|
| 1 | +package org.apache.phoenix.replication; |
| 2 | + |
| 3 | +import static org.apache.phoenix.replication.ReplicationLogGroup.ReplicationMode.STORE_AND_FORWARD; |
| 4 | +import static org.apache.phoenix.replication.ReplicationLogGroup.ReplicationMode.SYNC; |
| 5 | +import static org.apache.phoenix.replication.ReplicationLogGroup.ReplicationMode.SYNC_AND_FORWARD; |
| 6 | + |
| 7 | +import java.io.IOException; |
| 8 | + |
| 9 | +import org.apache.hadoop.fs.FileStatus; |
| 10 | +import org.apache.hadoop.fs.FileSystem; |
| 11 | +import org.apache.hadoop.fs.FileUtil; |
| 12 | +import org.apache.hadoop.fs.Path; |
| 13 | +import org.apache.phoenix.jdbc.ClusterType; |
| 14 | +import org.apache.phoenix.jdbc.HAGroupStateListener; |
| 15 | +import org.apache.phoenix.jdbc.HAGroupStoreManager; |
| 16 | +import org.apache.phoenix.jdbc.HAGroupStoreRecord; |
| 17 | +import org.apache.phoenix.replication.metrics.MetricsReplicationLogDiscovery; |
| 18 | +import org.apache.phoenix.replication.metrics.MetricsReplicationLogForwarderSourceFactory; |
| 19 | +import org.apache.phoenix.thirdparty.com.google.common.annotations.VisibleForTesting; |
| 20 | +import org.apache.phoenix.util.EnvironmentEdgeManager; |
| 21 | +import org.slf4j.Logger; |
| 22 | +import org.slf4j.LoggerFactory; |
| 23 | + |
| 24 | +/** |
| 25 | + * ReplicationLogDiscoveryForwarder manages the forwarding of the replication log |
| 26 | + * from the fallback cluster to the remote cluster. |
| 27 | + */ |
| 28 | +public class ReplicationLogDiscoveryForwarder extends ReplicationLogDiscovery { |
| 29 | + private static final Logger LOG = LoggerFactory.getLogger(ReplicationLogDiscoveryForwarder.class); |
| 30 | + |
| 31 | + private final ReplicationLogGroup logGroup; |
| 32 | + |
| 33 | + /** |
| 34 | + * Create a tracker for the replication logs in the fallback cluster. |
| 35 | + * |
| 36 | + * @param logGroup HAGroup |
| 37 | + * @return ReplicationLogTracker |
| 38 | + */ |
| 39 | + private static ReplicationLogTracker createLogTracker(ReplicationLogGroup logGroup) { |
| 40 | + ReplicationShardDirectoryManager localShardManager = logGroup.getFallbackShardManager(); |
| 41 | + return new ReplicationLogTracker( |
| 42 | + logGroup.conf, |
| 43 | + logGroup.getHAGroupName(), |
| 44 | + localShardManager, |
| 45 | + MetricsReplicationLogForwarderSourceFactory. |
| 46 | + getInstanceForTracker(logGroup.getHAGroupName())); |
| 47 | + } |
| 48 | + |
| 49 | + public ReplicationLogDiscoveryForwarder(ReplicationLogGroup logGroup) { |
| 50 | + super(createLogTracker(logGroup)); |
| 51 | + this.logGroup = logGroup; |
| 52 | + } |
| 53 | + |
| 54 | + @Override |
| 55 | + public String getExecutorThreadNameFormat() { |
| 56 | + return "ReplicationLogDiscoveryForwarder-" + logGroup.getHAGroupName() + "-%d"; |
| 57 | + } |
| 58 | + |
| 59 | + public void init() throws IOException { |
| 60 | + replicationLogTracker.init(); |
| 61 | + // Initialize the discovery only. Forwarding begins only when we switch to the |
| 62 | + // STORE_AND_FORWARD mode or SYNC_AND_FORWARD mode. |
| 63 | + super.init(); |
| 64 | + |
| 65 | + // Set up a listener to the ACTIVE_NOT_IN_SYNC state. This is needed because whenever any |
| 66 | + // RS switches to STORE_AND_FORWARD mode, other RS's in the cluster must move out of SYNC |
| 67 | + // mode. |
| 68 | + HAGroupStateListener activeNotInSync = (groupName, |
| 69 | + fromState, |
| 70 | + toState, |
| 71 | + modifiedTime, |
| 72 | + clusterType, |
| 73 | + lastSyncStateTimeInMs) -> { |
| 74 | + if (clusterType == ClusterType.LOCAL && HAGroupStoreRecord.HAGroupState.ACTIVE_NOT_IN_SYNC.equals(toState)) { |
| 75 | + LOG.info("Received ACTIVE_NOT_IN_SYNC event for {}", logGroup); |
| 76 | + // If the current mode is SYNC only then switch to SYNC_AND_FORWARD mode |
| 77 | + if (logGroup.checkAndSetMode(SYNC, SYNC_AND_FORWARD)) { |
| 78 | + // replication mode switched, notify the event handler |
| 79 | + try { |
| 80 | + logGroup.sync(); |
| 81 | + } catch (IOException e) { |
| 82 | + LOG.info("Failed to send sync event to {}", logGroup); |
| 83 | + } |
| 84 | + } |
| 85 | + } |
| 86 | + }; |
| 87 | + |
| 88 | + // Set up a listener to the ACTIVE_IN_SYNC state. This is needed because when the RS |
| 89 | + // switches back to SYNC mode, the other RS's in the cluster must move out of |
| 90 | + // SYNC_AND_FORWARD mode to SYNC mode. |
| 91 | + HAGroupStateListener activeInSync = (groupName, |
| 92 | + fromState, |
| 93 | + toState, |
| 94 | + modifiedTime, |
| 95 | + clusterType, |
| 96 | + lastSyncStateTimeInMs) -> { |
| 97 | + if (clusterType == ClusterType.LOCAL && HAGroupStoreRecord.HAGroupState.ACTIVE_IN_SYNC.equals(toState)) { |
| 98 | + LOG.info("Received ACTIVE_IN_SYNC event for {}", logGroup); |
| 99 | + // Set the current mode to SYNC |
| 100 | + if (logGroup.checkAndSetMode(SYNC_AND_FORWARD, SYNC)) { |
| 101 | + // replication mode switched, notify the event handler |
| 102 | + try { |
| 103 | + logGroup.sync(); |
| 104 | + } catch (IOException e) { |
| 105 | + LOG.info("Failed to send sync event to {}", logGroup); |
| 106 | + } |
| 107 | + } |
| 108 | + } |
| 109 | + }; |
| 110 | + |
| 111 | + HAGroupStoreManager haGroupStoreManager = logGroup.getHAGroupStoreManager(); |
| 112 | + haGroupStoreManager.subscribeToTargetState(logGroup.getHAGroupName(), |
| 113 | + HAGroupStoreRecord.HAGroupState.ACTIVE_NOT_IN_SYNC, ClusterType.LOCAL, activeNotInSync); |
| 114 | + haGroupStoreManager.subscribeToTargetState(logGroup.getHAGroupName(), |
| 115 | + HAGroupStoreRecord.HAGroupState.ACTIVE_IN_SYNC, ClusterType.LOCAL, activeInSync); |
| 116 | + } |
| 117 | + |
| 118 | + @Override |
| 119 | + protected void processFile(Path src) throws IOException { |
| 120 | + FileSystem srcFS = replicationLogTracker.getFileSystem(); |
| 121 | + FileStatus srcStat = srcFS.getFileStatus(src); |
| 122 | + long ts = EnvironmentEdgeManager.currentTimeMillis(); |
| 123 | + ReplicationShardDirectoryManager remoteShardManager = logGroup.getStandbyShardManager(); |
| 124 | + Path dst = remoteShardManager.getWriterPath(ts, logGroup.getServerName().getServerName()); |
| 125 | + long startTime = EnvironmentEdgeManager.currentTimeMillis(); |
| 126 | + FileUtil.copy(srcFS, srcStat, remoteShardManager.getFileSystem(), dst, false, false, conf); |
| 127 | + // successfully copied the file |
| 128 | + long endTime = EnvironmentEdgeManager.currentTimeMillis(); |
| 129 | + long copyTime = endTime - startTime; |
| 130 | + LOG.info("Copying file src={} dst={} size={} took {}ms", src, dst, srcStat.getLen(), copyTime); |
| 131 | + if (logGroup.getMode() == STORE_AND_FORWARD && |
| 132 | + isLogCopyThroughputAboveThreshold(srcStat.getLen(), copyTime)) { |
| 133 | + // start recovery by switching to SYNC_AND_FORWARD |
| 134 | + if (logGroup.checkAndSetMode(STORE_AND_FORWARD, SYNC_AND_FORWARD)) { |
| 135 | + // replication mode switched, notify the event handler |
| 136 | + try { |
| 137 | + logGroup.sync(); |
| 138 | + } catch (IOException e) { |
| 139 | + LOG.info("Failed to send sync event to {}", logGroup); |
| 140 | + } |
| 141 | + } |
| 142 | + } |
| 143 | + } |
| 144 | + |
| 145 | + @Override |
| 146 | + protected void processNoMoreRoundsLeft() throws IOException { |
| 147 | + // check if we are caught up so that we can transition to SYNC state |
| 148 | + // we are caught up when there are no files currently in the out progress directory |
| 149 | + // and no new files exist for ongoing round |
| 150 | + if (replicationLogTracker.getInProgressFiles().isEmpty() |
| 151 | + && replicationLogTracker.getNewFilesForRound(replicationLogTracker |
| 152 | + .getReplicationShardDirectoryManager() |
| 153 | + .getNextRound(getLastRoundProcessed())).isEmpty()) { |
| 154 | + LOG.info("Processed all the replication log files for {}", logGroup); |
| 155 | + // TODO ensure the mTime on the group store record is older than the wait sync timeout |
| 156 | + logGroup.setHAGroupStatusToSync(); |
| 157 | + } |
| 158 | + } |
| 159 | + |
| 160 | + /** |
| 161 | + * Determine if the throughput is above the configured threshold. If it is, then we can switch |
| 162 | + * to the SYNC_AND_FORWARD mode |
| 163 | + * |
| 164 | + * @param fileSize |
| 165 | + * @param copyTime |
| 166 | + * @return True if the throughput is good else false |
| 167 | + */ |
| 168 | + private boolean isLogCopyThroughputAboveThreshold(long fileSize, long copyTime) { |
| 169 | + // TODO: calculate throughput and check if is above configured threshold |
| 170 | + return true; |
| 171 | + } |
| 172 | + |
| 173 | + @Override |
| 174 | + protected MetricsReplicationLogDiscovery createMetricsSource() { |
| 175 | + return MetricsReplicationLogForwarderSourceFactory. |
| 176 | + getInstanceForDiscovery(logGroup.getHAGroupName()); |
| 177 | + } |
| 178 | + |
| 179 | + @VisibleForTesting |
| 180 | + protected ReplicationLogTracker getReplicationLogTracker() { |
| 181 | + return replicationLogTracker; |
| 182 | + } |
| 183 | +} |
0 commit comments