use of org.apache.cassandra.db.Keyspace in project cassandra by apache.
the class CompactionsTest method testNeedsCleanup.
@Test
@Ignore("making ranges based on the keys, not on the tokens")
public void testNeedsCleanup() {
Keyspace keyspace = Keyspace.open(KEYSPACE1);
ColumnFamilyStore store = keyspace.getColumnFamilyStore("CF_STANDARD1");
store.clearUnsafe();
// disable compaction while flushing
store.disableAutoCompaction();
// 201, 202, ... 208, 209
for (int i = 1; i < 10; i++) {
insertRowWithKey(i);
insertRowWithKey(i + 100);
insertRowWithKey(i + 200);
}
store.forceBlockingFlush();
assertEquals(1, store.getLiveSSTables().size());
SSTableReader sstable = store.getLiveSSTables().iterator().next();
// contiguous range spans all data
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 209)));
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 210)));
// separate ranges span all data
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 100, 109, 200, 209)));
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 109, 200, 210)));
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 100, 210)));
// one range is missing completely
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(100, 109, 200, 209)));
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 200, 209)));
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 100, 109)));
// the beginning of one range is missing
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(1, 9, 100, 109, 200, 209)));
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 101, 109, 200, 209)));
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 100, 109, 201, 209)));
// the end of one range is missing
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(0, 8, 100, 109, 200, 209)));
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 100, 108, 200, 209)));
assertTrue(CompactionManager.needsCleanup(sstable, makeRanges(0, 9, 100, 109, 200, 208)));
// some ranges don't contain any data
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 0, 0, 9, 50, 51, 100, 109, 150, 199, 200, 209, 300, 301)));
// same case, but with a middle range not covering some of the existing data
assertFalse(CompactionManager.needsCleanup(sstable, makeRanges(0, 0, 0, 9, 50, 51, 100, 103, 150, 199, 200, 209, 300, 301)));
}
use of org.apache.cassandra.db.Keyspace in project cassandra by apache.
the class CommitLogReplayer method blockForWrites.
/**
* Flushes all keyspaces associated with this replayer in parallel, blocking until their flushes are complete.
* @return the number of mutations replayed
*/
public int blockForWrites() {
for (Map.Entry<TableId, AtomicInteger> entry : commitLogReader.getInvalidMutations()) logger.warn("Skipped {} mutations from unknown (probably removed) CF with id {}", entry.getValue(), entry.getKey());
// wait for all the writes to finish on the mutation stage
FBUtilities.waitOnFutures(futures);
logger.trace("Finished waiting on mutations from recovery");
// flush replayed keyspaces
futures.clear();
boolean flushingSystem = false;
List<Future<?>> futures = new ArrayList<Future<?>>();
for (Keyspace keyspace : keyspacesReplayed) {
if (keyspace.getName().equals(SchemaConstants.SYSTEM_KEYSPACE_NAME))
flushingSystem = true;
futures.addAll(keyspace.flush());
}
// also flush batchlog incase of any MV updates
if (!flushingSystem)
futures.add(Keyspace.open(SchemaConstants.SYSTEM_KEYSPACE_NAME).getColumnFamilyStore(SystemKeyspace.BATCHES).forceFlush());
FBUtilities.waitOnFutures(futures);
return replayedCount.get();
}
use of org.apache.cassandra.db.Keyspace in project cassandra by apache.
the class IndexSummaryManager method getRestributionTransactions.
/**
* Marks the non-compacting sstables as compacting for index summary redistribution for all keyspaces/tables.
*
* @return Pair containing:
* left: total size of the off heap index summaries for the sstables we were unable to mark compacting (they were involved in other compactions)
* right: the transactions, keyed by table id.
*/
@SuppressWarnings("resource")
private Pair<Long, Map<TableId, LifecycleTransaction>> getRestributionTransactions() {
List<SSTableReader> allCompacting = new ArrayList<>();
Map<TableId, LifecycleTransaction> allNonCompacting = new HashMap<>();
for (Keyspace ks : Keyspace.all()) {
for (ColumnFamilyStore cfStore : ks.getColumnFamilyStores()) {
Set<SSTableReader> nonCompacting, allSSTables;
LifecycleTransaction txn;
do {
View view = cfStore.getTracker().getView();
allSSTables = ImmutableSet.copyOf(view.select(SSTableSet.CANONICAL));
nonCompacting = ImmutableSet.copyOf(view.getUncompacting(allSSTables));
} while (null == (txn = cfStore.getTracker().tryModify(nonCompacting, OperationType.INDEX_SUMMARY)));
allNonCompacting.put(cfStore.metadata.id, txn);
allCompacting.addAll(Sets.difference(allSSTables, nonCompacting));
}
}
long nonRedistributingOffHeapSize = allCompacting.stream().mapToLong(SSTableReader::getIndexSummaryOffHeapSize).sum();
return Pair.create(nonRedistributingOffHeapSize, allNonCompacting);
}
use of org.apache.cassandra.db.Keyspace in project cassandra by apache.
the class StorageProxy method doPaxos.
/**
* Performs the Paxos rounds for a given proposal, retrying when preempted until the timeout.
*
* <p>The main 'configurable' of this method is the {@code createUpdateProposal} method: it is called by the method
* once a ballot has been successfully 'prepared' to generate the update to 'propose' (and commit if the proposal is
* successful). That method also generates the result that the whole method will return. Note that due to retrying,
* this method may be called multiple times and does not have to return the same results.
*
* @param metadata the table to update with Paxos.
* @param key the partition updated.
* @param consistencyForPaxos the serial consistency of the operation (either {@link ConsistencyLevel#SERIAL} or
* {@link ConsistencyLevel#LOCAL_SERIAL}).
* @param consistencyForReplayCommits the consistency for the commit phase of "replayed" in-progress operations.
* @param consistencyForCommit the consistency for the commit phase of _this_ operation update.
* @param queryStartNanoTime the nano time for the start of the query this is part of. This is the base time for
* timeouts.
* @param casMetrics the metrics to update for this operation.
* @param createUpdateProposal method called after a successful 'prepare' phase to obtain 1) the actual update of
* this operation and 2) the result that the whole method should return. This can return {@code null} in the
* special where, after having "prepared" (and thus potentially replayed in-progress upgdates), we don't want
* to propose anything (the whole method then return {@code null}).
* @return the second element of the pair returned by {@code createUpdateProposal} (for the last call of that method
* if that method is called multiple times due to retries).
*/
private static RowIterator doPaxos(TableMetadata metadata, DecoratedKey key, ConsistencyLevel consistencyForPaxos, ConsistencyLevel consistencyForReplayCommits, ConsistencyLevel consistencyForCommit, long queryStartNanoTime, CASClientRequestMetrics casMetrics, Supplier<Pair<PartitionUpdate, RowIterator>> createUpdateProposal) throws UnavailableException, IsBootstrappingException, RequestFailureException, RequestTimeoutException, InvalidRequestException {
int contentions = 0;
Keyspace keyspace = Keyspace.open(metadata.keyspace);
AbstractReplicationStrategy latestRs = keyspace.getReplicationStrategy();
try {
consistencyForPaxos.validateForCas();
consistencyForReplayCommits.validateForCasCommit(latestRs);
consistencyForCommit.validateForCasCommit(latestRs);
long timeoutNanos = DatabaseDescriptor.getCasContentionTimeout(NANOSECONDS);
while (nanoTime() - queryStartNanoTime < timeoutNanos) {
// for simplicity, we'll do a single liveness check at the start of each attempt
ReplicaPlan.ForPaxosWrite replicaPlan = ReplicaPlans.forPaxos(keyspace, key, consistencyForPaxos);
latestRs = replicaPlan.replicationStrategy();
PaxosBallotAndContention pair = beginAndRepairPaxos(queryStartNanoTime, key, metadata, replicaPlan, consistencyForPaxos, consistencyForReplayCommits, casMetrics);
final UUID ballot = pair.ballot;
contentions += pair.contentions;
Pair<PartitionUpdate, RowIterator> proposalPair = createUpdateProposal.get();
// See method javadoc: null here is code for "stop here and return null".
if (proposalPair == null)
return null;
Commit proposal = Commit.newProposal(ballot, proposalPair.left);
Tracing.trace("CAS precondition is met; proposing client-requested updates for {}", ballot);
if (proposePaxos(proposal, replicaPlan, true, queryStartNanoTime)) {
// them), this is worth bothering.
if (!proposal.update.isEmpty())
commitPaxos(proposal, consistencyForCommit, true, queryStartNanoTime);
RowIterator result = proposalPair.right;
if (result != null)
Tracing.trace("CAS did not apply");
else
Tracing.trace("CAS applied successfully");
return result;
}
Tracing.trace("Paxos proposal not accepted (pre-empted by a higher ballot)");
contentions++;
Uninterruptibles.sleepUninterruptibly(ThreadLocalRandom.current().nextInt(100), TimeUnit.MILLISECONDS);
// continue to retry
}
} catch (CasWriteTimeoutException e) {
// Might be thrown by beginRepairAndPaxos. In that case, any contention that happened within the method and
// led up to the timeout was not accounted in our local 'contentions' variable and we add it now so it the
// contention recorded in the finally is correct.
contentions += e.contentions;
throw e;
} catch (WriteTimeoutException e) {
// Might be thrown by proposePaxos or commitPaxos
throw new CasWriteTimeoutException(e.writeType, e.consistency, e.received, e.blockFor, contentions);
} finally {
recordCasContention(metadata, key, casMetrics, contentions);
}
throw new CasWriteTimeoutException(WriteType.CAS, consistencyForPaxos, 0, consistencyForPaxos.blockFor(latestRs), contentions);
}
use of org.apache.cassandra.db.Keyspace in project cassandra by apache.
the class StorageProxy method wrapBatchResponseHandler.
// same as performWrites except does not initiate writes (but does perform availability checks).
private static WriteResponseHandlerWrapper wrapBatchResponseHandler(Mutation mutation, ConsistencyLevel consistencyLevel, ConsistencyLevel batchConsistencyLevel, WriteType writeType, BatchlogResponseHandler.BatchlogCleanup cleanup, long queryStartNanoTime) {
Keyspace keyspace = Keyspace.open(mutation.getKeyspaceName());
Token tk = mutation.key().getToken();
ReplicaPlan.ForTokenWrite replicaPlan = ReplicaPlans.forWrite(keyspace, consistencyLevel, tk, ReplicaPlans.writeNormal);
AbstractReplicationStrategy rs = replicaPlan.replicationStrategy();
AbstractWriteResponseHandler<IMutation> writeHandler = rs.getWriteResponseHandler(replicaPlan, null, writeType, queryStartNanoTime);
BatchlogResponseHandler<IMutation> batchHandler = new BatchlogResponseHandler<>(writeHandler, batchConsistencyLevel.blockFor(rs), cleanup, queryStartNanoTime);
return new WriteResponseHandlerWrapper(batchHandler, mutation);
}
Aggregations