[PeerMgr] Add Caching and Re-processing system (#263)

[peermgr]

- Add request cache with tests
- Add requestCache to peermgr
- refactored peer manager tests
- Added blockInfo struct, to allow sorting on the blockIndex
- added helper methods for cache, pickItem, pickFirstItem, removeHash,
findHash and refactored tests
- renamed requestcache to blockcache
- refactored peer manager to use block cache for block requests *only*
- added blockCallPeer function to handle block requests only
- refactored onDisconnect to add back any pending peer requests that the
disconnected peer did not complete into the peer manager queue

[peermgr/server]

- Modify onBlock handler in server, to send peermgr a BlockInfo struct

[peermgr/syncmgr/server]

- Modified blockIndex in BlockInfo to be uint32 and not uint64
- RequestBlocks in syncmgr now takes an index along with the hash
- modified syncmgr code to pass index along with hash in all methods

pkg/peermgr/blockcache.go

@@ -0,0 +1,155 @@
+package peermgr
+
+import (
+	"errors"
+	"sort"
+	"sync"
+
+	"github.com/CityOfZion/neo-go/pkg/wire/util"
+)
+
+var (
+	//ErrCacheLimit is returned when the cache limit is reached
+	ErrCacheLimit = errors.New("nomore items can be added to the cache")
+
+	//ErrNoItems is returned when pickItem is called and there are no items in the cache
+	ErrNoItems = errors.New("there are no items in the cache")
+
+	//ErrDuplicateItem is returned when you try to add the same item, more than once to the cache
+	ErrDuplicateItem = errors.New("this item is already in the cache")
+)
+
+//BlockInfo holds the necessary information that the cache needs
+// to sort and store block requests
+type BlockInfo struct {
+	BlockHash  util.Uint256
+	BlockIndex uint32
+}
+
+// Equals returns true if two blockInfo objects
+// have the same hash and the same index
+func (bi *BlockInfo) Equals(other BlockInfo) bool {
+	return bi.BlockHash.Equals(other.BlockHash) && bi.BlockIndex == other.BlockIndex
+}
+
+// indexSorter sorts the blockInfos by blockIndex.
+type indexSorter []BlockInfo
+
+func (is indexSorter) Len() int           { return len(is) }
+func (is indexSorter) Swap(i, j int)      { is[i], is[j] = is[j], is[i] }
+func (is indexSorter) Less(i, j int) bool { return is[i].BlockIndex < is[j].BlockIndex }
+
+//blockCache will cache any pending block requests
+// for the node when there are no available nodes
+type blockCache struct {
+	cacheLimit int
+	cacheLock  sync.Mutex
+	cache      []BlockInfo
+}
+
+func newBlockCache(cacheLimit int) *blockCache {
+	return &blockCache{
+		cache:      make([]BlockInfo, 0, cacheLimit),
+		cacheLimit: cacheLimit,
+	}
+}
+
+func (bc *blockCache) addBlockInfo(bi BlockInfo) error {
+	if bc.cacheLen() == bc.cacheLimit {
+		return ErrCacheLimit
+	}
+
+	bc.cacheLock.Lock()
+	defer bc.cacheLock.Unlock()
+
+	// Check for duplicates. slice will always be small so a simple for loop will work
+	for _, bInfo := range bc.cache {
+		if bInfo.Equals(bi) {
+			return ErrDuplicateItem
+		}
+	}
+	bc.cache = append(bc.cache, bi)
+
+	sort.Sort(indexSorter(bc.cache))
+
+	return nil
+}
+
+func (bc *blockCache) addBlockInfos(bis []BlockInfo) error {
+
+	if len(bis)+bc.cacheLen() > bc.cacheLimit {
+		return errors.New("too many items to add, this will exceed the cache limit")
+	}
+
+	for _, bi := range bis {
+		err := bc.addBlockInfo(bi)
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (bc *blockCache) cacheLen() int {
+	bc.cacheLock.Lock()
+	defer bc.cacheLock.Unlock()
+	return len(bc.cache)
+}
+
+func (bc *blockCache) pickFirstItem() (BlockInfo, error) {
+	return bc.pickItem(0)
+}
+
+func (bc *blockCache) pickAllItems() ([]BlockInfo, error) {
+
+	numOfItems := bc.cacheLen()
+
+	items := make([]BlockInfo, 0, numOfItems)
+
+	for i := 0; i < numOfItems; i++ {
+		bi, err := bc.pickFirstItem()
+		if err != nil {
+			return nil, err
+		}
+		items = append(items, bi)
+	}
+	return items, nil
+}
+
+func (bc *blockCache) pickItem(i uint) (BlockInfo, error) {
+	if bc.cacheLen() < 1 {
+		return BlockInfo{}, ErrNoItems
+	}
+
+	if i >= uint(bc.cacheLen()) {
+		return BlockInfo{}, errors.New("index out of range")
+	}
+
+	bc.cacheLock.Lock()
+	defer bc.cacheLock.Unlock()
+
+	item := bc.cache[i]
+	bc.cache = append(bc.cache[:i], bc.cache[i+1:]...)
+	return item, nil
+}
+
+func (bc *blockCache) removeHash(hashToRemove util.Uint256) error {
+	index, err := bc.findHash(hashToRemove)
+	if err != nil {
+		return err
+	}
+
+	_, err = bc.pickItem(uint(index))
+	return err
+}
+
+func (bc *blockCache) findHash(hashToFind util.Uint256) (int, error) {
+	bc.cacheLock.Lock()
+	defer bc.cacheLock.Unlock()
+	for i, bInfo := range bc.cache {
+		if bInfo.BlockHash.Equals(hashToFind) {
+			return i, nil
+		}
+	}
+	return -1, errors.New("hash cannot be found in the cache")
+}

pkg/peermgr/blockcache_test.go

@@ -0,0 +1,80 @@
+package peermgr
+
+import (
+	"math/rand"
+	"testing"
+
+	"github.com/CityOfZion/neo-go/pkg/wire/util"
+	"github.com/stretchr/testify/assert"
+)
+
+func TestAddBlock(t *testing.T) {
+
+	bc := &blockCache{
+		cacheLimit: 20,
+	}
+	bi := randomBlockInfo(t)
+
+	err := bc.addBlockInfo(bi)
+	assert.Equal(t, nil, err)
+
+	assert.Equal(t, 1, bc.cacheLen())
+
+	err = bc.addBlockInfo(bi)
+	assert.Equal(t, ErrDuplicateItem, err)
+
+	assert.Equal(t, 1, bc.cacheLen())
+}
+
+func TestCacheLimit(t *testing.T) {
+
+	bc := &blockCache{
+		cacheLimit: 20,
+	}
+
+	for i := 0; i < bc.cacheLimit; i++ {
+		err := bc.addBlockInfo(randomBlockInfo(t))
+		assert.Equal(t, nil, err)
+	}
+
+	err := bc.addBlockInfo(randomBlockInfo(t))
+	assert.Equal(t, ErrCacheLimit, err)
+
+	assert.Equal(t, bc.cacheLimit, bc.cacheLen())
+}
+func TestPickItem(t *testing.T) {
+
+	bc := &blockCache{
+		cacheLimit: 20,
+	}
+
+	for i := 0; i < bc.cacheLimit; i++ {
+		err := bc.addBlockInfo(randomBlockInfo(t))
+		assert.Equal(t, nil, err)
+	}
+
+	for i := 0; i < bc.cacheLimit; i++ {
+		_, err := bc.pickFirstItem()
+		assert.Equal(t, nil, err)
+	}
+
+	assert.Equal(t, 0, bc.cacheLen())
+}
+
+func randomUint256(t *testing.T) util.Uint256 {
+	rand32 := make([]byte, 32)
+	rand.Read(rand32)
+
+	u, err := util.Uint256DecodeBytes(rand32)
+	assert.Equal(t, nil, err)
+
+	return u
+}
+
+func randomBlockInfo(t *testing.T) BlockInfo {
+
+	return BlockInfo{
+		randomUint256(t),
+		rand.Uint32(),
+	}
+}

pkg/peermgr/peermgr.go

@@ -2,6 +2,7 @@ package peermgr
 
 import (
 	"errors"
+	"fmt"
 	"sync"
 
 	"github.com/CityOfZion/neo-go/pkg/wire/command"
@@ -9,6 +10,15 @@ import (
 	"github.com/CityOfZion/neo-go/pkg/wire/util"
 )
 
+const (
+	// blockCacheLimit is the maximum amount of pending requests that the cache can hold
+	pendingBlockCacheLimit = 20
+
+	//peerBlockCacheLimit is the maximum amount of inflight blocks that a peer can
+	// have, before they are flagged as busy
+	peerBlockCacheLimit = 1
+)
+
 var (
 	//ErrNoAvailablePeers is returned when a request for data from a peer is invoked
 	// but there are no available peers to request data from
@@ -28,6 +38,10 @@ type mPeer interface {
 }
 
 type peerstats struct {
+	// when a peer is sent a blockRequest
+	// the peermanager will track this using this blockCache
+	blockCache *blockCache
+	// all other requests will be tracked using the requests map
 	requests map[command.Type]bool
 }
 
@@ -35,12 +49,15 @@ type peerstats struct {
 type PeerMgr struct {
 	pLock sync.RWMutex
 	peers map[mPeer]peerstats
+
+	requestCache *blockCache
 }
 
 //New returns a new peermgr object
 func New() *PeerMgr {
 	return &PeerMgr{
-		peers: make(map[mPeer]peerstats),
+		peers:        make(map[mPeer]peerstats),
+		requestCache: newBlockCache(pendingBlockCacheLimit),
 	}
 }
 
@@ -52,7 +69,10 @@ func (pmgr *PeerMgr) AddPeer(peer mPeer) {
 	if _, exists := pmgr.peers[peer]; exists {
 		return
 	}
-	pmgr.peers[peer] = peerstats{requests: make(map[command.Type]bool)}
+	pmgr.peers[peer] = peerstats{
+		requests:   make(map[command.Type]bool),
+		blockCache: newBlockCache(peerBlockCacheLimit),
+	}
 	go pmgr.onDisconnect(peer)
 }
 
@@ -61,6 +81,8 @@ func (pmgr *PeerMgr) AddPeer(peer mPeer) {
 func (pmgr *PeerMgr) MsgReceived(peer mPeer, cmd command.Type) error {
 	pmgr.pLock.Lock()
 	defer pmgr.pLock.Unlock()
+
+	// if peer was unknown then disconnect
 	val, ok := pmgr.peers[peer]
 	if !ok {
 
@@ -76,6 +98,44 @@ func (pmgr *PeerMgr) MsgReceived(peer mPeer, cmd command.Type) error {
 	return nil
 }
 
+//BlockMsgReceived notifies the peer manager that we have received a
+// block message from a peer
+func (pmgr *PeerMgr) BlockMsgReceived(peer mPeer, bi BlockInfo) error {
+
+	// if peer was unknown then disconnect
+	val, ok := pmgr.peers[peer]
+	if !ok {
+
+		go func() {
+			peer.NotifyDisconnect()
+		}()
+
+		peer.Disconnect()
+		return ErrUnknownPeer
+	}
+
+	// // remove item from the peersBlock cache
+	err := val.blockCache.removeHash(bi.BlockHash)
+	if err != nil {
+		return err
+	}
+
+	// check if cache empty, if so then return
+	if pmgr.requestCache.cacheLen() == 0 {
+		return nil
+	}
+
+	// Try to clean an item from the pendingBlockCache, a peer has just finished serving a block request
+	cachedBInfo, err := pmgr.requestCache.pickFirstItem()
+	if err != nil {
+		return err
+	}
+
+	return pmgr.blockCallPeer(cachedBInfo, func(p mPeer) error {
+		return p.RequestBlocks([]util.Uint256{cachedBInfo.BlockHash})
+	})
+}
+
 // Len returns the amount of peers that the peer manager
 //currently knows about
 func (pmgr *PeerMgr) Len() int {
@@ -84,25 +144,34 @@ func (pmgr *PeerMgr) Len() int {
 	return len(pmgr.peers)
 }
 
-// RequestBlock will request  a block from the most
+// RequestBlock will request a block from the most
 // available peer. Then update it's stats, so we know that
 // this peer is busy
-func (pmgr *PeerMgr) RequestBlock(hash util.Uint256) error {
-	return pmgr.callPeerForCmd(command.Block, func(p mPeer) error {
-		return p.RequestBlocks([]util.Uint256{hash})
+func (pmgr *PeerMgr) RequestBlock(bi BlockInfo) error {
+	pmgr.pLock.Lock()
+	defer pmgr.pLock.Unlock()
+
+	err := pmgr.blockCallPeer(bi, func(p mPeer) error {
+		return p.RequestBlocks([]util.Uint256{bi.BlockHash})
 	})
+
+	if err == ErrNoAvailablePeers {
+		return pmgr.requestCache.addBlockInfo(bi)
+	}
+
+	return err
 }
 
 // RequestHeaders will request a headers from the most available peer.
 func (pmgr *PeerMgr) RequestHeaders(hash util.Uint256) error {
+	pmgr.pLock.Lock()
+	defer pmgr.pLock.Unlock()
 	return pmgr.callPeerForCmd(command.Headers, func(p mPeer) error {
 		return p.RequestHeaders(hash)
 	})
 }
 
 func (pmgr *PeerMgr) callPeerForCmd(cmd command.Type, f func(p mPeer) error) error {
-	pmgr.pLock.Lock()
-	defer pmgr.pLock.Unlock()
 	for peer, stats := range pmgr.peers {
 		if !stats.requests[cmd] {
 			stats.requests[cmd] = true
@@ -111,12 +180,48 @@ func (pmgr *PeerMgr) callPeerForCmd(cmd command.Type, f func(p mPeer) error) err
 	}
 	return ErrNoAvailablePeers
 }
+
+func (pmgr *PeerMgr) blockCallPeer(bi BlockInfo, f func(p mPeer) error) error {
+	for peer, stats := range pmgr.peers {
+		if stats.blockCache.cacheLen() < peerBlockCacheLimit {
+			err := stats.blockCache.addBlockInfo(bi)
+			if err != nil {
+				return err
+			}
+			return f(peer)
+		}
+	}
+	return ErrNoAvailablePeers
+}
+
 func (pmgr *PeerMgr) onDisconnect(p mPeer) {
 
 	// Blocking until peer is disconnected
 	p.NotifyDisconnect()
 
 	pmgr.pLock.Lock()
-	delete(pmgr.peers, p)
-	pmgr.pLock.Unlock()
+	defer func() {
+		delete(pmgr.peers, p)
+		pmgr.pLock.Unlock()
+	}()
+
+	// Add all of peers outstanding block requests into
+	// the peer managers pendingBlockRequestCache
+
+	val, ok := pmgr.peers[p]
+	if !ok {
+		return
+	}
+
+	pendingRequests, err := val.blockCache.pickAllItems()
+	if err != nil {
+		fmt.Println(err.Error())
+		return
+	}
+
+	err = pmgr.requestCache.addBlockInfos(pendingRequests)
+	if err != nil {
+		fmt.Println(err.Error())
+		return
+	}
 }

pkg/peermgr/peermgr_test.go

@@ -4,7 +4,6 @@ import (
 	"testing"
 
 	"github.com/CityOfZion/neo-go/pkg/wire/command"
-
 	"github.com/CityOfZion/neo-go/pkg/wire/util"
 	"github.com/stretchr/testify/assert"
 )
@@ -59,42 +58,65 @@ func TestRequestBlocks(t *testing.T) {
 	pmgr.AddPeer(peerB)
 	pmgr.AddPeer(peerC)
 
-	err := pmgr.RequestBlock(util.Uint256{})
+	firstBlock := randomBlockInfo(t)
+	err := pmgr.RequestBlock(firstBlock)
 	assert.Nil(t, err)
 
-	err = pmgr.RequestBlock(util.Uint256{})
+	secondBlock := randomBlockInfo(t)
+	err = pmgr.RequestBlock(secondBlock)
 	assert.Nil(t, err)
 
-	err = pmgr.RequestBlock(util.Uint256{})
+	thirdBlock := randomBlockInfo(t)
+	err = pmgr.RequestBlock(thirdBlock)
 	assert.Nil(t, err)
 
 	// Since the peer manager did not get a MsgReceived
 	// in between the block requests
 	// a request should be sent to all peers
+	// This is only true, if peerBlockCacheLimit == 1
 
 	assert.Equal(t, 1, peerA.blockRequested)
 	assert.Equal(t, 1, peerB.blockRequested)
 	assert.Equal(t, 1, peerC.blockRequested)
 
 	// Since the peer manager still has not received a MsgReceived
-	// another call to request blocks, will return a NoAvailablePeerError
+	// another call to request blocks, will add the request to the cache
+	// and return a nil err
 
-	err = pmgr.RequestBlock(util.Uint256{})
-	assert.Equal(t, ErrNoAvailablePeers, err)
+	fourthBlock := randomBlockInfo(t)
+	err = pmgr.RequestBlock(fourthBlock)
+	assert.Equal(t, nil, err)
+	assert.Equal(t, 1, pmgr.requestCache.cacheLen())
 
-	// If we tell the peer manager that peerA has given us a block
-	// then send another BlockRequest. It will go to peerA
-	// since the other two peers are still busy with their
-	// block requests
+	// If we tell the peer manager that we have received a block
+	// it will check the cache for any pending requests and send a block request if there are any.
+	// The request  will go to the peer who sent back the block corresponding to the first hash
+	//  since the other two peers are still busy with their block requests
 
-	pmgr.MsgReceived(peerA, command.Block)
-	err = pmgr.RequestBlock(util.Uint256{})
+	peer := findPeerwithHash(t, pmgr, firstBlock.BlockHash)
+	err = pmgr.BlockMsgReceived(peer, firstBlock)
 	assert.Nil(t, err)
 
-	assert.Equal(t, 2, peerA.blockRequested)
-	assert.Equal(t, 1, peerB.blockRequested)
-	assert.Equal(t, 1, peerC.blockRequested)
+	totalRequests := peerA.blockRequested + peerB.blockRequested + peerC.blockRequested
+	assert.Equal(t, 4, totalRequests)
+
+	// // cache should be empty now
+	assert.Equal(t, 0, pmgr.requestCache.cacheLen())
 }
+
+// The peer manager does not tell you what peer was sent a particular block request
+// For testing purposes, the following function will find that peer
+func findPeerwithHash(t *testing.T, pmgr *PeerMgr, blockHash util.Uint256) mPeer {
+	for peer, stats := range pmgr.peers {
+		_, err := stats.blockCache.findHash(blockHash)
+		if err == nil {
+			return peer
+		}
+	}
+	assert.Fail(t, "cannot find a peer with that hash")
+	return nil
+}
+
 func TestRequestHeaders(t *testing.T) {
 	pmgr := New()
 
@@ -152,7 +174,7 @@ func TestUnknownPeer(t *testing.T) {
 		quit:         make(chan bool),
 	}
 
-	err := pmgr.MsgReceived(unknownPeer, command.Block)
+	err := pmgr.MsgReceived(unknownPeer, command.Headers)
 	assert.Equal(t, true, unknownPeer.disconnected)
 	assert.Equal(t, ErrUnknownPeer, err)
 }

pkg/server/syncmgr.go

@@ -3,6 +3,8 @@ package server
 import (
 	"encoding/binary"
 
+	"github.com/CityOfZion/neo-go/pkg/peermgr"
+
 	"github.com/CityOfZion/neo-go/pkg/peer"
 	"github.com/CityOfZion/neo-go/pkg/syncmgr"
 	"github.com/CityOfZion/neo-go/pkg/wire/payload"
@@ -34,7 +36,10 @@ func (s *Server) onHeader(peer *peer.Peer, hdrsMessage *payload.HeadersMessage)
 }
 
 func (s *Server) onBlock(peer *peer.Peer, blockMsg *payload.BlockMessage) {
-	s.pmg.MsgReceived(peer, blockMsg.Command())
+	s.pmg.BlockMsgReceived(peer, peermgr.BlockInfo{
+		BlockHash:  blockMsg.Hash,
+		BlockIndex: blockMsg.Index,
+	})
 	s.smg.OnBlock(peer, blockMsg)
 }
 
@@ -50,8 +55,11 @@ func (s *Server) requestHeaders(hash util.Uint256) error {
 	return s.pmg.RequestHeaders(hash)
 }
 
-func (s *Server) requestBlock(hash util.Uint256) error {
-	return s.pmg.RequestBlock(hash)
+func (s *Server) requestBlock(hash util.Uint256, index uint32) error {
+	return s.pmg.RequestBlock(peermgr.BlockInfo{
+		BlockHash:  hash,
+		BlockIndex: index,
+	})
 }
 
 // getNextBlockHash searches the database for the blockHash

pkg/syncmgr/blockmode.go

@@ -32,7 +32,7 @@ func (s *Syncmgr) blockModeOnBlock(peer SyncPeer, block payload.Block) error {
 		if err != nil {
 			return err
 		}
-		err = s.cfg.RequestBlock(nextHash)
+		err = s.cfg.RequestBlock(nextHash, block.Index)
 		return err
 	}
 

pkg/syncmgr/config.go

@@ -18,7 +18,7 @@ type Config struct {
 
 	//RequestBlock will send a getdata request for the block
 	// with the hash passed as a parameter
-	RequestBlock func(hash util.Uint256) error
+	RequestBlock func(hash util.Uint256, index uint32) error
 
 	// GetNextBlockHash returns the block hash of the header infront of thr block
 	// at the tip of this nodes chain. This assumes that the node is not in sync

pkg/syncmgr/headermode.go

@@ -12,6 +12,7 @@ func (s *Syncmgr) headersModeOnHeaders(peer SyncPeer, hdrs []*payload.BlockBase)
 	// Note: For the un-optimised version, we move straight to blocksOnly mode
 
 	firstHash := hdrs[0].Hash
+	firstHdrIndex := hdrs[0].Index
 
 	err := s.cfg.ProcessHeaders(hdrs)
 	if err == nil {
@@ -19,7 +20,7 @@ func (s *Syncmgr) headersModeOnHeaders(peer SyncPeer, hdrs []*payload.BlockBase)
 		s.headerHash = hdrs[len(hdrs)-1].Hash
 
 		s.syncmode = blockMode
-		return s.cfg.RequestBlock(firstHash)
+		return s.cfg.RequestBlock(firstHash, firstHdrIndex)
 	}
 
 	// Check whether it is a validation error, or a database error

pkg/syncmgr/mockhelpers_test.go

@@ -46,7 +46,7 @@ func (s *syncTestHelper) FetchBlockAgain(util.Uint256) error {
 	return s.err
 }
 
-func (s *syncTestHelper) RequestBlock(util.Uint256) error {
+func (s *syncTestHelper) RequestBlock(util.Uint256, uint32) error {
 	s.blockFetchRequest++
 	return s.err
 }

pkg/syncmgr/normalmode.go

@@ -16,6 +16,7 @@ func (s *Syncmgr) normalModeOnHeaders(peer SyncPeer, hdrs []*payload.BlockBase)
 
 	lenHeaders := len(hdrs)
 	firstHash := hdrs[0].Hash
+	firstHdrIndex := hdrs[0].Index
 	lastHash := hdrs[lenHeaders-1].Hash
 
 	// Update syncmgr latest header
@@ -32,7 +33,7 @@ func (s *Syncmgr) normalModeOnHeaders(peer SyncPeer, hdrs []*payload.BlockBase)
 	// Bounds state that len > 1 && len!= 2000 & maxHeadersInMessage == 2000
 	// This means that we have less than 2k headers
 	s.syncmode = blockMode
-	return s.cfg.RequestBlock(firstHash)
+	return s.cfg.RequestBlock(firstHash, firstHdrIndex)
 }
 
 // normalModeOnBlock is called when the sync manager is normal mode