private void checkDataConsistencyWithMaster(
URI availableMasterId,
Master master,
NeoStoreXaDataSource nioneoDataSource,
TransactionIdStore transactionIdStore)
throws IOException {
long myLastCommittedTx = transactionIdStore.getLastCommittingTransactionId();
int myMaster = -1;
long myChecksum = 0;
TransactionMetadataCache.TransactionMetadata metadata =
nioneoDataSource
.getDependencyResolver()
.resolveDependency(LogicalTransactionStore.class)
.getMetadataFor(myLastCommittedTx);
myMaster = metadata.getMasterId();
myChecksum = metadata.getChecksum();
HandshakeResult handshake;
try (Response<HandshakeResult> response =
master.handshake(myLastCommittedTx, nioneoDataSource.getStoreId())) {
handshake = response.response();
requestContextFactory.setEpoch(handshake.epoch());
} catch (BranchedDataException e) {
// Rethrow wrapped in a branched data exception on our side, to clarify where the problem
// originates.
throw new BranchedDataException(
"The database stored on this machine has diverged from that "
+ "of the master. This will be automatically resolved.",
e);
} catch (RuntimeException e) {
// Checked exceptions will be wrapped as the cause if this was a serialized
// server-side exception
if (e.getCause() instanceof MissingLogDataException) {
/*
* This means the master was unable to find a log entry for the txid we just asked. This
* probably means the thing we asked for is too old or too new. Anyway, since it doesn't
* have the tx it is better if we just throw our store away and ask for a new copy. Next
* time around it shouldn't have to even pass from here.
*/
throw new StoreOutOfDateException(
"The master is missing the log required to complete the " + "consistency check",
e.getCause());
}
throw e;
}
if (myMaster != -1
&& (myMaster != handshake.txAuthor() || myChecksum != handshake.txChecksum())) {
String msg =
"The cluster contains two logically different versions of the database.. This will be "
+ "automatically resolved. Details: I (machineId:"
+ config.get(ClusterSettings.server_id)
+ ") think machineId for txId ("
+ myLastCommittedTx
+ ") is "
+ myMaster
+ ", but master (machineId:"
+ getServerId(availableMasterId)
+ ") says that it's "
+ handshake;
throw new BranchedDataException(msg);
}
msgLog.logMessage(
"Master id for last committed tx ok with highestTxId="
+ myLastCommittedTx
+ " with masterId="
+ myMaster,
true);
}
private void checkDataConsistency(
RequestContextFactory requestContextFactory,
NeoStoreXaDataSource nioneoDataSource,
URI masterUri)
throws Throwable {
// Must be called under lock on XaDataSourceManager
LifeSupport checkConsistencyLife = new LifeSupport();
TransactionIdStore txIdStore = null;
try {
MasterClient checkConsistencyMaster =
newMasterClient(masterUri, nioneoDataSource.getStoreId(), checkConsistencyLife);
checkConsistencyLife.start();
console.log("Checking store consistency with master");
txIdStore =
nioneoDataSource.getDependencyResolver().resolveDependency(TransactionIdStore.class);
checkDataConsistencyWithMaster(
masterUri, checkConsistencyMaster, nioneoDataSource, txIdStore);
console.log("Store is consistent");
/*
* Pull updates, since the store seems happy and everything. No matter how far back we are, this is just
* one thread doing the pulling, while the guard is up. This will prevent a race between all transactions
* that may start the moment the database becomes available, where all of them will pull the same txs from
* the master but eventually only one will get to apply them.
*/
console.log("Catching up with master");
resolver
.resolveDependency(TransactionCommittingResponseUnpacker.class)
.unpackResponse(
checkConsistencyMaster.pullUpdates(requestContextFactory.newRequestContext()));
console.log("Now consistent with master");
} catch (NoSuchLogVersionException e) {
msgLog.logMessage(
"Cannot catch up to master by pulling updates, because I cannot find the archived "
+ "logical log file that has the transaction I would start from. I'm going to copy the whole "
+ "store from the master instead.");
try {
stopServicesAndHandleBranchedStore(config.get(HaSettings.branched_data_policy));
} catch (Throwable throwable) {
msgLog.warn("Failed preparing for copying the store from the master instance", throwable);
}
throw e;
} catch (StoreUnableToParticipateInClusterException upe) {
console.log(
"The store is inconsistent. Will treat it as branched and fetch a new one from the master");
msgLog.warn(
"Current store is unable to participate in the cluster; fetching new store from master",
upe);
try {
stopServicesAndHandleBranchedStore(config.get(HaSettings.branched_data_policy));
} catch (IOException e) {
msgLog.warn("Failed while trying to handle branched data", e);
}
throw upe;
} catch (MismatchingStoreIdException e) {
console.log(
"The store does not represent the same database as master. Will remove and fetch a new one from master");
if (txIdStore.getLastCommittingTransactionId() == 0) {
msgLog.warn("Found and deleting empty store with mismatching store id " + e.getMessage());
stopServicesAndHandleBranchedStore(BranchedDataPolicy.keep_none);
} else {
msgLog.error("Store cannot participate in cluster due to mismatching store IDs");
}
throw e;
} finally {
checkConsistencyLife.shutdown();
}
}