2019-11-25 20:18:02 +00:00
|
|
|
import tables, heapqueue
|
|
|
|
import chronos
|
|
|
|
|
|
|
|
type
|
|
|
|
PeerType* = enum
|
|
|
|
None, Incoming, Outgoing
|
|
|
|
|
|
|
|
PeerFlags = enum
|
|
|
|
Acquired, DeleteOnRelease
|
|
|
|
|
|
|
|
PeerItem[T] = object
|
|
|
|
data: T
|
|
|
|
peerType: PeerType
|
|
|
|
flags: set[PeerFlags]
|
|
|
|
index: int
|
|
|
|
|
|
|
|
PeerIndex = object
|
|
|
|
data: int
|
|
|
|
cmp: proc(a, b: PeerIndex): bool {.closure, gcsafe.}
|
|
|
|
|
|
|
|
PeerPool*[A, B] = ref object
|
|
|
|
incNeEvent: AsyncEvent
|
|
|
|
outNeEvent: AsyncEvent
|
|
|
|
incQueue: HeapQueue[PeerIndex]
|
|
|
|
outQueue: HeapQueue[PeerIndex]
|
|
|
|
registry: Table[B, PeerIndex]
|
|
|
|
storage: seq[PeerItem[A]]
|
|
|
|
cmp: proc(a, b: PeerIndex): bool {.closure, gcsafe.}
|
|
|
|
maxPeersCount: int
|
|
|
|
maxIncPeersCount: int
|
|
|
|
maxOutPeersCount: int
|
|
|
|
curIncPeersCount: int
|
|
|
|
curOutPeersCount: int
|
|
|
|
acqIncPeersCount: int
|
|
|
|
acqOutPeersCount: int
|
|
|
|
|
2020-01-21 18:30:21 +00:00
|
|
|
PeerPoolError* = object of CatchableError
|
|
|
|
|
2019-11-25 20:18:02 +00:00
|
|
|
proc `<`*(a, b: PeerIndex): bool =
|
|
|
|
result = a.cmp(b, a)
|
|
|
|
|
|
|
|
proc fireEvent[A, B](pool: PeerPool[A, B], item: PeerItem[A]) {.inline.} =
|
|
|
|
if item.peerType == PeerType.Incoming:
|
|
|
|
pool.incNeEvent.fire()
|
|
|
|
elif item.peerType == PeerType.Outgoing:
|
|
|
|
pool.outNeEvent.fire()
|
|
|
|
|
|
|
|
proc waitEvent[A, B](pool: PeerPool[A, B],
|
|
|
|
filter: set[PeerType]) {.async.} =
|
|
|
|
if filter == {PeerType.Incoming, PeerType.Outgoing} or filter == {}:
|
|
|
|
var fut1 = pool.incNeEvent.wait()
|
|
|
|
var fut2 = pool.outNeEvent.wait()
|
|
|
|
try:
|
|
|
|
discard await one(fut1, fut2)
|
|
|
|
if fut1.finished:
|
|
|
|
if not(fut2.finished):
|
|
|
|
fut2.cancel()
|
|
|
|
pool.incNeEvent.clear()
|
|
|
|
else:
|
|
|
|
if not(fut1.finished):
|
|
|
|
fut1.cancel()
|
|
|
|
pool.outNeEvent.clear()
|
|
|
|
except CancelledError:
|
|
|
|
if not(fut1.finished):
|
|
|
|
fut1.cancel()
|
|
|
|
if not(fut2.finished):
|
|
|
|
fut2.cancel()
|
|
|
|
raise
|
|
|
|
elif PeerType.Incoming in filter:
|
|
|
|
await pool.incNeEvent.wait()
|
|
|
|
pool.incNeEvent.clear()
|
|
|
|
elif PeerType.Outgoing in filter:
|
|
|
|
await pool.outNeEvent.wait()
|
|
|
|
pool.outNeEvent.clear()
|
|
|
|
|
|
|
|
template getItem[A, B](pool: PeerPool[A, B],
|
|
|
|
filter: set[PeerType]): ptr PeerItem[A] =
|
|
|
|
doAssert((len(pool.outQueue) > 0) or (len(pool.incQueue) > 0))
|
|
|
|
var pindex: int
|
|
|
|
if filter == {PeerType.Incoming, PeerType.Outgoing}:
|
|
|
|
if len(pool.outQueue) > 0 and len(pool.incQueue) > 0:
|
|
|
|
# Don't think `<` is actually `<` here.
|
|
|
|
if pool.incQueue[0] < pool.outQueue[0]:
|
|
|
|
inc(pool.acqIncPeersCount)
|
|
|
|
pindex = pool.incQueue.pop().data
|
|
|
|
else:
|
|
|
|
inc(pool.acqOutPeersCount)
|
|
|
|
pindex = pool.outQueue.pop().data
|
|
|
|
else:
|
|
|
|
if len(pool.outQueue) > 0:
|
|
|
|
inc(pool.acqOutPeersCount)
|
|
|
|
pindex = pool.outQueue.pop().data
|
|
|
|
else:
|
|
|
|
inc(pool.acqIncPeersCount)
|
|
|
|
pindex = pool.incQueue.pop().data
|
|
|
|
else:
|
|
|
|
if PeerType.Outgoing in filter:
|
|
|
|
inc(pool.acqOutPeersCount)
|
|
|
|
pindex = pool.outQueue.pop().data
|
|
|
|
elif PeerType.Incoming in filter:
|
|
|
|
inc(pool.acqIncPeersCount)
|
|
|
|
pindex = pool.incQueue.pop().data
|
|
|
|
addr(pool.storage[pindex])
|
|
|
|
|
|
|
|
proc newPeerPool*[A, B](maxPeers = -1,
|
|
|
|
maxIncomingPeers = -1,
|
|
|
|
maxOutgoingPeers = -1): PeerPool[A, B] =
|
|
|
|
## Create new PeerPool.
|
|
|
|
##
|
|
|
|
## ``maxPeers`` - maximum number of peers allowed. All the peers which
|
|
|
|
## exceeds this number will be rejected (``addPeer()`` procedure will return
|
|
|
|
## ``false``). By default this number is infinite.
|
|
|
|
##
|
|
|
|
## ``maxIncomingPeers`` - maximum number of incoming peers allowed. All the
|
|
|
|
## incoming peers exceeds this number will be rejected. By default this
|
|
|
|
## number is infinite.
|
|
|
|
##
|
|
|
|
## ``maxOutgoingPeers`` - maximum number of outgoing peers allowed. All the
|
|
|
|
## outgoing peers exceeds this number will be rejected. By default this
|
|
|
|
## number if infinite.
|
|
|
|
##
|
|
|
|
## Please note, that if ``maxPeers`` is positive non-zero value, then equation
|
|
|
|
## ``maxPeers >= maxIncomingPeers + maxOutgoingPeers`` must be ``true``.
|
|
|
|
var res = PeerPool[A, B]()
|
|
|
|
if maxPeers != -1:
|
|
|
|
doAssert(maxPeers >= maxIncomingPeers + maxOutgoingPeers)
|
|
|
|
|
|
|
|
res.maxPeersCount = if maxPeers < 0: high(int)
|
|
|
|
else: maxPeers
|
|
|
|
res.maxIncPeersCount = if maxIncomingPeers < 0: high(int)
|
|
|
|
else: maxIncomingPeers
|
|
|
|
res.maxOutPeersCount = if maxOutgoingPeers < 0: high(int)
|
|
|
|
else: maxOutgoingPeers
|
|
|
|
res.incNeEvent = newAsyncEvent()
|
|
|
|
res.outNeEvent = newAsyncEvent()
|
|
|
|
res.incQueue = initHeapQueue[PeerIndex]()
|
|
|
|
res.outQueue = initHeapQueue[PeerIndex]()
|
|
|
|
res.registry = initTable[B, PeerIndex]()
|
|
|
|
res.storage = newSeq[PeerItem[A]]()
|
|
|
|
|
|
|
|
proc peerCmp(a, b: PeerIndex): bool {.closure, gcsafe.} =
|
|
|
|
let p1 = res.storage[a.data].data
|
|
|
|
let p2 = res.storage[b.data].data
|
|
|
|
result = p1 < p2
|
|
|
|
|
|
|
|
res.cmp = peerCmp
|
|
|
|
result = res
|
|
|
|
|
|
|
|
proc len*[A, B](pool: PeerPool[A, B]): int =
|
|
|
|
## Returns number of registered peers in PeerPool ``pool``. This number
|
|
|
|
## includes all the peers (acquired and available).
|
|
|
|
result = len(pool.registry)
|
|
|
|
|
|
|
|
proc lenAvailable*[A, B](pool: PeerPool[A, B],
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): int {.inline.} =
|
|
|
|
## Returns number of available peers in PeerPool ``pool`` which satisfies
|
|
|
|
## filter ``filter``.
|
|
|
|
if PeerType.Incoming in filter:
|
|
|
|
result = result + len(pool.incQueue)
|
|
|
|
if PeerType.Outgoing in filter:
|
|
|
|
result = result + len(pool.outQueue)
|
|
|
|
|
|
|
|
proc lenAcquired*[A, B](pool: PeerPool[A, B],
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): int {.inline.} =
|
|
|
|
## Returns number of acquired peers in PeerPool ``pool`` which satisifies
|
|
|
|
## filter ``filter``.
|
|
|
|
if PeerType.Incoming in filter:
|
|
|
|
result = result + pool.acqIncPeersCount
|
|
|
|
if PeerType.Outgoing in filter:
|
|
|
|
result = result + pool.acqOutPeersCount
|
|
|
|
|
|
|
|
proc deletePeer*[A, B](pool: PeerPool[A, B], peer: A, force = false): bool =
|
|
|
|
## Remove ``peer`` from PeerPool ``pool``.
|
|
|
|
##
|
|
|
|
## Deletion occurs immediately only if peer is available, otherwise it will
|
|
|
|
## be deleted only when peer will be released. You can change this behavior
|
|
|
|
## with ``force`` option.
|
|
|
|
mixin getKey
|
|
|
|
var key = getKey(peer)
|
|
|
|
if pool.registry.hasKey(key):
|
|
|
|
let pindex = pool.registry[key].data
|
|
|
|
var item = addr(pool.storage[pindex])
|
|
|
|
if (PeerFlags.Acquired in item[].flags):
|
|
|
|
if not(force):
|
|
|
|
item[].flags.incl(PeerFlags.DeleteOnRelease)
|
|
|
|
else:
|
|
|
|
if item[].peerType == PeerType.Incoming:
|
|
|
|
dec(pool.curIncPeersCount)
|
|
|
|
dec(pool.acqIncPeersCount)
|
|
|
|
elif item[].peerType == PeerType.Outgoing:
|
|
|
|
dec(pool.curOutPeersCount)
|
|
|
|
dec(pool.acqOutPeersCount)
|
|
|
|
# Cleanup storage with default item, and removing key from hashtable.
|
|
|
|
pool.storage[pindex] = PeerItem[A]()
|
|
|
|
pool.registry.del(key)
|
|
|
|
else:
|
|
|
|
if item[].peerType == PeerType.Incoming:
|
|
|
|
# If peer is available, then its copy present in heapqueue, so we need
|
|
|
|
# to remove it.
|
|
|
|
for i in 0 ..< len(pool.incQueue):
|
|
|
|
if pool.incQueue[i].data == pindex:
|
|
|
|
pool.incQueue.del(i)
|
|
|
|
break
|
|
|
|
dec(pool.curIncPeersCount)
|
|
|
|
elif item[].peerType == PeerType.Outgoing:
|
|
|
|
# If peer is available, then its copy present in heapqueue, so we need
|
|
|
|
# to remove it.
|
|
|
|
for i in 0 ..< len(pool.outQueue):
|
|
|
|
if pool.outQueue[i].data == pindex:
|
|
|
|
pool.outQueue.del(i)
|
|
|
|
break
|
|
|
|
dec(pool.curOutPeersCount)
|
|
|
|
# Cleanup storage with default item, and removing key from hashtable.
|
|
|
|
pool.storage[pindex] = PeerItem[A]()
|
|
|
|
pool.registry.del(key)
|
|
|
|
|
|
|
|
result = true
|
|
|
|
|
|
|
|
proc addPeer*[A, B](pool: PeerPool[A, B], peer: A, peerType: PeerType): bool =
|
|
|
|
## Add peer ``peer`` of type ``peerType`` to PeerPool ``pool``.
|
|
|
|
##
|
2020-01-21 18:30:21 +00:00
|
|
|
## Procedure returns ``false`` in case
|
|
|
|
## * if ``peer`` is already closed.
|
|
|
|
## * if ``pool`` currently has a maximum of peers.
|
|
|
|
## * if ``pool`` currently has a maximum of `Incoming` or `Outgoing` peers.
|
|
|
|
##
|
|
|
|
## Procedure returns ``true`` on success.
|
2019-11-25 20:18:02 +00:00
|
|
|
mixin getKey, getFuture
|
|
|
|
|
2020-01-21 18:30:21 +00:00
|
|
|
if peer.getFuture().finished:
|
|
|
|
return false
|
|
|
|
|
2019-11-25 20:18:02 +00:00
|
|
|
if len(pool.registry) >= pool.maxPeersCount:
|
|
|
|
return false
|
|
|
|
|
|
|
|
var item = PeerItem[A](data: peer, peerType: peerType,
|
|
|
|
index: len(pool.storage))
|
|
|
|
var key = getKey(peer)
|
|
|
|
|
|
|
|
if not(pool.registry.hasKey(key)):
|
|
|
|
pool.storage.add(item)
|
|
|
|
var pitem = addr(pool.storage[^1])
|
|
|
|
let pindex = PeerIndex(data: item.index, cmp: pool.cmp)
|
|
|
|
pool.registry[key] = pindex
|
|
|
|
|
|
|
|
proc onPeerClosed(udata: pointer) {.gcsafe.} =
|
|
|
|
discard pool.deletePeer(peer)
|
|
|
|
|
|
|
|
pitem[].data.getFuture().addCallback(onPeerClosed)
|
|
|
|
|
|
|
|
if peerType == PeerType.Incoming:
|
|
|
|
if pool.curIncPeersCount >= pool.maxIncPeersCount:
|
|
|
|
return false
|
|
|
|
else:
|
|
|
|
inc(pool.curIncPeersCount)
|
|
|
|
pool.incQueue.push(pindex)
|
|
|
|
pool.incNeEvent.fire()
|
|
|
|
elif peerType == PeerType.Outgoing:
|
|
|
|
if pool.curOutPeersCount >= pool.maxOutPeersCount:
|
|
|
|
return false
|
|
|
|
else:
|
|
|
|
inc(pool.curOutPeersCount)
|
|
|
|
pool.outQueue.push(pindex)
|
|
|
|
pool.outNeEvent.fire()
|
|
|
|
|
|
|
|
result = true
|
|
|
|
|
|
|
|
proc addIncomingPeer*[A, B](pool: PeerPool[A, B], peer: A): bool {.inline.} =
|
|
|
|
## Add incoming peer ``peer`` to PeerPool ``pool``.
|
|
|
|
##
|
|
|
|
## Returns ``true`` on success.
|
|
|
|
result = pool.addPeer(peer, PeerType.Incoming)
|
|
|
|
|
|
|
|
proc addOutgoingPeer*[A, B](pool: PeerPool[A, B], peer: A): bool {.inline.} =
|
|
|
|
## Add outgoing peer ``peer`` to PeerPool ``pool``.
|
|
|
|
##
|
|
|
|
## Returns ``true`` on success.
|
|
|
|
result = pool.addPeer(peer, PeerType.Outgoing)
|
|
|
|
|
|
|
|
proc acquire*[A, B](pool: PeerPool[A, B],
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): Future[A] {.async.} =
|
|
|
|
## Acquire peer from PeerPool ``pool``, which match the filter ``filter``.
|
|
|
|
doAssert(filter != {}, "Filter must not be empty")
|
|
|
|
while true:
|
|
|
|
var count = 0
|
|
|
|
if PeerType.Incoming in filter:
|
|
|
|
count = count + len(pool.incQueue)
|
|
|
|
if PeerType.Outgoing in filter:
|
|
|
|
count = count + len(pool.outQueue)
|
|
|
|
if count == 0:
|
|
|
|
await pool.waitEvent(filter)
|
|
|
|
else:
|
|
|
|
var item = pool.getItem(filter)
|
|
|
|
doAssert(PeerFlags.Acquired notin item[].flags)
|
|
|
|
item[].flags.incl(PeerFlags.Acquired)
|
|
|
|
result = item[].data
|
|
|
|
break
|
|
|
|
|
2020-01-21 18:30:21 +00:00
|
|
|
proc acquireNoWait*[A, B](pool: PeerPool[A, B],
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): A =
|
|
|
|
doAssert(filter != {}, "Filter must not be empty")
|
|
|
|
var count = 0
|
|
|
|
if PeerType.Incoming in filter:
|
|
|
|
count = count + len(pool.incQueue)
|
|
|
|
if PeerType.Outgoing in filter:
|
|
|
|
count = count + len(pool.outQueue)
|
|
|
|
if count < 1:
|
|
|
|
raise newException(PeerPoolError, "Not enough peers in pool")
|
|
|
|
var item = pool.getItem(filter)
|
|
|
|
doAssert(PeerFlags.Acquired notin item[].flags)
|
|
|
|
item[].flags.incl(PeerFlags.Acquired)
|
|
|
|
result = item[].data
|
|
|
|
|
2019-11-25 20:18:02 +00:00
|
|
|
proc release*[A, B](pool: PeerPool[A, B], peer: A) =
|
|
|
|
## Release peer ``peer`` back to PeerPool ``pool``
|
|
|
|
mixin getKey
|
|
|
|
var key = getKey(peer)
|
|
|
|
var titem = pool.registry.getOrDefault(key, PeerIndex(data: -1))
|
|
|
|
if titem.data >= 0:
|
|
|
|
let pindex = titem.data
|
|
|
|
var item = addr(pool.storage[pindex])
|
|
|
|
if PeerFlags.Acquired in item[].flags:
|
|
|
|
item[].flags.excl(PeerFlags.Acquired)
|
|
|
|
if PeerFlags.DeleteOnRelease in item[].flags:
|
|
|
|
if item[].peerType == PeerType.Incoming:
|
|
|
|
dec(pool.curIncPeersCount)
|
|
|
|
dec(pool.acqIncPeersCount)
|
|
|
|
elif item[].peerType == PeerType.Outgoing:
|
|
|
|
dec(pool.curOutPeersCount)
|
|
|
|
dec(pool.acqOutPeersCount)
|
|
|
|
pool.storage[pindex] = PeerItem[A]()
|
|
|
|
pool.registry.del(key)
|
|
|
|
else:
|
|
|
|
if item[].peerType == PeerType.Incoming:
|
|
|
|
pool.incQueue.push(titem)
|
|
|
|
dec(pool.acqIncPeersCount)
|
|
|
|
elif item[].peerType == PeerType.Outgoing:
|
|
|
|
pool.outQueue.push(titem)
|
|
|
|
dec(pool.acqOutPeersCount)
|
|
|
|
pool.fireEvent(item[])
|
|
|
|
|
|
|
|
proc release*[A, B](pool: PeerPool[A, B], peers: openarray[A]) {.inline.} =
|
|
|
|
## Release array of peers ``peers`` back to PeerPool ``pool``.
|
|
|
|
for item in peers:
|
|
|
|
pool.release(item)
|
|
|
|
|
|
|
|
proc acquire*[A, B](pool: PeerPool[A, B],
|
|
|
|
number: int,
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): Future[seq[A]] {.async.} =
|
|
|
|
## Acquire ``number`` number of peers from PeerPool ``pool``, which match the
|
|
|
|
## filter ``filter``.
|
|
|
|
doAssert(filter != {}, "Filter must not be empty")
|
|
|
|
var peers = newSeq[A]()
|
|
|
|
try:
|
|
|
|
if number > 0:
|
|
|
|
while true:
|
|
|
|
if len(peers) >= number:
|
|
|
|
break
|
|
|
|
var count = 0
|
|
|
|
if PeerType.Incoming in filter:
|
|
|
|
count = count + len(pool.incQueue)
|
|
|
|
if PeerType.Outgoing in filter:
|
|
|
|
count = count + len(pool.outQueue)
|
|
|
|
if count == 0:
|
|
|
|
await pool.waitEvent(filter)
|
|
|
|
else:
|
|
|
|
var item = pool.getItem(filter)
|
|
|
|
doAssert(PeerFlags.Acquired notin item[].flags)
|
|
|
|
item[].flags.incl(PeerFlags.Acquired)
|
|
|
|
peers.add(item[].data)
|
|
|
|
except CancelledError:
|
|
|
|
# If we got cancelled, we need to return all the acquired peers back to
|
|
|
|
# pool.
|
|
|
|
for item in peers:
|
|
|
|
pool.release(item)
|
|
|
|
peers.setLen(0)
|
|
|
|
raise
|
|
|
|
result = peers
|
|
|
|
|
2020-01-21 18:30:21 +00:00
|
|
|
proc acquireNoWait*[A, B](pool: PeerPool[A, B],
|
|
|
|
number: int,
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): seq[A] =
|
|
|
|
## Acquire ``number`` number of peers from PeerPool ``pool``, which match the
|
|
|
|
## filter ``filter``.
|
|
|
|
doAssert(filter != {}, "Filter must not be empty")
|
|
|
|
var peers = newSeq[A]()
|
|
|
|
var count = 0
|
|
|
|
if PeerType.Incoming in filter:
|
|
|
|
count = count + len(pool.incQueue)
|
|
|
|
if PeerType.Outgoing in filter:
|
|
|
|
count = count + len(pool.outQueue)
|
|
|
|
if count < number:
|
|
|
|
raise newException(PeerPoolError, "Not enough peers in pool")
|
|
|
|
for i in 0 ..< number:
|
|
|
|
var item = pool.getItem(filter)
|
|
|
|
doAssert(PeerFlags.Acquired notin item[].flags)
|
|
|
|
item[].flags.incl(PeerFlags.Acquired)
|
|
|
|
peers.add(item[].data)
|
|
|
|
result = peers
|
|
|
|
|
2019-11-25 20:18:02 +00:00
|
|
|
proc acquireIncomingPeer*[A, B](pool: PeerPool[A, B]): Future[A] {.inline.} =
|
|
|
|
## Acquire single incoming peer from PeerPool ``pool``.
|
|
|
|
pool.acquire({PeerType.Incoming})
|
|
|
|
|
|
|
|
proc acquireOutgoingPeer*[A, B](pool: PeerPool[A, B]): Future[A] {.inline.} =
|
|
|
|
## Acquire single outgoing peer from PeerPool ``pool``.
|
|
|
|
pool.acquire({PeerType.Outgoing})
|
|
|
|
|
|
|
|
proc acquireIncomingPeers*[A, B](pool: PeerPool[A, B],
|
|
|
|
number: int): Future[seq[A]] {.inline.} =
|
|
|
|
## Acquire ``number`` number of incoming peers from PeerPool ``pool``.
|
|
|
|
pool.acquire(number, {PeerType.Incoming})
|
|
|
|
|
|
|
|
proc acquireOutgoingPeers*[A, B](pool: PeerPool[A, B],
|
|
|
|
number: int): Future[seq[A]] {.inline.} =
|
|
|
|
## Acquire ``number`` number of outgoing peers from PeerPool ``pool``.
|
|
|
|
pool.acquire(number, {PeerType.Outgoing})
|
|
|
|
|
|
|
|
iterator peers*[A, B](pool: PeerPool[A, B],
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): A =
|
|
|
|
## Iterate over sorted list of peers.
|
|
|
|
##
|
|
|
|
## All peers will be sorted by equation `>`(Peer1, Peer2), so biggest values
|
|
|
|
## will be first.
|
|
|
|
var sorted = initHeapQueue[PeerIndex]()
|
|
|
|
for i in 0 ..< len(pool.storage):
|
|
|
|
if pool.storage[i].peerType in filter:
|
|
|
|
sorted.push(PeerIndex(data: i, cmp: pool.cmp))
|
|
|
|
while len(sorted) > 0:
|
|
|
|
let pindex = sorted.pop().data
|
|
|
|
yield pool.storage[pindex].data
|
|
|
|
|
|
|
|
iterator availablePeers*[A, B](pool: PeerPool[A, B],
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): A =
|
|
|
|
## Iterate over sorted list of available peers.
|
|
|
|
##
|
|
|
|
## All peers will be sorted by equation `>`(Peer1, Peer2), so biggest values
|
|
|
|
## will be first.
|
|
|
|
var sorted = initHeapQueue[PeerIndex]()
|
|
|
|
for i in 0 ..< len(pool.storage):
|
|
|
|
if (PeerFlags.Acquired notin pool.storage[i].flags) and
|
|
|
|
(pool.storage[i].peerType in filter):
|
|
|
|
sorted.push(PeerIndex(data: i, cmp: pool.cmp))
|
|
|
|
while len(sorted) > 0:
|
|
|
|
let pindex = sorted.pop().data
|
|
|
|
yield pool.storage[pindex].data
|
|
|
|
|
|
|
|
iterator acquiredPeers*[A, B](pool: PeerPool[A, B],
|
|
|
|
filter = {PeerType.Incoming,
|
|
|
|
PeerType.Outgoing}): A =
|
|
|
|
## Iterate over sorted list of acquired (non-available) peers.
|
|
|
|
##
|
|
|
|
## All peers will be sorted by equation `>`(Peer1, Peer2), so biggest values
|
|
|
|
## will be first.
|
|
|
|
var sorted = initHeapQueue[PeerIndex]()
|
|
|
|
for i in 0 ..< len(pool.storage):
|
|
|
|
if (PeerFlags.Acquired in pool.storage[i].flags) and
|
|
|
|
(pool.storage[i].peerType in filter):
|
|
|
|
sorted.push(PeerIndex(data: i, cmp: pool.cmp))
|
|
|
|
while len(sorted) > 0:
|
|
|
|
let pindex = sorted.pop().data
|
|
|
|
yield pool.storage[pindex].data
|
|
|
|
|
|
|
|
proc `[]`*[A, B](pool: PeerPool[A, B], key: B): A {.inline.} =
|
|
|
|
## Retrieve peer with key ``key`` from PeerPool ``pool``.
|
|
|
|
let pindex = pool.registry[key]
|
|
|
|
result = pool.storage[pindex.data]
|
|
|
|
|
|
|
|
proc `[]`*[A, B](pool: var PeerPool[A, B], key: B): var A {.inline.} =
|
|
|
|
## Retrieve peer with key ``key`` from PeerPool ``pool``.
|
|
|
|
let pindex = pool.registry[key]
|
|
|
|
result = pool.storage[pindex.data].data
|
|
|
|
|
|
|
|
proc hasPeer*[A, B](pool: PeerPool[A, B], key: B): bool {.inline.} =
|
|
|
|
## Returns ``true`` if peer with ``key`` present in PeerPool ``pool``.
|
|
|
|
result = pool.registry.hasKey(key)
|
|
|
|
|
|
|
|
proc getOrDefault*[A, B](pool: PeerPool[A, B], key: B): A {.inline.} =
|
|
|
|
## Retrieves the peer from PeerPool ``pool`` using key ``key``. If peer is
|
|
|
|
## not present, default initialization value for type ``A`` is returned
|
|
|
|
## (e.g. 0 for any integer type).
|
|
|
|
let pindex = pool.registry.getOrDefault(key, PeerIndex(data: -1))
|
|
|
|
if pindex.data >= 0:
|
|
|
|
result = pool.storage[pindex.data].data
|
|
|
|
|
|
|
|
proc getOrDefault*[A, B](pool: PeerPool[A, B], key: B,
|
|
|
|
default: A): A {.inline.} =
|
|
|
|
## Retrieves the peer from PeerPool ``pool`` using key ``key``. If peer is
|
|
|
|
## not present, default value ``default`` is returned.
|
|
|
|
let pindex = pool.registry.getOrDefault(key, PeerIndex(data: -1))
|
|
|
|
if pindex.data >= 0:
|
|
|
|
result = pool.storage[pindex.data].data
|
|
|
|
else:
|
|
|
|
result = default
|
|
|
|
|
|
|
|
proc clear*[A, B](pool: PeerPool[A, B]) =
|
|
|
|
## Performs PeerPool's ``pool`` storage and counters reset.
|
|
|
|
pool.incQueue.clear()
|
|
|
|
pool.outQueue.clear()
|
|
|
|
pool.registry.clear()
|
|
|
|
for i in 0 ..< len(pool.storage):
|
|
|
|
pool.storage[i] = PeerItem[A]()
|
|
|
|
pool.storage.setLen(0)
|
|
|
|
pool.curIncPeersCount = 0
|
|
|
|
pool.curOutPeersCount = 0
|
|
|
|
pool.acqIncPeersCount = 0
|
|
|
|
pool.acqOutPeersCount = 0
|
|
|
|
|
|
|
|
proc clearSafe*[A, B](pool: PeerPool[A, B]) {.async.} =
|
|
|
|
## Performs "safe" clear. Safe means that it first acquires all the peers
|
|
|
|
## in PeerPool, and only after that it will reset storage.
|
|
|
|
var acquired = newSeq[A]()
|
|
|
|
while len(pool.registry) > len(acquired):
|
|
|
|
var peers = await pool.acquire(len(pool.registry) - len(acquired))
|
|
|
|
for item in peers:
|
|
|
|
acquired.add(item)
|
|
|
|
pool.clear()
|