Fee Estimation via Fee rate Forecasters

ismaelsadeeq commented at 11:44 am on May 23, 2024: member

This PR aims to improve Bitcoin Core Fee Estimation by:

Reducing Overestimation: Address and mitigate the overestimation issues present in the current BlockPolicyEstimator. This issue has been documented and acknowledged by various sources:
- See Delving bitcoin posts
- Rene-pickhard Research
Mempool Awareness: Enable the fee estimator to be aware of the mempool state, allowing it to react faster and more accurately to rapidly changing fee market conditions, when targeting very short timeframes.
Empowering Node Users: Allow node users to be self-sovereign by using their node’s estimates, reducing reliance on third-party fee estimations.
Simplifying Strategy Integration: Simplify the process of adding new fee estimation strategies in the future.

The detailed design document for this PR can be found here. Please note that the design is subject to change as we refine the approach.

We are currently seeking conceptual acknowledgment for this PR.

This is a collaborative effort with @willcl-ark and incorporates insights from other contributors.

DrahtBot commented at 11:44 am on May 23, 2024: contributor

The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.

Code Coverage & Benchmarks

For details see: https://corecheck.dev/bitcoin/bitcoin/pulls/30157.

Reviews

See the guideline for information on the review process.

Type	Reviewers
Concept ACK	murchandamus, vasild

If your review is incorrectly listed, please react with 👎 to this comment and the bot will ignore it on the next update.

Conflicts

Reviewers, this pull request conflicts with the following ones:

#31664 (Fees: add Fee rate Forecaster Manager by ismaelsadeeq)
#31382 (kernel: Flush in ChainstateManager destructor by TheCharlatan)
#31282 (refactor: Make node_id a const& in RemoveBlockRequest by maflcko)
#31260 (scripted-diff: Type-safe settings retrieval by ryanofsky)
#30079 (Fee Estimation: Ignore all transactions that are CPFP’d by ismaelsadeeq)

If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.

DrahtBot added the label CI failed on May 23, 2024

DrahtBot commented at 3:18 pm on May 23, 2024: contributor

🚧 At least one of the CI tasks failed. Make sure to run all tests locally, according to the documentation.

Possibly this is due to a silent merge conflict (the changes in this pull request being incompatible with the current code in the target branch). If so, make sure to rebase on the latest commit of the target branch.

Leave a comment here, if you need help tracking down a confusing failure.

Debug: https://github.com/bitcoin/bitcoin/runs/25327260112

luke-jr commented at 5:20 pm on May 23, 2024: member

Make the fee estimator aware of the state of the mempool, allowing it to respond to changing conditions immediately.

The state of the node’s mempool may not accurately reflect the state of others’ mempools, and not even its own mempool when the block is found in the future. It isn’t a good single source of information. Perhaps it is a good idea to use it as a secondary source, but probably it should only ever adjust fee estimations upward, not down.

willcl-ark commented at 8:41 am on May 24, 2024: member

The state of the node’s mempool may not accurately reflect the state of others’ mempools, and not even its own mempool when the block is found in the future. It isn’t a good single source of information.

Correct. The rationale behind this set of changes can be summed up briefly as follows:

Add a new standalone modular fee estimation manager to which (many) Forcasters can be trivially added or removed (vs modifying BlockPolicyEstimator).
Provide two forcaster implementations which do not exist today ( 1. mempool-based and 2. previous-6-blocks-seen-in-mempool) to demonstrate functionality.

In the present, there is a strong tendency for users of Bitcoin Core to consult external fee estimation services when they want timely (next 1 or 2 blocks) confirmation of transactions, whilst also not overpaying. Examples of these include mempool.space, whatthefee.io, Samourai’s nextblock.is (now down), johoe, blockchair etc., with more popping up every month.

We also analysed/estimated Bitcoin Core users not using in-built estimation in a post here.

In our opinion having users feel the need to use external fee estimation services that they could equally have served to them by their own node, feels sub-optimal.

In addition to this, when users do use the current Bitcoin Core fee estimator, there are often times when they end up overpaying, see delving bitcoin post Mempool based fee estimation and various issues over the years e.g. #30009 . This is avoidable.

Work from a student of @renepickhardt link demonstrated something we also measured independently – that bitcoin core’s current estimates are often overpaying significantly following fee spikes. This effect can be directly mitigated by using a mempool-based estimation.

Perhaps it is a good idea to use it as a secondary source, but probably it should only ever adjust fee estimations upward, not down.

In this changeset we take the approach of using the lowest result from all “confident” Forcasters. The rationale is that we expect users wanting fast confirmation to have RBF enabled, allowing them to bump fees if we still undershoot.

Comments from @harding link explained that it may be possible for miners to artificially increase a strictly-mempool-based fee-rate. By taking the lower (confident) result from n Forcasters, we attempt to protect against this attack (and others like it), at the potential cost of having to RBF.

We do plan to add additional sanity checks to the mempool-based Forcaster as described in #27995, but these are not yet implemented. In any case, even without these additional checks we have been seeing much-improved short time-scale estimations from Bitcoin Core.

ismaelsadeeq force-pushed on May 29, 2024

DrahtBot removed the label CI failed on May 29, 2024

DrahtBot added the label Needs rebase on Jun 11, 2024

ismaelsadeeq force-pushed on Jul 3, 2024

DrahtBot removed the label Needs rebase on Jul 3, 2024

ismaelsadeeq force-pushed on Jul 3, 2024

DrahtBot added the label CI failed on Jul 3, 2024

DrahtBot commented at 5:02 pm on July 3, 2024: contributor

🚧 At least one of the CI tasks failed. Make sure to run all tests locally, according to the documentation.

Possibly this is due to a silent merge conflict (the changes in this pull request being incompatible with the current code in the target branch). If so, make sure to rebase on the latest commit of the target branch.

Leave a comment here, if you need help tracking down a confusing failure.

Debug: https://github.com/bitcoin/bitcoin/runs/27003108803

DrahtBot removed the label CI failed on Jul 3, 2024

ismaelsadeeq force-pushed on Jul 8, 2024

DrahtBot added the label Needs rebase on Jul 8, 2024

ismaelsadeeq force-pushed on Jul 8, 2024

DrahtBot removed the label Needs rebase on Jul 8, 2024

DrahtBot added the label Needs rebase on Jul 18, 2024

ismaelsadeeq force-pushed on Jul 24, 2024

DrahtBot commented at 12:14 pm on July 24, 2024: contributor

🚧 At least one of the CI tasks failed. Debug: https://github.com/bitcoin/bitcoin/runs/27856068740

Make sure to run all tests locally, according to the documentation.

The failure may happen due to a number of reasons, for example:

Possibly due to a silent merge conflict (the changes in this pull request being incompatible with the current code in the target branch). If so, make sure to rebase on the latest commit of the target branch.
A sanitizer issue, which can only be found by compiling with the sanitizer and running the affected test.
An intermittent issue.

Leave a comment here, if you need help tracking down a confusing failure.

DrahtBot added the label CI failed on Jul 24, 2024

willcl-ark added the label Needs Conceptual Review on Jul 24, 2024

DrahtBot removed the label Needs rebase on Jul 24, 2024

ismaelsadeeq force-pushed on Jul 24, 2024

DrahtBot removed the label CI failed on Jul 24, 2024

ismaelsadeeq force-pushed on Aug 13, 2024

DrahtBot added the label CI failed on Aug 13, 2024

ismaelsadeeq force-pushed on Aug 13, 2024

DrahtBot removed the label CI failed on Aug 13, 2024

hebasto added the label Needs CMake port on Aug 16, 2024

DrahtBot added the label Needs rebase on Aug 21, 2024

glozow added the label TX fees and policy on Aug 21, 2024

maflcko removed the label Needs CMake port on Aug 29, 2024

ismaelsadeeq force-pushed on Sep 3, 2024

ismaelsadeeq commented at 9:43 am on September 3, 2024: member

Rebased and added newly introduced files to cmakelist file instead makefile

DrahtBot removed the label Needs rebase on Sep 3, 2024

DrahtBot added the label Needs rebase on Sep 20, 2024

ismaelsadeeq force-pushed on Oct 14, 2024

DrahtBot removed the label Needs rebase on Oct 14, 2024

DrahtBot added the label CI failed on Oct 14, 2024

DrahtBot commented at 5:03 pm on October 14, 2024: contributor

🚧 At least one of the CI tasks failed. Debug: https://github.com/bitcoin/bitcoin/runs/31509135850

Try to run the tests locally, according to the documentation. However, a CI failure may still happen due to a number of reasons, for example:

Possibly due to a silent merge conflict (the changes in this pull request being incompatible with the current code in the target branch). If so, make sure to rebase on the latest commit of the target branch.
A sanitizer issue, which can only be found by compiling with the sanitizer and running the affected test.
An intermittent issue.

Leave a comment here, if you need help tracking down a confusing failure.

murchandamus commented at 5:57 pm on October 21, 2024: contributor

Concept ACK! Thank you for working on this.

vasild commented at 4:28 am on November 6, 2024: contributor

Concept ACK

Allow node users to be self-sovereign by using their node’s estimates, reducing reliance on third-party fee estimations.

having users feel the need to use external fee estimation services that they could equally have served to them by their own node, feels sub-optimal

Indeed, I do that :face_with_head_bandage:

ismaelsadeeq force-pushed on Nov 11, 2024

DrahtBot removed the label CI failed on Nov 11, 2024

ismaelsadeeq force-pushed on Nov 14, 2024

DrahtBot added the label CI failed on Nov 14, 2024

DrahtBot commented at 8:52 pm on November 14, 2024: contributor

🚧 At least one of the CI tasks failed. Debug: https://github.com/bitcoin/bitcoin/runs/33010015546

Try to run the tests locally, according to the documentation. However, a CI failure may still happen due to a number of reasons, for example:

Possibly due to a silent merge conflict (the changes in this pull request being incompatible with the current code in the target branch). If so, make sure to rebase on the latest commit of the target branch.
A sanitizer issue, which can only be found by compiling with the sanitizer and running the affected test.
An intermittent issue.

Leave a comment here, if you need help tracking down a confusing failure.

ismaelsadeeq force-pushed on Nov 14, 2024

ismaelsadeeq commented at 9:52 pm on November 14, 2024: member

Thanks for your conceptual review, @murchandamus and @vasild. There’s also another review by @remyers in the tracking issue: #30392 (comment).

I have updated this PR based on feedback from #30391 and in-person conversations:

I am now using the newly introduced FeeFrac datatype, which does not lose precision like CFeeRate.
Updated the CalculatePercentile function to return a monotonically decreasing estimate for high and low priority. The previous approach took the fee rate of the package at the exact percentile, which could be high due to the ancestors of the percentile package included previously in a sibling package.

Relevant commit: 2ac86a8fd7fe108078511ef2f6ad77e315591a2b. I’ve also added a test for this.

Added a functional test for the estimatefee RPC behavior (commit ac8f2caf050e73b038a8e03a51dfe69e1ea7aa19).
Removed the tracing commit after merging #26593, as it changes the tracing structure, which I haven’t reviewed yet. The tracing commit was relevant for collecting node’s estimation data for analysis, it can be added later on.

This PR has now received a conceptual review from a few contributors, so we can proceed to code and approach review.

DrahtBot removed the label CI failed on Nov 14, 2024

DrahtBot added the label Needs rebase on Nov 20, 2024

ismaelsadeeq force-pushed on Nov 25, 2024

DrahtBot removed the label Needs rebase on Nov 25, 2024

DrahtBot added the label Needs rebase on Dec 17, 2024

ismaelsadeeq force-pushed on Jan 2, 2025

DrahtBot removed the label Needs rebase on Jan 2, 2025

DrahtBot added the label Needs rebase on Jan 6, 2025

glozow referenced this in commit 66aa6a47bd on Jan 8, 2025

ismaelsadeeq force-pushed on Jan 15, 2025

DrahtBot added the label CI failed on Jan 15, 2025

ismaelsadeeq force-pushed on Jan 15, 2025

DrahtBot removed the label CI failed on Jan 15, 2025

DrahtBot removed the label Needs rebase on Jan 16, 2025

DrahtBot added the label Needs rebase on Jan 22, 2025

fees: add `ForecastResult` class

- This class represents the response returned by
  a fee rate forecaster.

66d4a29ab0

fees: add `ConfirmationTarget` struct

- Defines the types of confirmation targets for fee rate forecasters.

- This enable having a generic input for fee rate forecasters.

a29c038e1a

fees: add Forecaster abstract class

- This commit implements `Forecaster` abstract class
  as the base class of fee rate forecasters.

- Derived classes must provide concrete implementation
  of the virtual methods.

Co-authored-by: willcl-ark <will@256k1.dev>

0de1842f1b

fees: add `ForecastType` enum

- ForecastType will be used to identify forecasters.

- Each time a new forecaster is added, a corresponding
  enum value should be added to ForecastType.

- This allows users to identify which forecasting strategy
  was used to make a fee rate estimate.

35a0576ee2

fees: add `ForecasterMan` class

- Its a module for managing and utilising multiple
  fee rate forecasters to provide fee rate forecast.

- The ForecasterManager class allows for the registration of
  multiple fee rate forecasters.

Co-authored-by: willcl-ark <will@256k1.dev>

51347355a7

fees: refactor: rename policy_fee_tests.cpp to feerounder_tests.cpp

- Also remame the test suite name to match the new name.

435f2424d1

fees: refactor: rename fees to block_policy_estimator

- Also move it to policy/fees and update the includes
- Note: the block_policy_estimator.h include in block_policy_estimator.cpp was done manually.

86d46e71b5

fees: rename fees_args to block_policy_estimator_args

- Also move them to policy/fees/ and update includes
- Note: the block_policy_estimator_args.h include in block_policy_estimator_args.cpp was done manually.

0e6690e129

fees: return current block height in `estimateSmartFee` c67949ddad

fees: make block_policy_estimator a forecaster 956926cee3

fees: add block policy estimator to forecaster manager

- This changes `CBlockPolicyEstimator` to a shared pointer
  this gives us three advantages.
   - Registering to validation interface using shared pointer
   - Scheduling block policy estimator flushes using shared pointer
   - Registering block policy estimator to forecaster_man

ce7199070b

fees: add `forecastTypeToString` method

- This method converts a ForecastType enum to its
  string representation.

17e844276f

fees: add `CalculatePercentiles` function

- The CalculatePercentiles function, given
  a vector of feerates in the order they were added
  to the block, will return the 25th, 50th, 75th,
  and 95th percentile feerates.

- Also add a unit test for this function.

4892df53de

fees: add `MemPoolForecaster` class

- The mempool based fee rate forecaster generate a predicted fee rate estimate
  for a given confirmation target using the mempool unconfirmed transactions.

Co-authored-by: willcl-ark <will@256k1.dev>

8befea5451

test: add mempool forecaster unit test 115928a33c

fees: cache `MemPoolPolicyEstimator` forecasts

- Provide new estimates only when the time delta from previous
  forecast is older than 30 seconds.

- This caching helps avoid the high cost of frequently generating block templates,
  preventing users from inadvertently calling `estimateFee` repeatedly.

Co-authored-by: willcl-ark <will@256k1.dev>

aecb08b81c

fees: add `GetFeeEstimateFromForecasters` method

- Fallback to Block policy estimator estimates whenever mempool forecaster
  estimates are higher than block policy estimator.

3bf2c46428

===== End of FeeRateForecasterMan commits ===== c667bb1c6d

rpc: create an rpc `estimatefee`

- Given a confirmation target, we use fee estimator module that call all
available fee estimator forcasters and return the lowest fee rate that if
a transaction use will likely confirm in a given confirmation target.

Co-authored-by: willcl-ark <will@256k1.dev>

d1b786e5e2

test: test `estimatefee` rpc a405eefac9

test: add `estimatefee` functionality test 8982cfea52

===== End of RPC commits ===== 6d0c2c7a7a

ismaelsadeeq force-pushed on Jan 22, 2025

DrahtBot removed the label Needs rebase on Jan 22, 2025

Fee Estimation via Fee rate Forecasters #30157

Code Coverage & Benchmarks

Reviews

Conflicts