fanquake 34ae04d775 Merge bitcoin/bitcoin#21726: Improve Indices on pruned nodes via prune blockers ...

71c3f0356c move-only: Rename index + pruning functional test (Fabian Jahr)
de08932efa test: Update test for indices on pruned nodes (Fabian Jahr)
825d19839b Index: Allow coinstatsindex with pruning enabled (Fabian Jahr)
f08c9fb0c6 Index: Use prune locks for blockfilterindex (Fabian Jahr)
2561823531 blockstorage: Add prune locks to BlockManager (Fabian Jahr)
231fc7b035 refactor: Introduce GetFirstStoredBlock helper function (Fabian Jahr)

Pull request description:

  # Motivation
  The main motivation of this change and only behavior change noticeable by user is to allow running `coinstatsindex` on pruned nodes as has been requested [here for example](https://twitter.com/benthecarman/status/1388170854140452870?s=20).

  # Background
  `coinstatsindex` on pruned nodes can be enabled in a much simpler than it is done here but it comes with downside. The ability to run `blockfilterindex`on pruned nodes was added in #15946 but it also added the `blockfilterindex` as a dependency to `validation` and it introduced two new circular dependencies. Enabling `coinstatsindex` on pruned nodes in a similar way would add it as a dependency as well and introduce another circular dependency.

  Instead, this PR introduces a `m_prune_blockers` map to `BlockManager` as a flexible approach to block pruning. Entities like `blockfilterindex`, for example, can add a key and a height to block pruning over that height. These entities need to update that value to allow more pruning when they are ready.

  # Alternative approach
  Upon completing the first draft of this PR I found #19463 as an alternative that follows the same but follows a very different approach. I am listing the main differences here as I see them:
  - Usage of globals
  - Blocks pruning with a start and a stop height
  - Can persist blockers across restarts
  - Blockers can be set/unset via RPCs

  Personally, I don't think any of these are necessary to be added here but if the general approach or specific features are more appealing to reviewers I am happy to change to a solution based on that PR or port over specific parts of it here.

ACKs for top commit:
  mzumsande:
    Code review ACK 71c3f0356c
  ryanofsky:
    Code review ACK 71c3f0356c. Changes since last review: just tweaking comments and asserts, and rebasing
  w0xlt:
    tACK 71c3f0356c on signet.

Tree-SHA512: de7efda08b44aa31013fbebc47a02cd2de32db170b570f9643e1f013fee0e8e7ca3068952d1acc6e5e74a70910735c5f263437981ad73df841ad945b52d36b71

2022-04-26 19:42:45 +01:00

32 KiB

Executable File

Raw Blame History

View Raw