// check and initialize global configuration
private Configuration checkGlobalConfiguration() {
String configKind = _coordinator.getDbConfigPath(_serviceInfo.getName());
Configuration config =
_coordinator.queryConfiguration(_coordinator.getSiteId(), configKind, Constants.GLOBAL_ID);
if (config == null) {
// check if it is upgraded from previous version to yoda - configuration may be stored in
// znode /config. Since SeedProvider still need access that, so we remove the config
// from global in migration callback after migration is done.
config = _coordinator.queryConfiguration(configKind, Constants.GLOBAL_ID);
if (config != null) {
_log.info("Upgrade from pre-yoda release, move global config to new location");
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), config);
return config;
}
ConfigurationImpl cfg = new ConfigurationImpl();
cfg.setId(Constants.GLOBAL_ID);
cfg.setKind(configKind);
cfg.setConfig(Constants.SCHEMA_VERSION, this._serviceInfo.getVersion());
// persist configuration
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), cfg);
config = cfg;
}
return config;
}
// check and initialize versioned configuration
private Configuration checkVersionedConfiguration() {
String serviceVersion = _serviceInfo.getVersion();
String dbSchemaVersion = _dbClient.getSchemaVersion();
if (!serviceVersion.equals(dbSchemaVersion)) {
_log.warn(
"The db service version {} doesn't equals Db schema version {}, "
+ "set db service version to Db schema version",
serviceVersion,
dbSchemaVersion);
_serviceInfo.setVersion(dbSchemaVersion);
}
String kind =
_coordinator.getVersionedDbConfigPath(_serviceInfo.getName(), _serviceInfo.getVersion());
Configuration config =
_coordinator.queryConfiguration(_coordinator.getSiteId(), kind, _serviceInfo.getId());
if (config == null) {
// check if it is upgraded from previous version to yoda - configuration may be stored in
// znode /config
config = _coordinator.queryConfiguration(kind, _serviceInfo.getId());
if (config != null) {
_log.info("Upgrade from pre-2.5 release, move versioned dbconfig to new location");
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), config);
return config;
}
ConfigurationImpl cfg = new ConfigurationImpl();
cfg.setId(_serviceInfo.getId());
cfg.setKind(kind);
// persist configuration
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), cfg);
config = cfg;
}
return config;
}
public void setConfigValue(String key, String value) {
String configKind = _coordinator.getDbConfigPath(_serviceInfo.getName());
Configuration config =
_coordinator.queryConfiguration(_coordinator.getSiteId(), configKind, _serviceInfo.getId());
if (config != null) {
config.setConfig(key, value);
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), config);
}
}
private void removeStaleVersionedDbConfiguration() {
String configKind =
_coordinator.getVersionedDbConfigPath(_serviceInfo.getName(), _serviceInfo.getVersion());
List<Configuration> configs =
_coordinator.queryAllConfiguration(_coordinator.getSiteId(), configKind);
for (Configuration config : configs) {
if (isStaleConfiguration(config)) {
_coordinator.removeServiceConfiguration(_coordinator.getSiteId(), config);
_log.info("Remove stale version db config, id: {}", config.getId());
}
}
}
/**
* Checks and sets INIT_DONE state this means we are done with the actual cf changes on the
* cassandra side for the target version
*/
private void setDbConfigInitDone() {
String configKind =
_coordinator.getVersionedDbConfigPath(_serviceInfo.getName(), _serviceInfo.getVersion());
Configuration config =
_coordinator.queryConfiguration(_coordinator.getSiteId(), configKind, _serviceInfo.getId());
if (config != null) {
if (config.getConfig(DbConfigConstants.INIT_DONE) == null) {
config.setConfig(DbConfigConstants.INIT_DONE, Boolean.TRUE.toString());
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), config);
}
} else {
// we are expecting this to exist, because its initialized from checkVersionedConfiguration
throw new IllegalStateException("unexpected error, db versioned configuration is null");
}
}
public String getConfigValue(String key) {
String configKind = _coordinator.getDbConfigPath(_serviceInfo.getName());
Configuration config =
_coordinator.queryConfiguration(_coordinator.getSiteId(), configKind, _serviceInfo.getId());
if (config != null) {
return config.getConfig(key);
}
return null;
}
private void removeStaleServiceConfiguration() {
boolean isGeoDBSvc = isGeoDbsvc();
boolean resetAutoBootFlag = false;
String configKind = _coordinator.getDbConfigPath(_serviceInfo.getName());
List<Configuration> configs =
_coordinator.queryAllConfiguration(_coordinator.getSiteId(), configKind);
for (Configuration config : configs) {
if (isStaleConfiguration(config)) {
boolean autoboot = Boolean.parseBoolean(config.getConfig(DbConfigConstants.AUTOBOOT));
String configId = config.getId();
if (isGeoDBSvc && !autoboot && (configId.equals("geodb-4") || configId.equals("geodb-5"))) {
// for geodbsvc, if restore with the backup of 5 nodes to 3 nodes and the backup is made
// on the cluster that the 'autoboot=false' is set on vipr4 or vipr5
// we should set the autoboot=false on the current node or no node with autoboot=false
// TODO:This is a temporary/safest solution in Yoda, we'll provide a better soltuion post
// Yoda
resetAutoBootFlag = true;
}
if (isStaleConfiguration(config)) {
_coordinator.removeServiceConfiguration(_coordinator.getSiteId(), config);
_log.info("Remove stale db config, id: {}", config.getId());
}
}
}
if (resetAutoBootFlag) {
_log.info("set autoboot flag to false on {}", _serviceInfo.getId());
Configuration config =
_coordinator.queryConfiguration(
_coordinator.getSiteId(), configKind, _serviceInfo.getId());
config.setConfig(DbConfigConstants.AUTOBOOT, Boolean.FALSE.toString());
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), config);
}
}
/**
* Checks and registers db configuration information, this is one time when cluster is coming up
* for the first time
*/
private Configuration checkConfiguration() {
String configKind = _coordinator.getDbConfigPath(_serviceInfo.getName());
Configuration config =
_coordinator.queryConfiguration(_coordinator.getSiteId(), configKind, _serviceInfo.getId());
if (config == null) {
// check if it is upgraded from previous version to yoda - configuration may be stored in
// zk global area /config. Since SeedProvider still need access that, so we remove the config
// from global in migration callback after migration is done.
config = _coordinator.queryConfiguration(configKind, _serviceInfo.getId());
if (config != null) {
_log.info("Upgrade from pre-yoda release, move dbconfig to new location");
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), config);
return config;
}
// this is a new node
// 1. register its configuration with coordinator
// 2. assume autobootstrap configuration
// this means that when a node is added, it take 1/2 of biggest token rage and
// copies its data over
ConfigurationImpl cfg = new ConfigurationImpl();
cfg.setId(_serviceInfo.getId());
cfg.setKind(configKind);
cfg.setConfig(DbConfigConstants.NODE_ID, _coordinator.getInetAddessLookupMap().getNodeId());
cfg.setConfig(DbConfigConstants.AUTOBOOT, Boolean.TRUE.toString());
// check other existing db nodes
List<Configuration> configs =
_coordinator.queryAllConfiguration(_coordinator.getSiteId(), configKind);
if (configs.isEmpty()) {
// we are the first node - turn off autobootstrap
cfg.setConfig(DbConfigConstants.AUTOBOOT, Boolean.FALSE.toString());
}
// persist configuration
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), cfg);
config = cfg;
}
return config;
}
/** Check offline event info to see if dbsvc/geodbsvc on this node could get started */
private void checkDBOfflineInfo() {
Configuration config =
_coordinator.queryConfiguration(
_coordinator.getSiteId(), Constants.DB_DOWNTIME_TRACKER_CONFIG, _serviceInfo.getName());
DbOfflineEventInfo dbOfflineEventInfo = new DbOfflineEventInfo(config);
String localNodeId = _coordinator.getInetAddessLookupMap().getNodeId();
Long lastActiveTimestamp = dbOfflineEventInfo.geLastActiveTimestamp(localNodeId);
long zkTimeStamp =
(lastActiveTimestamp == null) ? TimeUtils.getCurrentTime() : lastActiveTimestamp;
File localDbDir = new File(dbDir);
Date lastModified = getLastModified(localDbDir);
boolean isDirEmpty = lastModified == null || localDbDir.list().length == 0;
long localTimeStamp = (isDirEmpty) ? TimeUtils.getCurrentTime() : lastModified.getTime();
_log.info("Service timestamp in ZK is {}, local file is: {}", zkTimeStamp, localTimeStamp);
long diffTime = (zkTimeStamp > localTimeStamp) ? (zkTimeStamp - localTimeStamp) : 0;
if (diffTime >= MAX_SERVICE_OUTAGE_TIME) {
String errMsg =
String.format(
"We detect database files on local disk are more than %s days older "
+ "than last time it was seen in the cluster. It may bring stale data into the database, "
+ "so the service cannot continue to boot. It may be the result of a VM snapshot rollback. "
+ "Please contact with EMC support engineer for solution.",
diffTime / TimeUtils.DAYS);
alertLog.error(errMsg);
throw new IllegalStateException(errMsg);
}
Long offlineTime = dbOfflineEventInfo.getOfflineTimeInMS(localNodeId);
if (!isDirEmpty && offlineTime != null && offlineTime >= MAX_SERVICE_OUTAGE_TIME) {
String errMsg =
String.format(
"This node is offline for more than %s days. It may bring stale data into "
+ "database, so the service cannot continue to boot. Please poweroff this node and follow our "
+ "node recovery procedure to recover this node",
offlineTime / TimeUtils.DAYS);
alertLog.error(errMsg);
throw new IllegalStateException(errMsg);
}
}
@Override
public void start() throws IOException {
if (_log.isInfoEnabled()) {
_log.info("Starting DB service...");
}
// Suppress Sonar violation of Lazy initialization of static fields should be synchronized
// start() method will be only called one time when startup dbsvc, so it's safe to ignore sonar
// violation
instance = this; // NOSONAR ("squid:S2444")
if (backCompatPreYoda) {
_log.info(
"Pre-yoda back compatible flag detected. Initialize local keystore/truststore for Cassandra native encryption");
initKeystoreAndTruststore();
_schemaUtil.setBackCompatPreYoda(true);
}
System.setProperty("cassandra.config", _config);
System.setProperty("cassandra.config.loader", CassandraConfigLoader.class.getName());
// Set to false to clear all gossip state for the node on restart.
//
// We encounter a weird Cassandra grossip issue(COP-19246) - some nodes are missing from gossip
// when rebooting the entire cluster simultaneously. Critical Gossip
// fields(ApplicationState.STATUS, ApplicationState.TOKENS)
// are not synchronized during handshaking. It looks like some problem caused by incorrect
// gossip version/generation
// at system local table. So add this option to cleanup local gossip state during reboot
//
// Make sure add-vdc/add-standby passed when you would remove this option in the future.
//
// We need make sure majority local nodes are added as seed nodes. Otherwise cassandra may not
// see other nodes if it loses
// connection to other sites
System.setProperty("cassandra.load_ring_state", "false");
// Nodes in new data center should not auto-bootstrap.
// See
// https://docs.datastax.com/en/cassandra/2.0/cassandra/operations/ops_add_dc_to_cluster_t.html
if (_schemaUtil.isStandby()) {
System.setProperty("cassandra.auto_bootstrap", "false");
}
InterProcessLock lock = null;
Configuration config = null;
StartupMode mode = null;
try {
// we use this lock to discourage more than one node bootstrapping / joining at the same time
// Cassandra can handle this but it's generally not recommended to make changes to schema
// concurrently
lock = getLock(getSchemaLockName());
config = checkConfiguration();
checkGlobalConfiguration();
checkVersionedConfiguration();
removeStaleConfiguration();
mode = checkStartupMode(config);
_log.info("Current startup mode is {}", mode);
// Check if service is allowed to get started by querying db offline info to avoid bringing
// back stale data.
// Skipping hibernate mode for node recovery procedure to recover the overdue node.
int nodeCount = ((CoordinatorClientImpl) _coordinator).getNodeCount();
if (nodeCount != 1 && mode.type != StartupMode.StartupModeType.HIBERNATE_MODE) {
checkDBOfflineInfo();
}
// this call causes instantiation of a seed provider instance, so the check*Configuration
// calls must be preceed it
removeCassandraSavedCaches();
mode.onPreStart();
if (_jmxServer != null) {
_jmxServer.start();
System.setProperty(
"com.sun.management.jmxremote.port", Integer.toString(_jmxServer.getPort()));
}
_service = new CassandraDaemon();
_service.init(null);
_service.start();
cassandraInitialized = true;
mode.onPostStart();
} catch (Exception e) {
if (mode != null && mode.type == StartupMode.StartupModeType.HIBERNATE_MODE) {
printRecoveryWorkAround(e);
}
_log.error("e=", e);
throw new IllegalStateException(e);
} finally {
if (lock != null) {
try {
lock.release();
} catch (Exception ignore) {
_log.debug("lock release failed");
}
}
}
if (config.getConfig(DbConfigConstants.JOINED) == null) {
config.setConfig(DbConfigConstants.JOINED, Boolean.TRUE.toString());
_coordinator.persistServiceConfiguration(_coordinator.getSiteId(), config);
}
_statusChecker.waitForAllNodesJoined();
_svcBeacon.start();
if (backCompatPreYoda) {
_log.info("Enable duplicated beacon in global area during pre-yoda upgrade");
startDupBeacon();
}
setDbInitializedFlag();
setDbConfigInitDone();
_dbClient.start();
if (_schemaUtil.isStandby()) {
String localDataRevision = getLocalDataRevision();
if (localDataRevision != null) {
_schemaUtil.checkDataRevision(localDataRevision);
}
}
// Setup the vdc information, so that login enabled before migration
if (!isGeoDbsvc()) {
_schemaUtil.checkAndSetupBootStrapInfo(_dbClient);
}
dbMgr.init();
if (_handler.run()) {
// Setup the bootstrap info root tenant, if root tenant migrated from local db, then skip it
if (isGeoDbsvc()) {
_schemaUtil.checkAndSetupBootStrapInfo(_dbClient);
} else {
_schemaUtil.checkAndInitStorageSystemTypes(_dbClient);
}
startBackgroundTasks();
_log.info("DB service started");
} else {
_log.error("DB migration failed. Skipping starting background tasks.");
}
}