * reconfiguration actions it must, and call
   * {@link #updateConfig(
   * com.eightkdata.mongowp.mongoserver.api.safe.library.v3m0.pojos.ReplicaSetConfig,
   * java.time.Instant, com.eightkdata.mongowp.OpTime) updateConfig}
   * with the appropiate arguments.
   * <p>
   * This call should be paired (with intervening network communication) with a call to
   * prepareHeartbeatRequest for the same "target".
   *
   * @param now                  the aproximated time when the response has been recived
   * @param networkRoundTripTime the time spent on network
   * @param target               the host that send the respond
   * @param hbResponse
   */
HeartbeatResponseAction processHeartbeatResponse(Instant now, Duration networkRoundTripTime, HostAndPort target, RemoteCommandResponse<ReplSetHeartbeatReply> hbResponse) {
    PingStats hbStats = getPingOrDefault(target);
    Preconditions.checkState(hbStats.getLastHeartbeatStartDate() != null, "It seems that a hb " + "response has been recived before it has been prepared");
    if (!hbResponse.isOk()) {
        hbStats.miss();
    } else {
        hbStats.hit(networkRoundTripTime);
    }
    boolean isUnauthorized = (hbResponse.getErrorCode() == ErrorCode.UNAUTHORIZED) || (hbResponse.getErrorCode() == ErrorCode.AUTHENTICATION_FAILED);
    Duration alreadyElapsed = Duration.between(hbStats.getLastHeartbeatStartDate(), now);
    Duration nextHeartbeatDelay;
    // determine next start time
    if (_rsConfig != null && (hbStats.getNumFailuresSinceLastStart() <= MAX_HEARTBEAT_RETRIES) && (alreadyElapsed.toMillis() < _rsConfig.getHeartbeatTimeoutPeriod())) {
        if (isUnauthorized) {
            nextHeartbeatDelay = HEARTBEAT_INTERVAL;
        } else {
            nextHeartbeatDelay = Duration.ZERO;
        }
    } else {
        nextHeartbeatDelay = HEARTBEAT_INTERVAL;
    }
    Optional<ReplSetHeartbeatReply> commandReply = hbResponse.getCommandReply();
    if (hbResponse.isOk() && commandReply.get().getConfig().isPresent()) {
        long currentConfigVersion = _rsConfig != null ? _rsConfig.getConfigVersion() : -2;
        ReplicaSetConfig newConfig = commandReply.get().getConfig().get();
        assert newConfig != null;
        if (newConfig.getConfigVersion() > currentConfigVersion) {
            HeartbeatResponseAction nextAction = HeartbeatResponseAction.makeReconfigAction().setNextHeartbeatDelay(nextHeartbeatDelay);
            return nextAction;
        } else {
            // target erroneously sent us one, even through it isn't newer.
            if (newConfig.getConfigVersion() < currentConfigVersion) {
                LOGGER.debug("Config version from heartbeat was older than ours.");
                LOGGER.trace("Current config: {}. Config from heartbeat: {}", _rsConfig, newConfig);
            } else {
                LOGGER.trace("Config from heartbeat response was same as ours.");
            }
        }
    }
    // so return early.
    if (_rsConfig == null) {
        HeartbeatResponseAction nextAction = HeartbeatResponseAction.makeNoAction();
        nextAction.setNextHeartbeatDelay(nextHeartbeatDelay);
        return nextAction;
    }
    OptionalInt memberIndexOpt = _rsConfig.findMemberIndexByHostAndPort(target);
    if (!memberIndexOpt.isPresent()) {
        LOGGER.debug("replset: Could not find {} in current config so ignoring --" + " current config: {}", target, _rsConfig);
        HeartbeatResponseAction nextAction = HeartbeatResponseAction.makeNoAction();
        nextAction.setNextHeartbeatDelay(nextHeartbeatDelay);
        return nextAction;
    }
    assert memberIndexOpt.isPresent();
    int memberIndex = memberIndexOpt.getAsInt();
    MemberHeartbeatData hbData = _hbdata.get(memberIndex);
    assert hbData != null;
    MemberConfig member = _rsConfig.getMembers().get(memberIndex);
    if (!hbResponse.isOk()) {
        if (isUnauthorized) {
            LOGGER.debug("setAuthIssue: heartbeat response failed due to authentication" + " issue for member _id: {}", member.getId());
            hbData.setAuthIssue(now);
        } else if (hbStats.getNumFailuresSinceLastStart() > MAX_HEARTBEAT_RETRIES || alreadyElapsed.toMillis() >= _rsConfig.getHeartbeatTimeoutPeriod()) {
            LOGGER.debug("setDownValues: heartbeat response failed for member _id:{}" + ", msg: {}", member.getId(), hbResponse.getErrorDesc());
            hbData.setDownValues(now, hbResponse.getErrorDesc());
        } else {
            LOGGER.trace("Bad heartbeat response from {}; trying again; Retries left: {}; " + "{} ms have already elapsed", target, MAX_HEARTBEAT_RETRIES - hbStats.getNumFailuresSinceLastStart(), alreadyElapsed.toMillis());
        }
    } else {
        ReplSetHeartbeatReply nonNullReply = commandReply.get();
        LOGGER.trace("setUpValues: heartbeat response good for member _id:{}, msg:  {}", member.getId(), nonNullReply.getHbmsg());
        hbData.setUpValues(now, member.getHostAndPort(), nonNullReply);
    }
    HeartbeatResponseAction nextAction = updateHeartbeatDataImpl(memberIndex, now);
    nextAction.setNextHeartbeatDelay(nextHeartbeatDelay);
    return nextAction;
}

Example 3 with MemberHeartbeatData

use of com.torodb.mongodb.commands.pojos.MemberHeartbeatData in project torodb by torodb.

the class TopologyCoordinator method updateHeartbeatDataImpl.

/**
   * Performs updating {@link #_hbdata} and {@link #_currentPrimaryIndex} for
   * {@link #processHeartbeatResponse(org.threeten.bp.Instant, org.threeten.bp.Duration,
   * com.google.common.net.HostAndPort,
   * com.eightkdata.mongowp.client.core.MongoConnection.RemoteCommandResponse,
   * com.eightkdata.mongowp.OpTime) }.
   */
private HeartbeatResponseAction updateHeartbeatDataImpl(int updatedConfigIndex, Instant now) {
    // the updated data supports that notion.  If not, erase our notion of who is primary.
    if (updatedConfigIndex == _currentPrimaryIndex) {
        final MemberHeartbeatData updatedHbData = _hbdata.get(updatedConfigIndex);
        assert updatedHbData != null;
        if (!updatedHbData.isUp() || updatedHbData.getState() != MemberState.RS_PRIMARY) {
            _currentPrimaryIndex = -1;
        }
    }
    HeartbeatResponseAction newAction;
    newAction = ifTwoPrimariesChecks(now);
    if (newAction != null) {
        return newAction;
    }
    // We do not believe that any remote is primary.
    assert _hbdata.stream().noneMatch(input -> input.isUp() && input.getState() == MemberState.RS_PRIMARY);
    assert _currentPrimaryIndex == -1;
    return HeartbeatResponseAction.makeNoAction();
}

Example 4 with MemberHeartbeatData

use of com.torodb.mongodb.commands.pojos.MemberHeartbeatData in project torodb by torodb.

the class TopologyCoordinator method getMaybeUpHostAndPorts.

/**
   * Retrieves a vector of HostAndPorts containing all nodes that are neither DOWN.
   */
List<HostAndPort> getMaybeUpHostAndPorts() {
    List<HostAndPort> upHosts = new ArrayList<>(_hbdata.size());
    for (int i = 0; i < _hbdata.size(); i++) {
        MemberHeartbeatData it = _hbdata.get(i);
        if (it.maybeUp()) {
            // skip DOWN nodes
            continue;
        }
        upHosts.add(_rsConfig.getMembers().get(i).getHostAndPort());
    }
    return upHosts;
}

Example 5 with MemberHeartbeatData

use of com.torodb.mongodb.commands.pojos.MemberHeartbeatData in project torodb by torodb.

the class TopologyCoordinator method shouldChangeSyncSource.

/**
   * Determines if a new sync source should be chosen, if a better candidate sync source is
   * available.
   *
   * It returns true if there exists a viable sync source member other than our current source,
   * whose oplog has reached an optime greater than the max sync source lag later than current
   * source's. It can return true in other scenarios (like if {@link #setForceSyncSourceIndex(int) }
   * has been called or if we don't have a current sync source.
   *
   * @param now is used to skip over currently blacklisted sync sources.
   * @return
   */
boolean shouldChangeSyncSource(HostAndPort currentSource, Instant now) {
    // If the user requested a sync source change, return true.
    if (_forceSyncSourceIndex != -1) {
        return true;
    }
    OptionalInt currentMemberIndex = _rsConfig.findMemberIndexByHostAndPort(currentSource);
    if (!currentMemberIndex.isPresent()) {
        return true;
    }
    assert _hbdata.get(currentMemberIndex.getAsInt()) != null;
    OpTime currentOpTime = _hbdata.get(currentMemberIndex.getAsInt()).getOpTime();
    if (currentOpTime == null) {
        // change.
        return false;
    }
    long currentSecs = currentOpTime.getSecs();
    long goalSecs = currentSecs + _maxSyncSourceLagSecs;
    for (int i = 0; i < _hbdata.size(); i++) {
        MemberHeartbeatData it = _hbdata.get(i);
        MemberConfig candidateConfig = _rsConfig.getMembers().get(i);
        OpTime itOpTime = it.getOpTime();
        if (itOpTime != null && it.isUp() && it.getState().isReadable() && !isBlacklistedMember(candidateConfig, now) && goalSecs < itOpTime.getSecs()) {
            LOGGER.info("changing sync target because current sync target's most recent OpTime " + "is {}  which is more than {} seconds behind member {} whose most recent " + "OpTime is {} ", currentOpTime, _maxSyncSourceLagSecs, candidateConfig.getHostAndPort(), itOpTime);
            return true;
        }
    }
    return false;
}