Follow neo-project/neo#2807. Notice that this data is not cached, our previous
implementation wasn't too and it shouldn't be a problem (not on the hot path).
They can stay in the memory pool forever because consensus process will never
accept these transactions (and maybe even block consensus process at all).
We're paging these hashes, so we need a previous full page and a current one
plus some cache for various requests. Storing 1M of hashes is 32M of memory
and it grows quickly. It also seriously affects node startup time, most of
what it's doing is reading these hashes, the longer the chain the more time it
needs to do that.
Notice that this doesn't change the underlying DB scheme in any way.
If we only have genesis block (or <2000 headers) then we might as well use
generic logic below with zero targetHash because genesis block has zero
PrevHash (and its hash will naturally be the last on the chain going
backwards).
Sometimes it can be hard to persist all changes at ones, the process
can take almost all RAM and a lot of time. Here's the example of reset
for mainnet from 2.4M to 1:
```
anna@kiwi:~/Documents/GitProjects/nspcc-dev/neo-go$ ./bin/neo-go db reset -m --height 1
2022-11-20T17:16:48.236+0300 INFO MaxBlockSize is not set or wrong, setting default value {"MaxBlockSize": 262144}
2022-11-20T17:16:48.236+0300 INFO MaxBlockSystemFee is not set or wrong, setting default value {"MaxBlockSystemFee": 900000000000}
2022-11-20T17:16:48.237+0300 INFO MaxTransactionsPerBlock is not set or wrong, using default value {"MaxTransactionsPerBlock": 512}
2022-11-20T17:16:48.237+0300 INFO MaxValidUntilBlockIncrement is not set or wrong, using default value {"MaxValidUntilBlockIncrement": 5760}
2022-11-20T17:16:48.240+0300 INFO restoring blockchain {"version": "0.2.6"}
2022-11-20T17:16:48.297+0300 INFO initialize state reset {"target height": 1}
2022-11-20T17:16:48.300+0300 INFO trying to reset blocks, transactions and AERs
2022-11-20T17:19:29.313+0300 INFO blocks, transactions ans AERs are reset {"took": "2m41.015126493s", "keys": 3958420}
...
```
To avoid OOM killer, split blocks reset into multiple stages. It increases
operation time due to intermediate DB persists, but makes things cleaner, the
result for almost the same DB height with the new approach:
```
anna@kiwi:~/Documents/GitProjects/nspcc-dev/neo-go$ ./bin/neo-go db reset -m --height 1
2022-11-20T17:39:42.023+0300 INFO MaxBlockSize is not set or wrong, setting default value {"MaxBlockSize": 262144}
2022-11-20T17:39:42.023+0300 INFO MaxBlockSystemFee is not set or wrong, setting default value {"MaxBlockSystemFee": 900000000000}
2022-11-20T17:39:42.023+0300 INFO MaxTransactionsPerBlock is not set or wrong, using default value {"MaxTransactionsPerBlock": 512}
2022-11-20T17:39:42.023+0300 INFO MaxValidUntilBlockIncrement is not set or wrong, using default value {"MaxValidUntilBlockIncrement": 5760}
2022-11-20T17:39:42.026+0300 INFO restoring blockchain {"version": "0.2.6"}
2022-11-20T17:39:42.071+0300 INFO initialize state reset {"target height": 1}
2022-11-20T17:39:42.073+0300 INFO trying to reset blocks, transactions and AERs
2022-11-20T17:40:11.735+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 1, "took": "29.66363737s", "keys": 210973}
2022-11-20T17:40:33.574+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 2, "took": "21.839208683s", "keys": 241203}
2022-11-20T17:41:29.325+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 3, "took": "55.750698386s", "keys": 250593}
2022-11-20T17:42:12.532+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 4, "took": "43.205892757s", "keys": 321896}
2022-11-20T17:43:07.978+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 5, "took": "55.445398156s", "keys": 334822}
2022-11-20T17:43:35.603+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 6, "took": "27.625292032s", "keys": 317131}
2022-11-20T17:43:51.747+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 7, "took": "16.144359017s", "keys": 355832}
2022-11-20T17:44:05.176+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 8, "took": "13.428733899s", "keys": 357690}
2022-11-20T17:44:32.895+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 9, "took": "27.718548783s", "keys": 393356}
2022-11-20T17:44:51.814+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 10, "took": "18.917954658s", "keys": 366492}
2022-11-20T17:45:07.208+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 11, "took": "15.392642196s", "keys": 326030}
2022-11-20T17:45:18.776+0300 INFO intermediate batch of removed blocks, transactions and AERs is persisted {"batches persisted": 12, "took": "11.568255716s", "keys": 299884}
2022-11-20T17:45:25.862+0300 INFO last batch of removed blocks, transactions and AERs is persisted {"batches persisted": 13, "took": "7.086079594s", "keys": 190399}
2022-11-20T17:45:25.862+0300 INFO blocks, transactions ans AERs are reset {"took": "5m43.791214084s", "overall persisted keys": 3966301}
...
```
We need to keep the headers information consistent with header batches
and headers. This comit fixes the bug with failing blockchain
initialization on recovering from state reset interrupted after the
second stage (blocks/txs/AERs removal):
```
anna@kiwi:~/Documents/GitProjects/nspcc-dev/neo-go$ ./bin/neo-go db reset -t --height 83000
2022-11-20T16:28:29.437+0300 INFO MaxValidUntilBlockIncrement is not set or wrong, using default value {"MaxValidUntilBlockIncrement": 5760}
2022-11-20T16:28:29.440+0300 INFO restoring blockchain {"version": "0.2.6"}
failed to create Blockchain instance: could not initialize blockchain: could not get header 1898cd356a4a2688ed1c6c7ba1fd6ba7d516959d8add3f8dd26232474d4539bd: key not found
```
Don't use cache because it's not yet initialized. Also, perform
safety checks only if state reset wasn't yet started. These fixes
alloww to solve the following problem while recovering from
interrupted state reset:
```
anna@kiwi:~/Documents/GitProjects/nspcc-dev/neo-go$ ./bin/neo-go db reset -t --height 83000
2022-11-20T15:51:31.431+0300 INFO MaxValidUntilBlockIncrement is not set or wrong, using default value {"MaxValidUntilBlockIncrement": 5760}
2022-11-20T15:51:31.434+0300 INFO restoring blockchain {"version": "0.2.6"}
failed to create Blockchain instance: could not initialize blockchain: current block height is 0, can't reset state to height 83000
```
We don't use all of the Stack functionality for it, so drop useless methods
and avoid some interface conversions. It increases single-node TPS by about
0.9%, so nothing really important there, but not a bad change either. Maybe it
can be reworked again with generics though.
Do not block subscribers until the unsubscription request to RPC server
is completed. Otherwise, another notification may be received from the
RPC server which will block the unsubscription process.
At the same time, fix event-based waiter. We must not block the receiver
channel during unsubscription because there's a chance that subsequent
event will be sent by the server. We need to read this event in order not
to block the WSClient's readloop.
Bad contract -> no contract. Unfortunately we've got a broken
6f1837723768f27a6f6a14452977e3e0e264f2cc contract on the mainnet which can't
be decoded (even though it had been saved successfully), so this is a
temporary fix for #2801 to be able to start mainnet node after shutdown.
There is a security issue found in github.com/btcsuite/btcd that we don't care
about (we're only using 256k1 implementation), but GitHub complains about
it. We could update to github.com/btcsuite/btcd/btcec/v2, but it's now just a
thin wrapper over github.com/decred/dcrd/dcrec/secp256k1/v4, so we better use
it directly.
When block is being spread through the network we can get a lot of invs with
the same hash. Some more stale nodes may also announce previous or some
earlier block. We can avoid full DB lookup for them and minimize inv handling
time (timeouts in inv handler had happened in #2744).
It doesn't affect tests, just makes node a little less likely to spend some
considerable amount of time in the inv handler.
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.
And determine the need for Null dynamically. For some reason the only dynamic
context is Contract.Call. CALLT is not dynamic and neither is a call from
native contract, go figure...
Which allows to enable/disable the service, change nodes, keys and other
settings. Unfortunately, atomic.Value doesn't allow Store(nil), so we have to
store a pointer there that can point to nil interface.
It can change the committee even if noone voted. Fixes state diff at block
390726 of T5 testnet where there are no transactions, but committee changes
because there were some registrations in previous 21 blocks.
It's not an ideal solution, but at least it solves the problem for
now. Caveats:
* consensus only needs one method, so it's mirrored to Blockchain
* rpcsrv uses core.* definition of the StateRoot (so technically it might as
well not have an internal Ledger), but it uses core already unfortunately
1. It's not good for pkg/core to import anything from pkg/neorpc.
2. The type is closely tied to the state package, even though it's not stored
in the DB
It makes a copy of the resulting set, so the lock can be released
earlier. This helps a lot with iterators that keep Seek() unfinished for a
long time,
