Price Oracle data-series logic uses `unordered_map` to update the Oracle object.
This results in different servers disagreeing on the order of that hash table.
Consequently, the generated ledgers will have different hashes.
The fix uses `map` instead to guarantee the order of the token pairs
in the data-series.
Due to the rounding, LPTokenBalance of the last
Liquidity Provider (LP), might not match this LP's
trustline balance. This fix sets LPTokenBalance on
last LP withdrawal to this LP's LPToken trustline
balance.
Single path AMM offer has to factor in the transfer in rate
when calculating the upper bound quality and the quality function
because single path AMM's offer quality is not constant.
This fix factors in the transfer fee in
BookStep::adjustQualityWithFees().
* Fix AMM offer rounding and low quality LOB offer blocking AMM:
A single-path AMM offer with account offer on DEX, is always generated
starting with the takerPays first, which is rounded up, and then
the takerGets, which is rounded down. This rounding ensures that the pool's
product invariant is maintained. However, when one of the offer's side
is XRP, this rounding can result in the AMM offer having a lower
quality, potentially causing offer generation to fail if the quality
is lower than the account's offer quality.
To address this issue, the proposed fix adjusts the offer generation process
to start with the XRP side first and always rounds it down. This results
in a smaller offer size, improving the offer's quality. Regardless if the offer
has XRP or not, the rounding is done so that the offer size is minimized.
This change still ensures the product invariant, as the other generated
side is the exact result of the swap-in or swap-out equations.
If a liquidity can be provided by both AMM and LOB offer on offer crossing
then AMM offer is generated so that it matches LOB offer quality. If LOB
offer quality is less than limit quality then generated AMM offer quality
is also less than limit quality and the offer doesn't cross. To address
this issue, if LOB quality is better than limit quality then use LOB
quality to generate AMM offer. Otherwise, don't use the quality to generate
AMM offer. In this case, limitOut() function in StrandFlow limits
the out amount to match strand's quality to limit quality and consume
maximum AMM liquidity.
Rounding in the payment engine is causing an assert to sometimes fire
with "dust" amounts. This is causing issues when running debug builds of
rippled. This issue will be addressed, but the assert is no longer
serving its purpose.
The AMM has an invariant for swaps where:
new_balance_1*new_balance_2 >= old_balance_1*old_balance_2
Due to rounding, this invariant could sometimes be violated (although by
very small amounts).
This patch introduces an amendment `fixAMMRounding` that changes the
rounding to always favor the AMM. Doing this should maintain the
invariant.
Co-authored-by: Bronek Kozicki
Co-authored-by: thejohnfreeman
It can be difficult to make transaction breaking changes to low level
code because the low level code does not have access to a ledger and the
current activated amendments in that ledger (the "rules"). This patch
adds global access to the current ledger rules as a `std::optional`. If
the optional is not seated, then there is no active transaction.
The `rotateWithLock` function holds a lock while it calls a callback
function that's passed in by the caller. This is a problematic design
that needs to be used very carefully. In this case, at least one caller
passed in a callback that eventually relocks the mutex on the same
thread, causing UB (a deadlock was observed). The caller was from
SHAMapStoreImpl, and it called `clearCaches`. This `clearCaches` can
potentially call `fetchNodeObject`, which tried to relock the mutex.
This patch resolves the issue by changing the mutex type to a
`recursive_mutex`. Ideally, the code should be rewritten so it doesn't
hold the mutex during the callback and the mutex should be changed back
to a regular mutex.
Co-authored-by: Ed Hennis <ed@ripple.com>
This amendment, `fixPreviousTxnID`, adds `PreviousTxnID` and
`PreviousTxnLgrSequence` as fields to all ledger objects that did
not already have them included (`DirectoryNode`, `Amendments`,
`FeeSettings`, `NegativeUNL`, and `AMM`). This makes it much easier
to go through the history of these ledger objects.
When calculating reward shares, the amount should always be rounded
down. If the `fixUniversalNumber` amendment is not active, this works
correctly. If it is not active, then the amount is incorrectly rounded
up. This patch introduces an amendment so it will be rounded down.
This fixes a case where a peer can desync under a certain timing
circumstance--if it reaches a certain point in consensus before it receives
proposals.
This was noticed under high transaction volumes. Namely, when we arrive at the
point of deciding whether consensus is reached after minimum establish phase
duration but before having received any proposals. This could be caused by
finishing the previous round slightly faster and/or having some delay in
receiving proposals. Existing behavior arrives at consensus immediately after
the minimum establish duration with no proposals. This causes us to desync
because we then close a non-validated ledger. The change in this PR causes us to
wait for a configured threshold before making the decision to arrive at
consensus with no proposals. This allows validators to catch up and for brief
delays in receiving proposals to be absorbed. There should be no drawback since,
with no proposals coming in, we needn't be in a huge rush to jump ahead.
We do not currently enforce that incoming peer connection does not have
remote_endpoint which is already used (either by incoming or outgoing
connection), hence already stored in slots_. If we happen to receive a
connection from such a duplicate remote_endpoint, it will eventually result in a
crash (when disconnecting) or weird behavior (when updating slot state), as a
result of an apparently matching remote_endpoint in slots_ being used by a
different connection.
This amendment fixes an edge case where an empty DID object can be
created. It adds an additional check to ensure that DIDs are
non-empty when created, and returns a `tecEMPTY_DID` error if the DID
would be empty.
* telENV_RPC_FAILED is a new code, reserved exclusively
for unit tests when RPC fails. This will
make those types of errors distinct and easier to test
for when expected and/or diagnose when not.
* Output RPC command result when result is not expected.
We are currently using old version 0.6.2 of `xxhash`, as a verbatim copy and paste of its header file `xxhash.h`. Switch to the more recent version 0.8.2. Since this version is in Conan Center (and properly protects its ABI by keeping the state object incomplete), add it as a Conan requirement. Switch to the SIMD instructions (in the new `XXH3` family) supported by the new version.
This algorithm is about an order of magnitude faster than the existing
algorithm (about 10x faster for encoding and about 15x faster for
decoding - including the double hash for the checksum). The algorithms
use gcc's int128 (fast MS version will have to wait, in the meantime MS
falls back to the slow code).
