Examples of com.google.common.util.concurrent.RateLimiter$WarmingUp

Package com.google.common.util.concurrent

Examples of com.google.common.util.concurrent.RateLimiter$WarmingUp

com.google.common.util.concurrent.RateLimiter
This implements the following function: ^ throttling | 3*stable + / interval | /. (cold) | / . | / . <-- "warmup period" is the area of the trapezoid between 2*stable + / . halfPermits and maxPermits interval | / . | / . | / . stable +----------/ WARM . } interval | . UP . } <-- this rectangle (from 0 to maxPermits, and | . PERIOD. } height== stableInterval) defines the cooldown period, | . . } and we want cooldownPeriod== warmupPeriod |---------------------------------> storedPermits (halfPermits) (maxPermits) Before going into the details of this particular function, let's keep in mind the basics: 1) The state of the RateLimiter (storedPermits) is a vertical line in this figure. 2) When the RateLimiter is not used, this goes right (up to maxPermits) 3) When the RateLimiter is used, this goes left (down to zero), since if we have storedPermits, we serve from those first 4) When _unused_, we go right at the same speed (rate)! I.e., if our rate is 2 permits per second, and 3 unused seconds pass, we will always save 6 permits (no matter what our initial position was), up to maxPermits. If we invert the rate, we get the "stableInterval" (interval between two requests in a perfectly spaced out sequence of requests of the given rate). Thus, if you want to see "how much time it will take to go from X storedPermits to X+K storedPermits?", the answer is always stableInterval * K. In the same example, for 2 permits per second, stableInterval is 500ms. Thus to go from X storedPermits to X+6 storedPermits, we require 6 * 500ms = 3 seconds. In short, the time it takes to move to the right (save K permits) is equal to the rectangle of width == K and height == stableInterval. 4) When _used_, the time it takes, as explained in the introductory class note, is equal to the integral of our function, between X permits and X-K permits, assuming we want to spend K saved permits. In summary, the time it takes to move to the left (spend K permits), is equal to the area of the function of width == K. Let's dive into this function now: When we have storedPermits <= halfPermits (the left portion of the function), then we spend them at the exact same rate that fresh permits would be generated anyway (that rate is 1/stableInterval). We size this area to be equal to _half_ the specified warmup period. Why we need this? And why half? We'll explain shortly below (after explaining the second part). Stored permits that are beyond halfPermits, are mapped to an ascending line, that goes from stableInterval to 3 * stableInterval. The average height for that part is 2 * stableInterval, and is sized appropriately to have an area _equal_ to the specified warmup period. Thus, by point (4) above, it takes "warmupPeriod" amount of time to go from maxPermits to halfPermits. BUT, by point (3) above, it only takes "warmupPeriod / 2" amount of time to return back to maxPermits, from halfPermits! (Because the trapezoid has double the area of the rectangle of height stableInterval and equivalent width). We decided that the "cooldown period" time should be equivalent to "warmup period", thus a fully saturated RateLimiter (with zero stored permits, serving only fresh ones) can go to a fully unsaturated (with storedPermits == maxPermits) in the same amount of time it takes for a fully unsaturated RateLimiter to return to the stableInterval -- which happens in halfPermits, since beyond that point, we use a horizontal line of "stableInterval" height, simulating the regular rate. Thus, we have figured all dimensions of this shape, to give all the desired properties: - the width is warmupPeriod / stableInterval, to make cooldownPeriod == warmupPeriod - the slope starts at the middle, and goes from stableInterval to 3*stableInterval so to have halfPermits being spend in double the usual time (half the rate), while their respective rate is steadily ramping up

            logger.debug("average hinted-row column size is {}; using pageSize of {}", averageColumnSize, pageSize);
        }


        // rate limit is in bytes per second. Uses Double.MAX_VALUE if disabled (set to 0 in cassandra.yaml).
        int throttleInKB = DatabaseDescriptor.getHintedHandoffThrottleInKB();
        RateLimiter rateLimiter = RateLimiter.create(throttleInKB == 0 ? Double.MAX_VALUE : throttleInKB * 1024);


        while (true)
        {
            QueryFilter filter = QueryFilter.getSliceFilter(epkey, new QueryPath(SystemTable.HINTS_CF), startColumn, ByteBufferUtil.EMPTY_BYTE_BUFFER, false, pageSize);
            ColumnFamily hintsPage = ColumnFamilyStore.removeDeleted(hintStore.getColumnFamily(filter), (int)(System.currentTimeMillis() / 1000));
            if (pagingFinished(hintsPage, startColumn))
            {
                if (ByteBufferUtil.EMPTY_BYTE_BUFFER.equals(startColumn))
                {
                    // we've started from the beginning and could not find anything (only maybe tombstones)
                    break;
                }
                else
                {
                    // restart query from the first column until we read an empty row;
                    // that will tell us everything was delivered successfully with no timeouts
                    startColumn = ByteBufferUtil.EMPTY_BYTE_BUFFER;
                    continue;
                }


            }


            for (final IColumn hint : hintsPage.getSortedColumns())
            {
                // Skip tombstones:
                // if we iterate quickly enough, it's possible that we could request a new page in the same millisecond
                // in which the local deletion timestamp was generated on the last column in the old page, in which
                // case the hint will have no columns (since it's deleted) but will still be included in the resultset
                // since (even with gcgs=0) it's still a "relevant" tombstone.
                if (!hint.isLive())
                    continue;


                startColumn = hint.name();


                ByteBuffer[] components = comparator.split(hint.name());
                int version = Int32Type.instance.compose(components[1]);
                DataInputStream in = new DataInputStream(ByteBufferUtil.inputStream(hint.value()));
                RowMutation rm;
                try
                {
                    rm = RowMutation.serializer.deserialize(in, version);
                }
                catch (UnknownColumnFamilyException e)
                {
                    logger.debug("Skipping delivery of hint for deleted columnfamily", e);
                    deleteHint(hostIdBytes, hint.name(), hint.maxTimestamp());
                    continue;
                }


                MessageOut<RowMutation> message = rm.createMessage();
                rateLimiter.acquire(message.serializedSize(MessagingService.current_version));
                WrappedRunnable callback = new WrappedRunnable()
                {
                    public void runMayThrow() throws IOException
                    {
                        rowsReplayed.incrementAndGet();

View Full Code Here

     * allow for a more memory efficient solution if we know the sstable don't overlap (see
     * LeveledCompactionStrategy for instance).
     */
    public List<ICompactionScanner> getScanners(Collection<SSTableReader> sstables, Range<Token> range)
    {
        RateLimiter limiter = CompactionManager.instance.getRateLimiter();
        ArrayList<ICompactionScanner> scanners = new ArrayList<ICompactionScanner>();
        for (SSTableReader sstable : sstables)
            scanners.add(sstable.getScanner(range, limiter));
        return scanners;
    }

View Full Code Here

        int pageSize = calculatePageSize();
        logger.debug("Using pageSize of {}", pageSize);


        // rate limit is in bytes per second. Uses Double.MAX_VALUE if disabled (set to 0 in cassandra.yaml).
        int throttleInKB = DatabaseDescriptor.getHintedHandoffThrottleInKB();
        RateLimiter rateLimiter = RateLimiter.create(throttleInKB == 0 ? Double.MAX_VALUE : throttleInKB * 1024);


        delivery:
        while (true)
        {
            QueryFilter filter = QueryFilter.getSliceFilter(epkey,
                                                            new QueryPath(SystemTable.HINTS_CF),
                                                            startColumn,
                                                            ByteBufferUtil.EMPTY_BYTE_BUFFER,
                                                            false,
                                                            pageSize);


            ColumnFamily hintsPage = ColumnFamilyStore.removeDeleted(hintStore.getColumnFamily(filter),
                                                                     (int) (System.currentTimeMillis() / 1000));


            if (pagingFinished(hintsPage, startColumn))
                break;


            // check if node is still alive and we should continue delivery process
            if (!FailureDetector.instance.isAlive(endpoint))
            {
                logger.info("Endpoint {} died during hint delivery; aborting ({} delivered)", endpoint, rowsReplayed);
                return;
            }


            List<WriteResponseHandler> responseHandlers = Lists.newArrayList();


            for (final IColumn hint : hintsPage.getSortedColumns())
            {
                // check if hints delivery has been paused during the process
                if (hintedHandOffPaused)
                {
                    logger.debug("Hints delivery process is paused, aborting");
                    break delivery;
                }


                // Skip tombstones:
                // if we iterate quickly enough, it's possible that we could request a new page in the same millisecond
                // in which the local deletion timestamp was generated on the last column in the old page, in which
                // case the hint will have no columns (since it's deleted) but will still be included in the resultset
                // since (even with gcgs=0) it's still a "relevant" tombstone.
                if (!hint.isLive())
                    continue;


                startColumn = hint.name();


                ByteBuffer[] components = comparator.split(hint.name());
                int version = Int32Type.instance.compose(components[1]);
                DataInputStream in = new DataInputStream(ByteBufferUtil.inputStream(hint.value()));
                RowMutation rm;
                try
                {
                    rm = RowMutation.serializer.deserialize(in, version);
                }
                catch (UnknownColumnFamilyException e)
                {
                    logger.debug("Skipping delivery of hint for deleted columnfamily", e);
                    deleteHint(hostIdBytes, hint.name(), hint.maxTimestamp());
                    continue;
                }
                catch (IOException e)
                {
                    throw new AssertionError(e);
                }


                MessageOut<RowMutation> message = rm.createMessage();
                rateLimiter.acquire(message.serializedSize(MessagingService.current_version));
                Runnable callback = new Runnable()
                {
                    public void run()
                    {
                        rowsReplayed.incrementAndGet();

View Full Code Here


        // rate limit is in bytes per second. Uses Double.MAX_VALUE if disabled (set to 0 in cassandra.yaml).
        // max rate is scaled by the number of nodes in the cluster (CASSANDRA-5272).
        int throttleInKB = DatabaseDescriptor.getHintedHandoffThrottleInKB()
                           / (StorageService.instance.getTokenMetadata().getAllEndpoints().size() - 1);
        RateLimiter rateLimiter = RateLimiter.create(throttleInKB == 0 ? Double.MAX_VALUE : throttleInKB * 1024);


        delivery:
        while (true)
        {
            QueryFilter filter = QueryFilter.getSliceFilter(epkey,
                                                            new QueryPath(SystemTable.HINTS_CF),
                                                            startColumn,
                                                            ByteBufferUtil.EMPTY_BYTE_BUFFER,
                                                            false,
                                                            pageSize);


            ColumnFamily hintsPage = ColumnFamilyStore.removeDeleted(hintStore.getColumnFamily(filter),
                                                                     (int) (System.currentTimeMillis() / 1000));


            if (pagingFinished(hintsPage, startColumn))
                break;


            // check if node is still alive and we should continue delivery process
            if (!FailureDetector.instance.isAlive(endpoint))
            {
                logger.info("Endpoint {} died during hint delivery; aborting ({} delivered)", endpoint, rowsReplayed);
                return;
            }


            List<WriteResponseHandler> responseHandlers = Lists.newArrayList();


            for (final IColumn hint : hintsPage.getSortedColumns())
            {
                // check if hints delivery has been paused during the process
                if (hintedHandOffPaused)
                {
                    logger.debug("Hints delivery process is paused, aborting");
                    break delivery;
                }


                // Skip tombstones:
                // if we iterate quickly enough, it's possible that we could request a new page in the same millisecond
                // in which the local deletion timestamp was generated on the last column in the old page, in which
                // case the hint will have no columns (since it's deleted) but will still be included in the resultset
                // since (even with gcgs=0) it's still a "relevant" tombstone.
                if (!hint.isLive())
                    continue;


                startColumn = hint.name();


                ByteBuffer[] components = comparator.split(hint.name());
                int version = Int32Type.instance.compose(components[1]);
                DataInputStream in = new DataInputStream(ByteBufferUtil.inputStream(hint.value()));
                RowMutation rm;
                try
                {
                    rm = RowMutation.serializer.deserialize(in, version);
                }
                catch (UnknownColumnFamilyException e)
                {
                    logger.debug("Skipping delivery of hint for deleted columnfamily", e);
                    deleteHint(hostIdBytes, hint.name(), hint.maxTimestamp());
                    continue;
                }
                catch (IOException e)
                {
                    throw new AssertionError(e);
                }


                Map<UUID, Long> truncationTimesCache = new HashMap<UUID, Long>();
                for (UUID cfId : ImmutableSet.copyOf((rm.getColumnFamilyIds())))
                {
                    Long truncatedAt = truncationTimesCache.get(cfId);
                    if (truncatedAt == null)
                    {
                        ColumnFamilyStore cfs = Table.open(rm.getTable()).getColumnFamilyStore(cfId);
                        truncatedAt = cfs.getTruncationTime();
                        truncationTimesCache.put(cfId, truncatedAt);
                    }


                    if (hint.maxTimestamp() < truncatedAt)
                    {
                        logger.debug("Skipping delivery of hint for truncated columnfamily {}" + cfId);
                        rm = rm.without(cfId);
                    }
                }


                if (rm.isEmpty())
                {
                    deleteHint(hostIdBytes, hint.name(), hint.maxTimestamp());
                    continue;
                }


                MessageOut<RowMutation> message = rm.createMessage();
                rateLimiter.acquire(message.serializedSize(MessagingService.current_version));
                Runnable callback = new Runnable()
                {
                    public void run()
                    {
                        rowsReplayed.incrementAndGet();

View Full Code Here


        output.println("total,interval_op_rate,interval_key_rate,latency/95th/99th,elapsed_time");


        int itemsPerThread = client.getKeysPerThread();
        int modulo = client.getNumKeys() % threadCount;
        RateLimiter rateLimiter = RateLimiter.create(client.getMaxOpsPerSecond());


        // creating required type of the threads for the test
        for (int i = 0; i < threadCount; i++) {
            if (i == threadCount - 1)
                itemsPerThread += modulo; // last one is going to handle N + modulo items

View Full Code Here

        waitForRawInserts.await("Failed to add raw data");
    }


    //@Test(dependsOnMethods = "insertRawData")
    public void queryRawDataAsync() throws Exception {
        RateLimiter readPermits = RateLimiter.create(50);


        log.info("Running queryRawDataAsync");
        long start = System.currentTimeMillis();


        DateTime startTime = hour(3).minusHours(1).minusSeconds(1);

View Full Code Here

        if (opCount < 0)
            workQueue = new ContinuousWorkQueue(50);
        else
            workQueue = FixedWorkQueue.build(opCount);


        RateLimiter rateLimiter = null;
        // TODO : move this to a new queue wrapper that gates progress based on a poisson (or configurable) distribution
        if (settings.rate.opRateTargetPerSecond > 0)
            rateLimiter = RateLimiter.create(settings.rate.opRateTargetPerSecond);


        final StressMetrics metrics = new StressMetrics(output, settings.log.intervalMillis);

View Full Code Here


        // rate limit is in bytes per second. Uses Double.MAX_VALUE if disabled (set to 0 in cassandra.yaml).
        // max rate is scaled by the number of nodes in the cluster (CASSANDRA-5272).
        int throttleInKB = DatabaseDescriptor.getHintedHandoffThrottleInKB()
                           / (StorageService.instance.getTokenMetadata().getAllEndpoints().size() - 1);
        RateLimiter rateLimiter = RateLimiter.create(throttleInKB == 0 ? Double.MAX_VALUE : throttleInKB * 1024);


        boolean finished = false;
        delivery:
        while (true)
        {
            long now = System.currentTimeMillis();
            QueryFilter filter = QueryFilter.getSliceFilter(epkey,
                                                            SystemKeyspace.HINTS_CF,
                                                            startColumn,
                                                            Composites.EMPTY,
                                                            false,
                                                            pageSize,
                                                            now);


            ColumnFamily hintsPage = ColumnFamilyStore.removeDeleted(hintStore.getColumnFamily(filter), (int) (now / 1000));


            if (pagingFinished(hintsPage, startColumn))
            {
                logger.info("Finished hinted handoff of {} rows to endpoint {}", rowsReplayed, endpoint);
                finished = true;
                break;
            }


            // check if node is still alive and we should continue delivery process
            if (!FailureDetector.instance.isAlive(endpoint))
            {
                logger.info("Endpoint {} died during hint delivery; aborting ({} delivered)", endpoint, rowsReplayed);
                break;
            }


            List<WriteResponseHandler> responseHandlers = Lists.newArrayList();
            for (final Cell hint : hintsPage)
            {
                // check if hints delivery has been paused during the process
                if (hintedHandOffPaused)
                {
                    logger.debug("Hints delivery process is paused, aborting");
                    break delivery;
                }


                // Skip tombstones:
                // if we iterate quickly enough, it's possible that we could request a new page in the same millisecond
                // in which the local deletion timestamp was generated on the last column in the old page, in which
                // case the hint will have no columns (since it's deleted) but will still be included in the resultset
                // since (even with gcgs=0) it's still a "relevant" tombstone.
                if (!hint.isLive())
                    continue;


                startColumn = hint.name();


                int version = Int32Type.instance.compose(hint.name().get(1));
                DataInputStream in = new DataInputStream(ByteBufferUtil.inputStream(hint.value()));
                Mutation mutation;
                try
                {
                    mutation = Mutation.serializer.deserialize(in, version);
                }
                catch (UnknownColumnFamilyException e)
                {
                    logger.debug("Skipping delivery of hint for deleted columnfamily", e);
                    deleteHint(hostIdBytes, hint.name(), hint.timestamp());
                    continue;
                }
                catch (IOException e)
                {
                    throw new AssertionError(e);
                }


                for (UUID cfId : mutation.getColumnFamilyIds())
                {
                    if (hint.timestamp() <= SystemKeyspace.getTruncatedAt(cfId))
                    {
                        logger.debug("Skipping delivery of hint for truncated columnfamily {}", cfId);
                        mutation = mutation.without(cfId);
                    }
                }


                if (mutation.isEmpty())
                {
                    deleteHint(hostIdBytes, hint.name(), hint.timestamp());
                    continue;
                }


                MessageOut<Mutation> message = mutation.createMessage();
                rateLimiter.acquire(message.serializedSize(MessagingService.current_version));
                Runnable callback = new Runnable()
                {
                    public void run()
                    {
                        rowsReplayed.incrementAndGet();

View Full Code Here

     */
    class CompactionScannerFactory implements EntryLogger.EntryLogListener {
        List<Offset> offsets = new ArrayList<Offset>();


        EntryLogScanner newScanner(final EntryLogMetadata meta) {
            final RateLimiter rateLimiter = RateLimiter.create(compactionRate);
            return new EntryLogScanner() {
                @Override
                public boolean accept(long ledgerId) {
                    return meta.containsLedger(ledgerId);
                }


                @Override
                public void process(final long ledgerId, long offset, ByteBuffer entry)
                        throws IOException {
                    rateLimiter.acquire();
                    synchronized (CompactionScannerFactory.this) {
                        if (offsets.size() > maxOutstandingRequests) {
                            waitEntrylogFlushed();
                        }
                        entry.getLong(); // discard ledger id, we already have it

View Full Code Here


        // rate limit is in bytes per second. Uses Double.MAX_VALUE if disabled (set to 0 in cassandra.yaml).
        // max rate is scaled by the number of nodes in the cluster (CASSANDRA-5272).
        int throttleInKB = DatabaseDescriptor.getHintedHandoffThrottleInKB()
                           / (StorageService.instance.getTokenMetadata().getAllEndpoints().size() - 1);
        RateLimiter rateLimiter = RateLimiter.create(throttleInKB == 0 ? Double.MAX_VALUE : throttleInKB * 1024);


        delivery:
        while (true)
        {
            QueryFilter filter = QueryFilter.getSliceFilter(epkey,
                                                            new QueryPath(SystemTable.HINTS_CF),
                                                            startColumn,
                                                            ByteBufferUtil.EMPTY_BYTE_BUFFER,
                                                            false,
                                                            pageSize);


            ColumnFamily hintsPage = ColumnFamilyStore.removeDeleted(hintStore.getColumnFamily(filter),
                                                                     (int) (System.currentTimeMillis() / 1000));


            if (pagingFinished(hintsPage, startColumn))
                break;


            // check if node is still alive and we should continue delivery process
            if (!FailureDetector.instance.isAlive(endpoint))
            {
                logger.info("Endpoint {} died during hint delivery; aborting ({} delivered)", endpoint, rowsReplayed);
                return;
            }


            List<WriteResponseHandler> responseHandlers = Lists.newArrayList();


            for (final IColumn hint : hintsPage.getSortedColumns())
            {
                // check if hints delivery has been paused during the process
                if (hintedHandOffPaused)
                {
                    logger.debug("Hints delivery process is paused, aborting");
                    break delivery;
                }


                // Skip tombstones:
                // if we iterate quickly enough, it's possible that we could request a new page in the same millisecond
                // in which the local deletion timestamp was generated on the last column in the old page, in which
                // case the hint will have no columns (since it's deleted) but will still be included in the resultset
                // since (even with gcgs=0) it's still a "relevant" tombstone.
                if (!hint.isLive())
                    continue;


                startColumn = hint.name();


                ByteBuffer[] components = comparator.split(hint.name());
                int version = Int32Type.instance.compose(components[1]);
                DataInputStream in = new DataInputStream(ByteBufferUtil.inputStream(hint.value()));
                RowMutation rm;
                try
                {
                    rm = RowMutation.serializer.deserialize(in, version);
                }
                catch (UnknownColumnFamilyException e)
                {
                    logger.debug("Skipping delivery of hint for deleted columnfamily", e);
                    deleteHint(hostIdBytes, hint.name(), hint.maxTimestamp());
                    continue;
                }
                catch (IOException e)
                {
                    throw new AssertionError(e);
                }


                Map<UUID, Long> truncationTimesCache = new HashMap<UUID, Long>();
                for (UUID cfId : ImmutableSet.copyOf((rm.getColumnFamilyIds())))
                {
                    Long truncatedAt = truncationTimesCache.get(cfId);
                    if (truncatedAt == null)
                    {
                        ColumnFamilyStore cfs = Table.open(rm.getTable()).getColumnFamilyStore(cfId);
                        truncatedAt = cfs.getTruncationTime();
                        truncationTimesCache.put(cfId, truncatedAt);
                    }


                    if (hint.maxTimestamp() < truncatedAt)
                    {
                        logger.debug("Skipping delivery of hint for truncated columnfamily {}" + cfId);
                        rm = rm.without(cfId);
                    }
                }


                if (rm.isEmpty())
                {
                    deleteHint(hostIdBytes, hint.name(), hint.maxTimestamp());
                    continue;
                }


                MessageOut<RowMutation> message = rm.createMessage();
                rateLimiter.acquire(message.serializedSize(MessagingService.current_version));
                Runnable callback = new Runnable()
                {
                    public void run()
                    {
                        rowsReplayed.incrementAndGet();

View Full Code Here

0 1 2 3 4 5

TOP

Related Classes of com.google.common.util.concurrent.RateLimiter$WarmingUp

com.elastisys.scale.commons.logreplayer.BurstingApacheLogReplayer

com.google.common.util.concurrent.SmoothRateLimiter.SmoothBursty

com.google.common.util.concurrent.SmoothRateLimiter.SmoothWarmingUp

org.apache.bookkeeper.bookie.GarbageCollectorThread$CompactionScannerFactory

org.apache.cassandra.db.BatchlogManager

org.apache.cassandra.db.compaction.AbstractCompactionStrategy

org.apache.cassandra.db.HintedHandOffManager

org.apache.cassandra.stress.StressAction

org.jclouds.smartos.SmartOSHostController

org.rhq.server.metrics.MetricsPerfTests

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.