dumptxoutset rollback feature does not take forks into account #32817

1. 
   sipa commented at 12:35 pm on June 26, 2025: member

   In the dumptxoutset RPC, when the rollback feature is used with a height H, it will issue an InvalidateBlock on the active-chain block at height H+1. However, if an inactive branch of the chain exists that forks off at height H or below, and extends to block H+1 or later (or has equivalent PoW to that), InvalidateBlock will reorg there, and dump will be for the tip of that branch, rather than the height-H ancestor of the main chain.

   I have not tested this to verify, so it is possible I am missing something.

2. 
   Sjors commented at 2:03 pm on June 26, 2025: member

   cc @fjahr
3. 
   fjahr commented at 3:55 pm on June 26, 2025: contributor

   Sigh, yeah, that’s probably right. When taken to an extreme (aka Testnet4) that’s a really annoying problem as I have experienced in #31117. Since I have to deal with this anyway so I will address this soon.
4. 
   maflcko commented at 3:59 pm on June 26, 2025: member

   Isn’t this addressed by the throw here?

   https://github.com/bitcoin/bitcoin/blob/67ea4b9994e668dcea5e5d0f62f886d92e3737dc/src/rpc/blockchain.cpp#L3111

   I think there is still a race when calling submitblock, but this doesn’t seem related to forks per se.

5. 
   sipa commented at 4:03 pm on June 26, 2025: member

   @maflcko Right! So indeed, it won’t (ignoring race conditions) dump the wrong thing - but if you have a fork at a certain height, you’ll just be unable to dump at that height.
6. 
   maflcko commented at 4:08 pm on June 26, 2025: member

   you’ll just be unable to dump at that height.

   Could do a manual invalidateblock of the low-work forks to work around the error, but yea, could be annoying if the forks are more than 1, or longer than 1 block.

7. maflcko added the label RPC/REST/ZMQ on Jun 26, 2025
8. 
   mzumsande commented at 4:48 pm on June 26, 2025: contributor

   We could also automate it: Iterate over the block index to find all fork blocks at the target height and invalidate those too, only then do the invalidation on the main chain block / do the dump, and finally reconsider everything. Though I’m not sure if that would be overkill.
9. 
   Sjors commented at 9:51 am on June 27, 2025: member

   This goes back to the discussion in #29565 that we shouldn’t be using block invalidation as a hack to trigger the chain state manager into rolling back. Instead we should perform the rollback directly (though no need for a new RPC).

Contributors
sipa Sjors fjahr maflcko mzumsande

Labels
RPC/REST/ZMQ