Although BlockMinter could reattach a repository session to its cache of potential blocks,
and these blocks would in turn reattach that repository session to their transactions,
further transaction-specific fields (e.g. creator PublicKeyAccount) were not being updated.
This would lead to NPEs like the following:
Exception in thread "BlockMinter" java.lang.NullPointerException
at org.qortal.repository.hsqldb.HSQLDBRepository.cachePreparedStatement(HSQLDBRepository.java:587)
at org.qortal.repository.hsqldb.HSQLDBRepository.prepareStatement(HSQLDBRepository.java:569)
at org.qortal.repository.hsqldb.HSQLDBRepository.checkedExecute(HSQLDBRepository.java:609)
at org.qortal.repository.hsqldb.HSQLDBAccountRepository.getBalance(HSQLDBAccountRepository.java:327)
at org.qortal.account.Account.getConfirmedBalance(Account.java:72)
at org.qortal.transaction.MessageTransaction.isValid(MessageTransaction.java:200)
at org.qortal.block.Block.areTransactionsValid(Block.java:1190)
at org.qortal.block.Block.isValid(Block.java:1137)
at org.qortal.controller.BlockMinter.run(BlockMinter.java:301)
where the Account has an associated repository session which is now obsolete.
This commit reverts BlockMinter back to obtaining a repository session before entering main loop.
To prevent a single or very small number of minters receiving the rewards for an entire tier, share bins can now require "activation". This adds the requirement that a minimum number of accounts must be present in a share bin before it is considered active. When inactive, the rewards and minters are added to the previous tier.
Summary of new functionality:
- If a share bin has more than one, but less than 30 accounts present, the rewards and accounts are shifted to the previous share bin.
- This process is iterative, so the accounts can shift through multiple tiers until the minimum number of accounts is met, OR the share bin's starting level is less than shareBinActivationMinLevel.
- Applies to level 7+, so that no backwards support is needed. It will only take effect once the first account reaches level 7.
This requires hot swapping the sharesByLevel data to combine tiers where needed, so is a considerable shift away from the immutable array that was in place previously.
All existing and new unit tests are now passing, however a lot more testing will be needed.
Online account credit is a more useful definition of "minting" than block signing, from the user's perspective. Should bring UI minting/syncing status in line with the core's systray status.
- Reduce concurrent reward share limit from 6 to 3 (or from 5 to 2 when including self share) - as per community vote.
- Founders remain at 6 (5 when including self share) - also decided in community vote.
- When all slots are being filled, require that at least one is a self share, so that not all can be used for sponsorship.
- Activates at future undecided timestamp.
We already mark peers as misbehaved if they returned invalid signatures, but this wasn't sufficient when multiple copies of the same invalid block exist on the network (e.g. after a hard fork). In these cases, we need to be more proactive to avoid syncing with these peers, to increase the chances of preserving other candidate blocks.
Previously, a peer would be continuously considered not 'old' if it had a connection attempt in the past day. This prevented some peers from being removed, causing nodes to hold a large repository of peers. On slower systems, this large number of known peers resulted in low numbers of outbound connections being made, presumably because of the time taken to iterate through dataset, using up a lot of allKnownPeers lock time.
On devices that experienced the problem, it could be solved by deleting all known peers. This adds confidence that the old peers were the problem.
Previously, a peer would be continuously considered not 'old' if it had a connection attempt in the past day. This prevented some peers from being removed, causing nodes to hold a large repository of peers. On slower systems, this large number of known peers resulted in low numbers of outbound connections being made, presumably because of the time taken to iterate through dataset, using up a lot of allKnownPeers lock time.
On devices that experienced the problem, it could be solved by deleting all known peers. This adds confidence that the old peers were the problem.
CalDescent
catbref
QuickMythril
qortish