Skip to content

Latest commit

 

History

History
304 lines (235 loc) · 17.5 KB

full-rbf.md

File metadata and controls

304 lines (235 loc) · 17.5 KB
Raw

Full RBF

Disclaimer: this is my personal understanding from reading everything I can and speaking to as many people as possible. I'll indicate when I'm giving my own opinion. Otherwise, I've tried to be open-minded and fair but there may be unintentional biases.

Background

RBF and Full RBF

Replace-by-Fee is a mempool policy that allows nodes to decide between conflicting unconfirmed transactions based on feerate. Prior to this policy, the mempool accepted whichever transaction it saw first.

The definition of Full RBF has changed over time. In this document, it means a policy of accepting transaction replacements even if the original transactions did not signal BIP125 replaceability.

Recently

June 2021: Antoine Riard posted about a plan to add a -mempoolfullrbf option in v24.0 (scheduled for September/October 2022), soliciting opinions and concerns.

June 2022: Antoine Riard opened #25353 to add -mempoolfullrbf option

July 2022: Bitcoin Core merged #25353

October 2022: Dario Sneidermanis posted concerns about full RBF and the -mempoolfullrbf option in Bitcoin Core creating problems for businesses accepting unconfirmed transactions as final. He requested removal of the -mempoolfullrbf option from v24.0 to give Muun and other applications more time to prepare.

This post led to various discusssions and mailing list posts, mainly about full RBF in general:

  • Many responded emphasizing that the default Bitcoin Core RBF policy had not changed.
  • Anthony Towns posted in agreement with the idea that network-wide deployment of full RBF, not just the individual node option, is on the table. He suggested, as a way to give protocols time to test and businesses time to adapt, temporarily restricting the option to testnet-only and creating a timeline for deployment of default true full RBF on mainnet.
  • Sergej Kotliar posted about issues with full RBF, disagreeing with the idea that double spend risk is already high and that LN can replace on-chain "zeroconf" payments. He provided some data from Bitrefill's experiences. He says that, in practice, "fewer than 1 in a million" double spend attempts succeed. He notes only 15% of their payments use LN and a past experiment defaulting to bech32 simply drove users away. He expects that, if Bitrefill stopped accepting unconfirmed payments, "the majority of the current 85% of bitcoin users that pay onchain would just not use bitcoin anymore."
  • John Carvalho posted about general issues with RBF, believing it to make zeroconf payment risks less manageable. In the same post mentioned above, Sergej Kotliar also described the free option problem as one of the issues with replacements in general. See American call option section below.

Several full-RBF-related pull requests have been opened to Bitcoin Core.

  • #26287 implemented the request to remove the option on mainnet. It was closed after opposition from several reviewers.
  • #26323 is a PR to enable full RBF by default, but at a predetermined time (May 1, 2023). If users set the -mempoolfullrbf=1 option, it does not go into effect until then. Users can also set the -mempoolfullrbf=0 option to continue requiring signaling past that day.
  • #26305 is a PR to enable full RBF by default, implementing the second part of what Antoine had posted to the mailing list. It is not intended for v24.0. Most reviewers are indicating concept ACK for v25.0 or some other later time.

Summaries

This is my understanding of the situation after reading and speaking with people 1:1. There are still people I haven't managed to get ahold of and there could be some biases here, but I've tried my best and apologize if I've unfairly represented anyone's arguments.

Full RBF in general

There seems to be 3 groups of thought:

  1. we should actively move towards Full RBF

  2. we should try to prevent Full RBF for now, but end up there later

  3. we should try to prevent Full RBF from ever happening

The first position is largely held by protocol developers, and the second by some businesses. The last position seems to be very rare.

Why we should actively move towards Full RBF

  1. Full RBF is the natural state of the network. The point of Bitcoin blocks, PoW, etc. is to solve double-spending; there was never any guarantee of finality for unconfirmed transactions. When a miner receives two conflicting transactions, the incentive-compatible policy is to take the one giving them higher fees. Using an RBF policy helps nodes have a more accurate picture of which transaction is likely to confirm. A security assumption of "miners will be nice" when a non-signaling tx has a higher-feerate conflict, that is significantly weaker than "miners will do the rational thing."

  2. Full RBF is already observable, so we should be prepared to update our mempool policy to give us a more accurate view of what will be mined. If full RBF becomes widespread and my node is still assuming opt-in to be honored, it will reject transactions that are likely to be mined and potentially blind me to double-spend attempts.

  3. Enabling full RBF closes various signaling-based DoS/pinning attacks.

Common counterarguments:

  • Accepting unconfirmed transactions as final is actually safe in practice.
  • LN attacks are also uncommon in practice.

Why we should try to prevent Full RBF for now

  1. Full RBF will create a higher risk of double-spends for businesses accepting unconfirmed transactions as final. Accepting non-signaling unconfirmed transactions is unsafe in theory, but currently quite safe in practice, and a common way for Bitcoin users to send fast payments. LN exists, but the ecosystem is still building the technological and UX layers to make that option accessible to many users, and adoption is still low. Removing the option of using on-chain unconfirmed transactions might not actually encourage LN adoption, but drive users away instead. Dario Sneidermanis post

Common counterarguments:

  • Zeroconf really isn't safe; double spends are possible today.
  • If zeroconf is safe in practice, it's not really due to full RBF not happening, but because businesses are using precautions such as requiring high feerates and waiting to detect raced conflicting transactions in other (sometimes many) mempools. Businesses are also typically protected legally from users double-spending them.

Why we should try to prevent Full RBF from ever happening

  1. RBF in general creates a higher risk of double-spends for businesses accepting unconfirmed transactions as final. The current risk is close to zero. These businesses should be able to continue making these assumptions because it enables "fast" on-chain payments.

  2. American call option: if the exchange rate changes between transaction creation and confirmation, the user can replace the transaction to "cancel" it and create a new one.

Common counterarguments:

The -mempoolfullrbf option

Why we should remove the -mempoolfullrbf option for now

  1. Existence of the option cannot be distinguished from Bitcoin Core endorsing it. Bitcoin Core has not turned it on by default, but by adding the option, has signaled that it's safe for users to have this option.

  2. Adding the option makes it easier to run full RBF, so miners will turn it on, increasing the chances of non-signaling replacements propagating in practice. Only a minority is enough for non-signaled replacements to happen. Dario Sneidermanis post

  3. If this leads to full RBF before businesses are ready, they may be attacked and/or be forced to narrow the options for their (many) users. That could be bad for Bitcoin.

Why we should keep the -mempoolfullrbf

  1. One user of a software product should not be able to take options away from other users. We have some users saying "I want the option to configure my personal node to do this" and other users saying "I don't want other users to have this option because _." It seems unreasonable to say "yeah, sorry, you can't have this feature, not because we can't support it, but because another user said so."

  2. This is holding Bitcoin Core responsible for the decisions of individual node operators. It hasn't changed the default behavior. It added an option. Having this option doesn't mean full RBF happens; full RBF could still happen if we removed it. Pieter Wuille post

Common counterarguments:

  • If adding the option doesn't matter, then removing it shouldn't matter either. Anthony Towns post

Questions and Factual Disagreements

I think it's worth having a section on these fundamental disagreements, as they cause people to talk past one another in many discussions.

  • The prevalence of full RBF on the network
    • Peter Todd reports that he has observed replacements of non-signaling transactions using opentimestamps. He notes that they are rare, and it's possible that these replacements are due to other factors.
    • He also posted about this in 2016. That thread reads similarly to this year's discussion.
  • The current risk of double-spends by accepting "zeroconf"
  • The level of support for RBF and LN in the Bitcoin ecosystem
  • How much "notice" Bitcoin Core has given
    • Murch and others (particularly during the irc meeting) say that "we have been talking about full-rbf for seven years."
    • Antoine Riard posted to the mailing list in June 2021, more than 1 year before the option was merged.
    • David Harding notes that the post perhaps doesn't fully capture the potential impacts and risks of adding the option. Many people believe it can directly cause full RBF to be common on the network, so it should have been communicated that businesses must be ready for full RBF by then.

I have a few open questions, and would really appreciate being able to ask miners them:

  • If v24.0 contains the -mempoolfullrbf option, will any miners turn it on? If so, how soon?
  • If we removed the -mempoolfullrbf option, would any miners still end up doing full RBF (e.g. running with a patch)?

Moving Forward

Detecting Full RBF on the network

We should try to monitor how prevalent full RBF is on the network because, if it becomes very normal to see non-signaling replacements, it should be our firm recommendation to turn -mempoolfullrbf=1 on. A few things to keep an eye on:

  • Peter Todd's opentimestamps. Each transaction has its fees listed; anything higher than 182sat is a non-signaled replacement.
  • Individual experiments of non-signaled replacements.
  • Double-spend attempts on Muun, Bitrefill, and businesses that accept unconfirmed transactions as final.
  • Full RBF in other pieces of software. For example, implementing full RBF or adding behavior to assume replacements can happen without signaling.

Bitcoin Core v24.0

Bitcoin Core v24.0 was scheduled for release 19th of October 2022. Bug fixes also delayed the release, but I think this is the last blocker. I don't think "we're too close to change anything" is a good argument, but the urgency exists.

Full RBF was discussed in the weekly bitcoin-core-dev irc meeting. No decision was made to merge or move forward with any of PRs #26323 or #26287. Many supported #26305, but for a longer time period (1-2 releases or 1-2 years from now). Many disliked the idea of removing an option from users, especially if the option is desired and a safe practice for the individual node operator.

The biggest question to me is "will adding this option cause miners to switch to full RBF?" and I think it needs to be answered by miners and not through speculation. Just like full RBF proponents cannot definitively say "miners will do it because it's rational," I don't think people can say "miners will do it because it's easy to flip a config."

Dario posted an analysis of the available PRs from Muun's perspective. It is decidedly pro-full RBF but is interested in some predictability. For example, if it is expected that everyone will switch to full RBF at a predetermined time ~6 months from now, Muun has time to implement and deploy their solutions.

Rollout Timeline

Multiple people have suggested a timeline for coordinated deployment of full RBF:

I personally also believe the safest way to deploy full RBF is in a coordinated fashion aligning with when people agree the ecosystem would be ready. While this is not a consensus change, it is much safer for everyone on the network to switch over at the same time.

As mentioned earlier, if full RBF is starts happening on the network and most Bitcoin Core users don't have a way to configure their node to accept non-signaled replacements, their mempools become inaccurate. Even if we monitor the network closely and miners actively communicate that they are accepting non-signaling replacements, it is much safer to have nodes switch automatically than to try to tell everyone to restart with -mempoolfullrbf=1.

A timeline would put pressure on businesses to figure technical and UX things out, but something like 2 years would be very reasonable in my view. Similarly, if miners had been planning on switching to full RBF after the 2024 halving (i.e. because fees matter more as the subsidy decreases), this would line up with their plans so (I think) they probably wouldn't feel the need to do it through a patch.

However, I also think that merging some kind of lockin-on-timeout activation can cause an onslaught of accusations that Bitcoin Core is trying to force Full RBF. Trying to ask businesses who would be affected by full RBF "when might you be ready for full RBF?" seems to always be met with "never!" Though I think it's still worth trying.

Businesses that need help dealing with full RBF

Payment Replacement Detection

Payment Replacement Prevention