mondrian.rolap.agg
Class SegmentCacheManager

java.lang.Object
  mondrian.rolap.agg.SegmentCacheManager

public class SegmentCacheManager
extends Object

Active object that maintains the "global cache" (in JVM, but shared between connections using a particular schema) and "external cache" (as implemented by a SegmentCache.

Segment states

State	Meaning
Local	Initial state of a segment

Decisions to be reviewed

1. Create variant of actor that processes all requests synchronously, and does not need a thread. This would be a more 'embedded' mode of operation (albeit with worse scale-out).

2. Move functionality into AggregationManager?

3. Delete RolapStar.lookupOrCreateAggregation(mondrian.rolap.agg.AggregationKey) and RolapStar.lookupSegment(mondrian.rolap.agg.AggregationKey) and RolapStar.lookupAggregationShared (formerly RolapStar.lookupAggregation).

Moved methods

(Keeping track of where methods came from will make it easier to merge to the mondrian-4 code line.)

1. RolapStar.getCellFromCache(mondrian.rolap.agg.CellRequest, mondrian.rolap.RolapAggregationManager.PinSet) moved from Aggregation.getCellValue

Done

1. Obsolete CountingAggregationManager, and property mondrian.rolap.agg.enableCacheHitCounters.

2. AggregationManager becomes non-singleton.

3. SegmentCacheWorker methods and segmentCache field become non-static. initCache() is called on construction. SegmentCache is passed into constructor (therefore move ServiceDiscovery into client). AggregationManager (or maybe MondrianServer) is another constructor parameter.

5. Move SegmentHeader, SegmentBody, ConstrainedColumn into mondrian.spi. Leave behind dependencies on mondrian.rolap.agg. In particular, put code that converts Segment + SegmentWithData to and from SegmentHeader + SegmentBody (e.g. SegmentHeader#forSegment) into a utility class. (Do this as CLEANUP, after functionality is complete?)

6. Move functionality Aggregation to Segment. Long-term, Aggregation should not be used as a 'gatekeeper' to Segment. Remove Aggregation fields columns and axes.

9. Obsolete RolapStar.cacheAggregations. Similar effect will be achieved by removing the 'jvm cache' from the chain of caches.

10. Rename Aggregation.Axis to SegmentAxis.

11. Remove Segment.setData and instead split out subclass SegmentWithData. Now segment is immutable. You don't have to wait for its state to change. You wait for a Future<SegmentWithData> to become ready.

12. Remove methods: RolapCube.checkAggregateModifications, RolapStar.checkAggregateModifications, RolapSchema.checkAggregateModifications, RolapStar.pushAggregateModificationsToGlobalCache, RolapSchema.pushAggregateModificationsToGlobalCache, RolapCube.pushAggregateModificationsToGlobalCache.

13. Add new implementations of Future: CompletedFuture and SlotFuture.

14. Remove methods:

• Remove SegmentLoader.loadSegmentsFromCache - creates a SegmentHeader that has PRECISELY same specification as the requested segment, very unlikely to have a hit
• Remove SegmentLoader.loadSegmentFromCacheRollup
• Break up SegmentLoader.cacheSegmentData, and place code that is called after a segment has arrived

13. Fix flush. Obsolete Aggregation.flush, and RolapStar.flush, which called it.

18. SegmentCacheManager#locateHeaderBody (and maybe other methods) call SegmentCacheWorker.get(mondrian.spi.SegmentHeader), and that's a slow blocking call. Make waits for segment futures should be called from a worker or client, not an agent.

Ideas and tasks

7. RolapStar.localAggregations and .sharedAggregations. Obsolete sharedAggregations.

8. Longer term. Move RolapStar.Bar.segmentRefs to Execution. Would it still be thread-local?

10. Call DataSourceChangeListener.isAggregationChanged(mondrian.rolap.agg.AggregationKey). Previously called from RolapStar.checkAggregateModifications, now never called.

12. We can quickly identify segments affected by a flush using SegmentCacheIndex.intersectRegion(java.lang.String, mondrian.util.ByteString, java.lang.String, java.lang.String, java.lang.String, mondrian.spi.SegmentColumn[]). But then what? Options:

1. Option #1. Pull them in, trim them, write them out? But: causes a lot of I/O, and we may never use these segments. Easiest.
2. Option #2. Mark the segments in the index as needing to be trimmed; trim them when read, and write out again. But: doesn't propagate to other nodes.
3. Option #3. (Best?) Write a mapping SegmentHeader->Restrictions into the cache. Less I/O than #1. Method "SegmentCache.addRestriction(SegmentHeader, CacheRegion)"

14. Move AggregationManager.getCellFromCache(mondrian.rolap.agg.CellRequest) somewhere else. It's concerned with local segments, not the global/external cache.

15. Method to convert SegmentHeader + SegmentBody to Segment + SegmentWithData is imperfect. Cannot parse predicates, compound predicates. Need mapping in star to do it properly and efficiently? SegmentBuilder.SegmentConverter is a hack that can be removed when this is fixed. See SegmentBuilder.toSegment(mondrian.spi.SegmentHeader, mondrian.rolap.RolapStar, mondrian.rolap.BitKey, mondrian.rolap.RolapStar.Column[], mondrian.rolap.RolapStar.Measure, java.util.List). Also see #20.

17. Revisit the strategy for finding segments that can be copied from global and external cache into local cache. The strategy of sending N CellRequests at a time, then executing SQL to fill in the gaps, is flawed. We need to maximize N in order to reduce segment fragmentation, but if too high, we blow memory. BasicQueryTest.testAnalysis is an example of this. Instead, we should send cell-requests in batches (is ~1000 the right size?), identify those that can be answered from global or external cache, return those segments, but not execute SQL until the end of the phase. If so, CellRequestQuantumExceededException be obsoleted.

19. Tracing. a. Remove or re-purpose FastBatchingCellReader.pendingCount; b. Add counter to measure requests satisfied by calling peek(mondrian.rolap.agg.CellRequest).

20. Obsolete SegmentDataset and its implementing classes. SegmentWithData can use SegmentBody instead. Will save copying.

21. Obsolete CombiningGenerator.

22. SegmentHeader.constrain(mondrian.spi.SegmentColumn[]) is broken for N-dimensional regions where N > 1. Each call currently creates N more 1-dimensional regions, but should create 1 more N-dimensional region. SegmentHeader.excludedRegions should be a list of SegmentColumn arrays.

23. All code that calls Future.get() should probably handle CancellationException.

24. Obsolete handler. Indirection doesn't win anything.

Author:

jhyde

Nested Class Summary

static interface	SegmentCacheManager.Command<T>
static class	SegmentCacheManager.FlushCommand Command to flush a particular region from cache.
static class	SegmentCacheManager.FlushResult Result of a SegmentCacheManager.FlushCommand.
(package private) static interface	SegmentCacheManager.Message
class	SegmentCacheManager.SegmentCacheIndexRegistry Registry of all the indexes that were created for this cache manager, per RolapStar.
static interface	SegmentCacheManager.Visitor Visitor for messages (commands and events).

Field Summary

ExecutorService	cacheExecutor Executor with which to send requests to external caches.
SegmentCache	compositeCache
List<SegmentCacheWorker>	segmentCacheWorkers
ExecutorService	sqlExecutor Executor with which to execute SQL requests.
Thread	thread

Constructor Summary

SegmentCacheManager(MondrianServer server)

Method Summary

<T> T	execute(SegmentCacheManager.Command<T> command)
void	externalSegmentCreated(SegmentHeader header, MondrianServer server) Tells the cache that a segment is newly available in an external cache.
void	externalSegmentDeleted(SegmentHeader header, MondrianServer server) Tells the cache that a segment is no longer available in an external cache.
SegmentBuilder.SegmentConverter	getConverter(RolapStar star, SegmentHeader header)
SegmentCacheManager.SegmentCacheIndexRegistry	getIndexRegistry()
void	loadFailed(RolapStar star, SegmentHeader header, Throwable throwable) Informs cache manager that a segment load failed.
void	loadSucceeded(RolapStar star, SegmentHeader header, SegmentBody body) Adds a segment to segment index.
SegmentWithData	peek(CellRequest request) Makes a quick request to the aggregation manager to see whether the cell value required by a particular cell request is in external cache.
void	printCacheState(CacheControl.CellRegion region, PrintWriter pw, Locus locus)
void	remove(RolapStar star, SegmentHeader header) Removes a segment from segment index.
void	shutdown() Shuts down this cache manager and all active threads and indexes.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

thread

public final Thread thread

cacheExecutor

public final ExecutorService cacheExecutor

Executor with which to send requests to external caches.

sqlExecutor

public final ExecutorService sqlExecutor

Executor with which to execute SQL requests.

TODO: create using factory and/or configuration parameters. Executor should be shared within MondrianServer or target JDBC database.

segmentCacheWorkers

public final List<SegmentCacheWorker> segmentCacheWorkers

compositeCache

public final SegmentCache compositeCache

Constructor Detail

SegmentCacheManager

public SegmentCacheManager(MondrianServer server)

Method Detail

execute

public <T> T execute(SegmentCacheManager.Command<T> command)

getIndexRegistry

public SegmentCacheManager.SegmentCacheIndexRegistry getIndexRegistry()

loadSucceeded

public void loadSucceeded(RolapStar star,
                          SegmentHeader header,
                          SegmentBody body)

Adds a segment to segment index.

Called when a SQL statement has finished loading a segment.

Does not add the segment to the external cache. That is a potentially long-duration operation, better carried out by a worker.

Parameters:

header - segment header

body - segment body

loadFailed

public void loadFailed(RolapStar star,
                       SegmentHeader header,
                       Throwable throwable)

Informs cache manager that a segment load failed.

Called when a SQL statement receives an error while loading a segment.

Parameters:

header - segment header

throwable - Error

remove

public void remove(RolapStar star,
                   SegmentHeader header)

Removes a segment from segment index.

Call is asynchronous. It comes back immediately.

Does not remove it from the external cache.

Parameters:

header - segment header

externalSegmentCreated

public void externalSegmentCreated(SegmentHeader header,
                                   MondrianServer server)

Tells the cache that a segment is newly available in an external cache.

externalSegmentDeleted

public void externalSegmentDeleted(SegmentHeader header,
                                   MondrianServer server)

Tells the cache that a segment is no longer available in an external cache.

printCacheState

public void printCacheState(CacheControl.CellRegion region,
                            PrintWriter pw,
                            Locus locus)

shutdown

public void shutdown()

Shuts down this cache manager and all active threads and indexes.

getConverter

public SegmentBuilder.SegmentConverter getConverter(RolapStar star,
                                                    SegmentHeader header)

peek

public SegmentWithData peek(CellRequest request)

Makes a quick request to the aggregation manager to see whether the cell value required by a particular cell request is in external cache.

'Quick' is relative. It is an asynchronous request (due to the aggregation manager being an actor) and therefore somewhat slow. If the segment is in cache, will save batching up future requests and re-executing the query. Win should be particularly noticeable for queries running on a populated cache. Without this feature, every query would require at least two iterations.

Request does not issue SQL to populate the segment. Nor does it try to find existing segments for rollup. Those operations can wait until next phase.

Client is responsible for adding the segment to its private cache.

Parameters:

request - Cell request

Returns:

Segment with data, or null if not in cache