validation: don't accept txs to the mempool while loading a snapshot #36346

pull 0xShadowX wants to merge 1 commits into bitcoin:master from 0xShadowX:fix-loadtxoutset-mempool-recheck changing 3 files +82 −0

0xShadowX commented at 5:16 PM on September 26, 2026: none
ActivateSnapshot checks that the mempool is empty before it starts, but PopulateAndValidateSnapshot then runs without cs_main, which takes minutes on mainnet. A transaction entering the mempool in that window (e.g. through sendrawtransaction) hits the assert in AddChainstate:
```
Assertion failed: (!prev_chainstate.m_mempool || prev_chainstate.m_mempool->size() == 0), function AddChainstate, file validation.cpp, line 6201.
```
By then the base blockhash file has been written, so the node needs -reindex to start again.

This sets a flag on ChainstateManager while ActivateSnapshot runs, and AcceptToMemoryPool and ProcessNewPackage reject with snapshot-loading while it's set. testmempoolaccept still works. A second loadtxoutset started while one is running is also rejected.

test_mempool_blocked_during_load in feature_assumeutxo.py pauses the load through a FIFO, checks that sendrawtransaction and submitpackage are rejected and the mempool stays empty, then lets the load finish. On master the transaction is accepted and the node aborts. The test is skipped where os.mkfifo isn't available.
DrahtBot added the label Validation on Sep 26, 2026
DrahtBot commented at 5:16 PM on September 26, 2026: 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/36346.



Reviews

See the guideline and AI policy for information on the review process. A summary of reviews will appear here.
fjahr commented at 10:07 PM on September 26, 2026: contributor

I haven't looked at the code or verified the issue but it seems to me the correct fix would be to prevent the user from putting a tx into the mempool when we don't want one to get there instead of letting them do it and then fail, even if it's a graceful failure with this change.

validation: don't accept txs to the mempool while loading a snapshot

ActivateSnapshot checks that the mempool is empty, then releases cs_main
while PopulateAndValidateSnapshot runs, which can take minutes. A
transaction added to the mempool in that window (e.g. through
sendrawtransaction) made AddChainstate hit its empty-mempool assert after
the base blockhash file was already written, and the node then needed
-reindex to start again.

Set a flag on ChainstateManager for the duration of ActivateSnapshot and
reject transactions and packages in AcceptToMemoryPool and
ProcessNewPackage while it is set. test_accept calls still work since
they don't add anything to the mempool. A second concurrent loadtxoutset
is rejected while the flag is set.

ca21d8f5fb

0xShadowX force-pushed on Sep 27, 2026
0xShadowX renamed this:
~~validation: recheck mempool before activating a snapshot~~
validation: don't accept txs to the mempool while loading a snapshot
on Sep 27, 2026
0xShadowX commented at 7:55 AM on September 27, 2026: none

Makes sense, changed it to that. While a snapshot is loading, AcceptToMemoryPool and ProcessNewPackage now reject with "snapshot-loading" (test_accept still works), so the load no longer fails because of the mempool. Updated the description and the test.

Contributors

Labels

validation: don't accept txs to the mempool while loading a snapshot #36346

Code Coverage & Benchmarks

Reviews