Prune locks

luke-jr commented at 4:03 am on July 8, 2020: member

This adds the ability to have any number of internal or external locks to prevent pruning specific block ranges.

Some examples where this would be useful:

Ensuring any known Core wallet won’t become unable to sync due to pruning when it is unloaded. (Included in this PR)
Ensuring pruning doesn’t go too far for a desired-functional wallet backup. (Questionable, since you probably lose the node if you need the backup anyway?)
Allowing users to temporarily disable block indexes and filters, without pruning cutting them off from catching up later.
Avoiding pruning out from under external wallets (Armory, EPS, etc) that might need block data.
External block indexes/filters, etc.

By design, users retain complete control over prune locks, and can delete them out from under applications using them. This is to avoid circumstances where a prune lock has been set, by an application that may no longer be used.

luke-jr force-pushed on Jul 8, 2020

DrahtBot commented at 4:15 am on July 8, 2020: 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/19463.

Reviews

See the guideline for information on the review process.

Type	Reviewers
Concept ACK	fjahr

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:

#31860 (init: Take lock on blocks directory in BlockManager ctor by TheCharlatan)
#29652 (wallet: Avoid potentially writing incorrect best block locator by ryanofsky)
#29418 (rpc: provide per message stats for global traffic via new RPC ‘getnetmsgstats’ by vasild)

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 RPC/REST/ZMQ on Jul 8, 2020

DrahtBot added the label UTXO Db and Indexes on Jul 8, 2020

DrahtBot added the label Validation on Jul 8, 2020

DrahtBot added the label Wallet on Jul 8, 2020

andronoob commented at 4:41 am on July 9, 2020: none

I think redownloading specified block could be much more useful, just like the case that scantxoutset gives the corresponding block height, but the actual block had already been pruned - the ability to redownload (and then rescan) specified block fix this problem.

andronoob commented at 4:45 am on July 9, 2020: none

Recently I heard (could be inaccurate/misunderstanding of myself) that some projects like BTCPay didn’t keep full transaction data of received funds. Only the UTXO entry (txid, vout, amount and scriptPubkey) was kept. Therefore they are facing problems to fix the BIP143 vulnerability, because the full data of the previous transaction had been already discarded. I think the ability to redownload & rescan specified block(s) can relieve this problem.

luke-jr commented at 4:04 pm on July 9, 2020: member

I think redownloading specified block could be much more useful, just like the case that scantxoutset gives the corresponding block height, but the actual block had already been pruned - the ability to redownload (and then rescan) specified block fix this problem.

That’s useful too, but for different use cases. You wouldn’t want to be constantly redownloading the same block over and over. (Prune locks might be useful for it, though, to prevent the redownloaded block from being immediately repruned before use.)

Recently I heard (could be inaccurate/misunderstanding of myself) that some projects like BTCPay didn’t keep full transaction data of received funds. Only the UTXO entry (txid, vout, amount and scriptPubkey) was kept.

Aside, that’s a pretty serious bug. It’s the receiver’s responsibility to rebroadcast the transaction until it confirms…

laanwj added the label Block storage on Jul 9, 2020

laanwj removed the label Wallet on Jul 9, 2020

laanwj removed the label Validation on Jul 9, 2020

laanwj removed the label UTXO Db and Indexes on Jul 9, 2020

luke-jr force-pushed on Jul 24, 2020

luke-jr force-pushed on Jul 28, 2020

andronoob commented at 7:18 am on July 28, 2020: none

Allowing users to temporarily disable block indexes and filters, without pruning cutting them off from catching up later.

Honestly I don’t get what’s the point of this - to save a little more disk space under extreme conditions, like almostly exhausted disk space, so that the node could then keep running for a littel more time?

External block indexes/filters, etc.

This seems to justify the above point, however I think one of the most-wanted features is “external (and pluggable) block storage”, rather than “external indices/filters etc”, although the later also seems to be nice (despite both would complicate the things further).

Avoiding pruning out from under external wallets (Armory, EPS, etc) that might need block data.

Ensuring any known Core wallet won’t become unable to sync due to pruning when it is unloaded. (Included in this PR)

Then, (without some filter/matching mechanism) it will in fact disable (pause) pruning until the wallet reloads, won’t it? Because the node won’t know whether a newly received block is unrelated (prunable), therefore the only choice is to stop pruning.

It has been a significant usability problem since quite a long time ago, that the user cannot load/import wallets/keys/addrs freely - nobody likes triggering the troubles of blockchain rescanning.

I think #15946 (together with something like #15845) should be able to fix such usability problem in a simpler,better way, that:

With local blockfilterindex, the blocks could be simply discarded (pruned). Then, the once pruned blocks could be redownloaded according to local blockfilterindex while reloading wallet/importing keys/addrs - no need to worry about whether a block could be needed in the future anymore.

However I still agree that a button allowing the user to simply “pause/resume” pruning would be nice, too - I hope that this could be displayed on some dashboard.

andronoob commented at 7:58 am on July 28, 2020: none

That’s useful too, but for different use cases. You wouldn’t want to be constantly redownloading the same block over and over. (Prune locks might be useful for it, though, to prevent the redownloaded block from being immediately repruned before use.)

Yes, prune locks indeedly have its utility, especially to lock needed/related historical blocks. I just think that I see greater problems.

luke-jr force-pushed on Jul 29, 2020

luke-jr commented at 5:45 pm on July 29, 2020: member

Removed wallet support for this PR, rebased, added test, and passing CI.

in src/rpc/blockchain.cpp:930 in 8d43c3cbc0 outdated

925+                {
926+                    {RPCResult::Type::OBJ, "", "",
927+                    {
928+                        {RPCResult::Type::STR, "id", "A unique identifier for the lock"},
929+                        {RPCResult::Type::STR, "desc", "A description of the lock's purpose"},
930+                        {RPCResult::Type::STR, "height", "A description of the lock's purpose"},

kristapsk commented at 6:52 pm on July 29, 2020:

This line should be removed I think (duplicate “height” with wrong description).

luke-jr commented at 7:44 pm on July 29, 2020:

Fixed

luke-jr force-pushed on Jul 29, 2020

DrahtBot added the label Needs rebase on Sep 22, 2020

luke-jr force-pushed on Nov 19, 2020

DrahtBot removed the label Needs rebase on Nov 19, 2020

DrahtBot added the label Needs rebase on Dec 1, 2020

fjahr commented at 0:01 am on April 19, 2021: contributor

Concept ACK on solving this for internal purposes. I started working on the same issue with #21726 before I found this. I am only focused on finding a simple solution for internal issues, specifically indices that are still syncing. I am not sure if there is much need for the additional features that this PR provides (RPCs etc.). But happy to consolidate ideas/approaches if that seems useful to other reviewers.

luke-jr force-pushed on Jul 7, 2021

luke-jr commented at 0:39 am on July 7, 2021: member

Rebased

DrahtBot removed the label Needs rebase on Jul 7, 2021

DrahtBot added the label Needs rebase on Jul 20, 2021

luke-jr commented at 7:31 pm on February 4, 2022: member

Closing this in favour of #21726 for now (this PR had some more functionality, but that can be added in later)

luke-jr closed this on Feb 4, 2022

fanquake referenced this in commit 34ae04d775 on Apr 26, 2022

luke-jr reopened this on Nov 12, 2022

luke-jr force-pushed on Nov 12, 2022

luke-jr commented at 5:04 am on November 12, 2022: member

Rebased

DrahtBot removed the label Needs rebase on Nov 12, 2022

luke-jr force-pushed on Nov 12, 2022

luke-jr commented at 4:58 pm on November 12, 2022: member

A new circular dependency in the form of “node/blockstorage -> txdb -> node/blockstorage” appears to have been introduced.

CI is falsely equating headers and source files. I don’t see a clean way to resolve this (besides fixing CI to distinguish properly)

DrahtBot added the label Needs rebase on Feb 17, 2023

achow101 commented at 3:29 pm on April 25, 2023: member

Are you still working on this?

luke-jr force-pushed on Jul 8, 2023

luke-jr commented at 1:58 am on July 8, 2023: member

Are you still working on this?

Yes, though it seems even rebasing on branch-25 still results in conflicts, so I’ll have to get back to it after Knots v25.

achow101 commented at 4:15 pm on September 20, 2023: member

Are you still working on this?

achow101 marked this as a draft on Sep 20, 2023

achow101 commented at 4:16 pm on September 20, 2023: member

Converting to draft as this is not ready for review.

luke-jr force-pushed on Dec 19, 2023

DrahtBot added the label CI failed on Dec 19, 2023

DrahtBot removed the label Needs rebase on Dec 19, 2023

luke-jr force-pushed on Dec 19, 2023

luke-jr force-pushed on Dec 21, 2023

DrahtBot removed the label CI failed on Dec 21, 2023

luke-jr marked this as ready for review on Dec 21, 2023

luke-jr commented at 11:13 pm on December 21, 2023: member

Rebased

luke-jr requested review from kristapsk on Dec 21, 2023

DrahtBot added the label CI failed on Jan 24, 2024

achow101 removed review request from kristapsk on Apr 9, 2024

achow101 requested review from mzumsande on Apr 9, 2024

DrahtBot marked this as a draft on Apr 23, 2024

DrahtBot added the label Needs rebase on Jul 11, 2024

luke-jr force-pushed on Sep 3, 2024

Roll back (tip - 1) prune locks when doing a reorg 8ee1214157

Move prune lock checking into BlockManager 0b4bd4e134

blockstorage: Add height_last to PruneLockInfo f4b3368b13

Add internal interfaces for prune locks

Including desc member of PruneLockInfo for a human-readable description

b4e1c9614e

Support for persisting prune locks in blocks/index db 4822c21812

RPC: blockchain: Add listprunelocks and setprunelock methods b64d1e9b7b

QA: Test prune locks via RPC 196b501198

RPC/Blockchain: Optimise setprunelock "delete all" slightly ef060ac6a5

Bugfix: QA/fuzz: Add listprunelocks and setprunelock to RPC_COMMANDS_SAFE_FOR_FUZZING 1d1c92d40d

Bugfix: RPC: blockchain: Actually include "temporary" flag in listprunelocks result 162f0dba2f

luke-jr force-pushed on Apr 6, 2025

luke-jr marked this as ready for review on Apr 6, 2025

DrahtBot removed the label Needs rebase on Apr 6, 2025

DrahtBot removed the label CI failed on Apr 6, 2025

DrahtBot closed this on Apr 7, 2025

DrahtBot reopened this on Apr 7, 2025

in src/validation.cpp:3094 in 8ee1214157 outdated

3087@@ -3088,13 +3088,13 @@ bool Chainstate::DisconnectTip(BlockValidationState& state, DisconnectedBlockTra
3088              Ticks<MillisecondsDouble>(SteadyClock::now() - time_start));
3089 
3090     {
3091-        // Prune locks that began at or after the tip should be moved backward so they get a chance to reorg
3092+        // Prune locks that began around the tip should be moved backward so they get a chance to reorg
3093         const int max_height_first{pindexDelete->nHeight - 1};
3094         for (auto& prune_lock : m_blockman.m_prune_locks) {
3095-            if (prune_lock.second.height_first <= max_height_first) continue;
3096+            if (prune_lock.second.height_first < max_height_first) continue;

mzumsande commented at 5:00 pm on April 7, 2025:

Commit 8ee12141570140f56f5cfefd793825a8787656a6: would be good to have an explanation why this is being changed in the commit message.

I guess this means that if we have a chain at height N, have an index synced to height N-1 and disconnect the tip, we would change the prune lock of the index from N-1 to N-2. Why?

in src/node/blockstorage.h:100 in f4b3368b13 outdated

 96@@ -97,6 +97,7 @@ struct CBlockIndexHeightOnlyComparator {
 97 
 98 struct PruneLockInfo {
 99     int height_first{std::numeric_limits<int>::max()}; //! Height of earliest block that should be kept and not pruned
100+    int height_last{std::numeric_limits<int>::max()}; //! Height of latest block that should be kept and not pruned

mzumsande commented at 8:53 pm on April 7, 2025:

Is there a use case for allowing users to specify height_last, which would allow blocks higher than that to be pruned, so that anything requiring to sync the entire chain at a later point wouldn’t work anymore?

in src/rpc/blockchain.cpp:972 in b64d1e9b7b outdated

967+        }
968+    } else {
969+        if (lockid == "*") throw JSONRPCError(RPC_INVALID_PARAMETER, "id \"*\" only makes sense when deleting");
970+        if (!height_param[0].isNum()) throw JSONRPCError(RPC_TYPE_ERROR, "Invalid start height");
971+        lock_info.height_first = height_param[0].getInt<uint64_t>();
972+        if (!height_param[1].isNull()) {

mzumsande commented at 8:56 pm on April 7, 2025:

Should probably return an error if height_first > height_last.

mzumsande commented at 9:40 pm on April 7, 2025: contributor

I can see how this could be useful for external applications in theory, but it would be good to know if there is an actual demand for this - I wonder if there are any applications out there that would actually use this / generate prune locks if this option was available for them?

Prune locks #19463

Code Coverage & Benchmarks

Reviews

Conflicts