* treat connected/handshaked peers separately in the discoverer, save
"original" address for connected ones, it can be a name instead of IP and
it's important to keep it to avoid reconnections
* store name->IP mapping for seeds if and when they're connected to avoid
reconnections
* block seed if it's detected to be our own node (which is often the case for
small private networks)
* add an event for handshaked peers in the server, connected but
non-handshaked ones are not really helpful for MinPeers or GetAddr logic
Fixes#2796.
Every 1000 blocks seems to be OK for big networks (that only had done some
initial requests previously and then effectively never requested addresses
again because there was a sufficient number of addresses), won't hurt smaller
ones as well (that effectively keep doing this on every connect/disconnect,
peer changes are very rare there, but when they happen we want to have some
quick reaction to these changes).
32 is a very good number, but we all know 42 is a better one. And it can even
be proven by tests with higher peaking TPS values.
You may wonder why is it so good? Because we're using packet-switching
networks mostly and a packet is a packet almost irrespectively of how bit it
is. Yet a packet has some maximum possible size (hi, MTU) and this size most
of the time is 1500 (or a little less than that, hi VPN). Subtract IP header
(20 for IPv4 or 40 for IPv6 not counting options), TCP header (another 20) and
Neo message/payload headers (~8 for this case) and we have just a little more
than 1400 bytes for our dear hashes. Which means that in a single packet most
of the time we can have 42-44 of them, maybe 45. Choosing between these
numbers is not hard then.
We have AttemptConnPeers that is closely related, the more we have there the
bigger the network supposedly is, so it's much better than magic minPoolCount.
Sometimes we already have it, but it's not yet processed, so we can save on
getdata request. It only affects very high-speed networks like 4-1 scenario
and it doesn't affect it a lot, but still we can do it.
This is not exactly the protocol-level batching as was tried in #1770 and
proposed by neo-project/neo#2365, but it's a TCP-level change in that we now
Write() a set of messages and given that Go sets up TCP sockets with
TCP_NODELAY by default this is a substantial change, we have less packets
generated with the same amount of data. It doesn't change anything on properly
connected networks, but the ones with delays benefit from it a lot.
This also improves queueing because we no longer generate 32 messages to
deliver on transaction's GetData, it's just one stream of bytes with 32
messages inside.
Do the same with GetBlocksByIndex, we can have a lot of messages there too.
But don't forget about potential peer DoS attacks, if a peer is to request a
lot of big blocks we need to flush them before we process the whole set.
This allows to naturally scale transaction processing if we have some peer
that is sending a lot of them while others are mostly silent. It also can help
somewhat in the event we have 50 peers that all send transactions. 4+1
scenario benefits a lot from it, while 7+2 slows down a little. Delayed
scenarios don't care.
Surprisingly, this also makes disconnects (#2744) much more rare, 4-node
scenario almost never sees it now. Most probably this is the case where peers
affect each other a lot, single-threaded transaction receiver can be slow
enough to trigger some timeout in getdata handler of its peer (because it
tries to push a number of replies).
It makes sense in general (further narrowing down the time window when
transactions are processed by consensus thread) and it improves block times a
little too, especially in the 7+2 scenario.
Related to #2744.
Until the consensus process starts for a new block and until it really needs
some transactions we can spare some cycles by not delivering transactions to
it. In tests this doesn't affect TPS, but makes block delays a bit more
stable. Related to #2744, I think it also may cause timeouts during
transaction processing (waiting on the consensus process channel while it does
something dBFT-related).
When the network is big enough, MinPeers may be suboptimal for good network
connectivity, but if we know the network size we can do some estimation on the
number of sufficient peers.
They can fail right in the getPeers or they can fail later when packet send
is attempted. Of course they can complete handshake in-between these events,
but most likely they won't and we'll waste more resources on this attempt. So
rule out bad peers immediately.
Drop EnqueueP2PPacket, replace EnqueueHPPacket with EnqueueHPMessage. We use
Enqueue* when we have a specific per-peer message, it makes zero sense
duplicating serialization code for it (unlike Broadcast*).
Follow the general rules of broadcasts, even though it's somewhat different
from Inv, we just want to get some reply from our neighbors to see if we're
behind. We don't strictly need all neighbors for it.
We have a number of queues for different purposes:
* regular broadcast queue
* direct p2p queue
* high-priority queue
And two basic egress scenarios:
* direct p2p messages (replies to requests in Server's handle* methods)
* broadcasted messages
Low priority broadcasted messages:
* transaction inventories
* block inventories
* notary inventories
* non-consensus extensibles
High-priority broadcasted messages:
* consensus extensibles
* getdata transaction requests from consensus process
* getaddr requests
P2P messages are a bit more complicated, most of the time they use p2p queue,
but extensible message requests/replies use HP queue.
Server's handle* code is run from Peer's handleIncoming, every peer has this
thread that handles incoming messages. When working with the peer it's
important to reply to requests and blocking this thread until we send (queue)
a reply is fine, if the peer is slow we just won't get anything new from
it. The queue used is irrelevant wrt this issue.
Broadcasted messages are radically different, we want them to be delivered to
many peers, but we don't care about specific ones. If it's delivered to 2/3 of
the peers we're fine, if it's delivered to more of them --- it's not an
issue. But doing this fairly is not an easy thing, current code tries performing
unblocked sends and if this doesn't yield enough results it then blocks (but
has a timeout, we can't wait indefinitely). But it does so in sequential
manner, once the peer is chosen the code will wait for it (and only it) until
timeout happens.
What can be done instead is an attempt to push the message to all of the peers
simultaneously (or close to that). If they all deliver --- OK, if some block
and wait then we can wait until _any_ of them pushes the message through (or
global timeout happens, we still can't wait forever). If we have enough
deliveries then we can cancel pending ones and it's again not an error if
these canceled threads still do their job.
This makes the system more dynamic and adds some substantial processing
overhead, but it's a networking code, any of this overhead is much lower than
the actual packet delivery time. It also allows to spread the load more
fairly, if there is any spare queue it'll get the packet and release the
broadcaster. On the next broadcast iteration another peer is more likely to be
chosen just because it didn't get a message previously (and had some time to
deliver already queued messages).
It works perfectly in tests, with optimal networking conditions we have much
better block times and TPS increases by 5-25%% depending on the scenario.
I'd go as far as to say that it fixes the original problem of #2678, because
in this particular scenario we have empty queues in ~100% of the cases and
this new logic will likely lead to 100% fan out in this case (cancelation just
won't happen fast enough). But when the load grows and there is some waiting
in the queue it will optimize out the slowest links.
Peers can be slow, very slow, slow enough to affect node's regular
operation. We can't wait for them indefinitely, there has to be a timeout for
send operations.
This patch uses TimePerBlock as a reference for its timeout. It's relatively
big and it doesn't affect tests much, 4+1 scenarios tend to perform a little
worse with while 7+2 scenarios work a little better. The difference is in some
percents, but all of these tests easily have 10-15% variations from run to
run.
It's an important step in making our gossip better because we can't have any
behavior where neighbors directly block the node forever, refs. #2678 and
Blockchain's notificationDispatcher sends events to channels and these
channels must be read from. Unfortunately, regular service shutdown procedure
does unsubscription first (outside of the read loop) and only then drains the
channel. While it waits for unsubscription request to be accepted
notificationDispatcher can try pushing more data into the same channel which
will lead to a deadlock. Reading in the same method solves this, any number of
events can be pushed until unsub channel accepts the data.
Unsubscribe and drain first, then return from the Shutdown method. It's
important wrt to subsequent chain shutdown process (normally it's closed right
after the network server).
Some tests are failing on Windows due to slow runners with errors like the following:
```
2022-02-09T17:11:20.3127016Z --- FAIL: TestGetData/transaction (1.82s)
2022-02-09T17:11:20.3127385Z server_test.go:500:
2022-02-09T17:11:20.3127878Z Error Trace: server_test.go:500
2022-02-09T17:11:20.3128533Z server_test.go:520
2022-02-09T17:11:20.3128978Z Error: Condition never satisfied
2022-02-09T17:11:20.3129479Z Test: TestGetData/transaction
```
Consensus can require conflicting transactions and it can require more
transactions than mempool can fit, all of this should work. Transactions will
be checked anyway using its secondary mempool. See the scenario from #668.
Notice that it makes the node accept Extensible payloads with any category
which is the same way C# node works. We're trusting Extensible senders,
improper payloads are harmless until they DoS the network, but we have some
protections against that too (and spamming with proper category doesn't differ
a lot).
Use circular buffer which is a bit more appropriate. The problem is that
priority queue accepts and stores equal items which wastes memory even in
normal usage scenario, but it's especially dangerous if the node is stuck for
some reason. In this case it'll accept from peers and put into queue the same
blocks again and again leaking memory up to OOM condition.
Notice that queue length calculation might be wrong in case circular buffer
wraps, but it's not very likely to happen (usually blocks not coming from the
queue are added by consensus and it's not very fast in doing so).
In this commit:
1. Request unknown MPT nodes from peers. Note, that StateSync module itself
shouldn't be responsible for nodes requests, that's a server duty.
2. Do not request the same node twice, check if it is in storage
already. If so, then the only thing remaining is to update refcounter.
It requires only two methods from Blockchainer: AddBlock and
BlockHeight. New interface will allow to easily reuse the block queue
for state exchange purposes.
Refactor code and be fine with sending to just 2/3 of proper peers. Previously
it was an edge case, but it can be a normal thing to do also as broadcasting
to everyone is obviously too expensive and excessive (hi, #608).
Baseline (four node, 10 workers):
RPS 8180.760 8137.822 7858.358 7820.011 8051.076 ≈ 8010 ± 2.04%
TPS 7819.831 7521.172 7519.023 7242.965 7426.000 ≈ 7506 ± 2.78%
CPU % 41.983 38.775 40.606 39.375 35.537 ≈ 39.3 ± 6.15%
Mem MB 2947.189 2743.658 2896.688 2813.276 2863.108 ≈ 2853 ± 2.74%
Patched:
RPS 9714.567 9676.102 9358.609 9371.408 9301.372 ≈ 9484 ± 2.05% ↑ 18.40%
TPS 8809.796 8796.854 8534.754 8661.158 8426.162 ≈ 8646 ± 1.92% ↑ 15.19%
CPU % 44.980 45.018 33.640 29.645 43.830 ≈ 39.4 ± 18.41% ↑ 0.25%
Mem MB 2989.078 2976.577 2306.185 2351.929 2910.479 ≈ 2707 ± 12.80% ↓ 5.12%
There is a nuance with this patch however. While typically it works the way
outlined above, sometimes it works like this:
RPS ≈ 6734.368
TPS ≈ 6299.332
CPU ≈ 25.552%
Mem ≈ 2706.046MB
And that's because the log looks like this:
DeltaTime, TransactionsCount, TPS
5014, 44212, 8817.710
5163, 49690, 9624.249
5166, 49523, 9586.334
5189, 49693, 9576.604
5198, 49339, 9491.920
5147, 49559, 9628.716
5192, 49680, 9568.567
5163, 49750, 9635.871
5183, 49189, 9490.450
5159, 49653, 9624.540
5167, 47945, 9279.079
5179, 2051, 396.022
5015, 4, 0.798
5004, 0, 0.000
5003, 0, 0.000
5003, 0, 0.000
5003, 0, 0.000
5003, 0, 0.000
5004, 0, 0.000
5003, 2925, 584.649
5040, 49099, 9741.865
5161, 49718, 9633.404
5170, 49228, 9521.857
5179, 49773, 9610.543
5167, 47253, 9145.152
5202, 49788, 9570.934
5177, 47704, 9214.603
5209, 46610, 8947.975
5249, 49156, 9364.831
5163, 18284, 3541.352
5072, 174, 34.306
On a network with 4 CNs and 1 RPC node there is 1/256 probability that a block
won't be broadcasted to RPC node, so it won't see it until ping timeout kicks
in. While it doesn't see a block it can't accept new incoming transactions so
the bench gets stuck basically. To me that's an acceptable trade-off because
normal networks are much larger than that and the effect of this patch is way
more important there, but still that's what we have and we need to take into
account.
send() can return errStateMismatch, errGone and errBusy. errGone means the
peer is dead and it won't ever be active again, it doesn't make sense retrying
sends to it. errStateMismatch is technically "not yet ready", but we can't
wait for it either, no one knows how much will it take to complete
handshake. So only errBusy means we can retry.
So keep track of dead peers and adjust tries counting appropriately.
It doesn't change much, we can't magically get more valid peers and if some
die while we're iterating we'd detect that by an error returned from send().
