fetch strategy:
Before:
- block fetching commands for different accounts were in the same wait
group, making them dependent on each iteration.
- transfers loading command was checking database for new unloaded
blocks on timeout and was in the same wait group with block fetching, so
it was often blocked until all block fetching commands finish for
iteration.
Now:
- block fetching commands run independently for each account
- transfers fetching command is run once on startup for unloaded blocks
from DB
- fetch history blocks commands are launched once on startup for
accounts with no full history loaded
- transfers are loaded on each iteration of block range check
without waiting for all ranges to be checked
- reverted a change that stopped looking for ERC20 transfers if no nonce
and balance change found within a block range for ETH
- implemented sending EventRecentHistoryReady event at a proper time
- moved EventFetchingRecentHistory event to Strategy type as it does not make
sense to send this event in loop
- moved iterating through blocks logic to inside of `loadTransfers` command, which
now accepts a block range.
- reuse `uniqueHeaders` function in commands.go
- clean up
Updates #10246
were stored in cache by pointers, which caused falsy cache hits in loop
because pointers with same address were created for different block
numbers. Now cache uses block numbers of uint64 as key, which can
overflow but it is not a problem since we use this cache for values
comparison, not as user data.
Fix crash on nil pointer in log.
Remove some unused code.
Protect nonceRanges with mutex while reading.
Updates #10246
Ivan Belyakov
Dario Gabriel Lipicar