Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Further speed up transactions view load #6579

Conversation

stejbac
Copy link
Contributor

@stejbac stejbac commented Feb 12, 2023

This PR is a collection of optimisations to the Funds/Transactions display, via the TransactionsView.updateList method. The much earlier PR #5120 also attempted to optimise the transactions view load, but did not completely fix the main quadratic time issue (cost ~ transations count * tradable count, due to this many calls to TransactionAwareTradable.isRelatedToTransaction) and since then it has slowed down again a bit.

The main optimisation of this PR is to partition the wallet transactions pseudo-randomly into 64 buckets and use a bit array structure (which could be considered a degenerate Bloom filter with k, the number of hash functions, equal to 1), to filter the list of all tradable-transaction pairs prior to calling isRelatedToTransaction on each of them. While technically still quadratic time, it cuts down the number of calls (though not the runtime) roughly 40 times (at least when there are similar numbers of BSQ swap trades and escrow trades).

The other main speedup is fixing an issue where the set of all tradables was needlessly rebuild for every transaction listed. In addition to that, some further modest optimisations to isRelatedToTransaction and the TransactionsListItem constructor were made by introducing some appropriate simple caching, cleaning up the code slightly and removing an unused listener.

The following is a call tree of the transactions view load from JProfiler, prior to the optimisations (in which the load was taking 5-15 seconds on my old Bisq instance with ~1300 trades and ~3000 transactions):
Screenshot-2023-2-12 Call Tree

The following is the same call tree after the optimisations, repeatedly tabbing to the view 10 times (instead of just once):
Screenshot-2023-2-12 Call Tree(1)

Replace the "Optional.ofNullable(...)..." constructs with more direct
code using short-circuit operators, as this is shorter and a little
faster. Also use "trade.get[Deposit|Payout]TxId()" instead of the code
"trade.get[Deposit|Payout]TxId().getTxId()", as (upon inspection of the
code) there should never be a case where the deposit/payout transaction
field of a Trade object is set but the respective txID field is null (or
set to an inconsistent value).

Also remove a redundant 'RefundManager.getDisputesAsObservableList'
method call, which was also slowing things down slightly.

The minor speedups afforded by the above are important because the
method 'TransactionAwareTrade.isRelatedToTransaction' is called a
quadratic number of times and consequently a major bottleneck when
loading the Transactions view.
Use a mutable static tuple field to cache the last result of
'Sha256Hash.toString', which is used to get the ID string of the input
tx, when calling 'TransactionAwareTrade.isRelatedToTransaction'. In this
way, consecutive calls to 'isRelatedToTransaction' on the same input tx
(over all the past trades, as done by 'TransactionsView.updateList') are
sped up significantly, since hex encoding the txId is a bottleneck.
1. Expression lambda -> method reference;
2. Statement lambda -> expression lambda;
3. Field -> local variable.
Move the line,

   Set<Tradable> tradables = tradableRepository.getAll();

to the top level of 'TransactionsView.updateList', instead of needlessly
calling 'TradableRepository.getAll' (which builds a new set every
invocation) for each wallet transaction being iterated over.

This was causing a significant slowdown of the view load.
Inline a local variable, to eliminate another minor Sha256Hash.toString
hotspot in the Transactions view load, this time coming from
'TransactionsAwareOpenOffer.isRelatedToTransaction'. This is helpful in
the case that the user has a large number of (possibly disabled) BSQ
swap offers.
Use a crude Bloom filter (of sorts) to cut down the quadratic number of
calls to 'TransactionAwareTradable.isRelatedToTransaction' (that is, one
for each tx-tradable pair) during the Transactions view load. In this
way, we may reduce the number of calls roughly 40-fold, for a Bisq
instance with similar numbers of BSQ swap trades and escrow trades.

(Sadly, profiling does not show a 40-fold reduction in the size of the
'isRelatedToTransaction' hotspot, likely due to the remaining calls
being expensive ones involving disputed trades or unusual txs with
nonzero locktime, e.g. dust attacks or funds from Electrum wallets.)

To this end, partition the wallet transactions into 64 pseudo-randomly
chosen buckets (with a dedicated bucket for txs which might be delayed
payouts, namely those with nonzero locktime). Add an interface method,
'TransactionAwareTradable.getRelatedTransactionFilter', which returns an
IntStream of all the indices of buckets where a related tx may plausibly
be found. Where this is unclear, e.g. for trades involved in a dispute,
just return everything (that is, the range 0..63 inclusive).

Add a class, 'RelatedTransactionFilterSlices', that holds a provided
list of TransactionAwareTradable instances and 64 bitsets of all the
slices through their respective filters (each realised as 64-bit word
instead of a streams of integers). In this way, a list of tradables
plausibly related to any given tx may be quickly found by simply
selecting the appropriate bitset of the 64 (by the tx bucket index).
(This also avoids a redundant call to 'Tradable.getShortId', shaving off
another percent or two from the Transactions view load.)
Eliminate a minor quadratic time bug, caused by the unnecessary addition
of a (BtcWalletService) TxConfidenceListener for each list item in the
Transactions view. (Since the confidence listeners are internally held
in a CopyOnWriteArraySet, this sadly runs in quadratic time, slowing
down the Transactions view load a little.)

The confidence listener is apparently redundant because of a set of
calls to 'TransactionsListItem.cleanup' immediately upon construction of
the item list, which removes all the listeners just added. (This code
appears to date from at least February 2016, in commit c70df86.)

(The confidence indicators are kept up to date by simply reloading the
entire list upon each wallet change event.)
Make 'FormattingUtils.formatDateTime(Date, boolean)' two or three times
faster by caching the 'java.text.DateFormat' objects used to format the
given date & time in either the local or UTC time zone. Since these
formatters are not thread safe (though they may be reused), use a
ThreadLocal to store them (as a tuple, along with the current locale to
make sure it hasn't changed since the last invocation).

This is a minor hotspot for the Transactions view load, since the date
strings in TransactionsListItem are formatted eagerly (via DisplayUtils)
for the purpose of filtering the items by substring match. (Other list
views seem to format dates lazily, though it's possible that there will
be speedups elsewhere in the UI.)
In the (hopefully rare) case that the user has multiple past trades that
end in arbitration, the entire wallet tx output set was scanned once for
every such trade (via 'TransactionAwareTrade.isRefundPayoutTx' calls),
to look for any outputs matching the payout address. This potentially
causes a slowdown of the Transaction view load for each new arbitration
case added. To avoid this problem, cache the last set of recipient
address strings of the provided tx, as the next call to
'isRefundPayoutTx' is likely to be for the same tx.

Also check that there is exactly one input (the multisig input) for any
candidate delayed payout tx, to speed up 'isDelayedPayoutTx' in case the
wallet contains many unusual txs with nonzero locktime.
Copy link
Contributor

@alejandrogarcia83 alejandrogarcia83 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

utACK

Excellent job!

@alejandrogarcia83 alejandrogarcia83 merged commit 581fbd1 into bisq-network:master Feb 12, 2023
@alejandrogarcia83 alejandrogarcia83 added this to the v1.9.10 milestone Feb 12, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants