[seidroid]

Backport of #4321 to release/v6.7.

[cursor]

PR Summary

Medium Risk
Touches MVCC prune lifecycle and Pebble compaction timing; low logic change but affects production disk usage during pruning.

Overview
Fixes a state-store pruning disk spike by closing the full-DB scan iterator before post-prune compaction runs, instead of holding it open until function return.

An open Pebble iterator pins readState, which prevents reclaiming obsolete SSTables while compactPrunedRange runs (often slowly on large deleted spans). The change adds an idempotent closeItr helper (defer still handles errors mid-scan) and logs compaction duration via seilog (pruneDescending: compacted pruned range).

^{Reviewed by Cursor Bugbot for commit d1aa21b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

The latest Buf updates on your PR. Results from workflow Buf / buf (pull_request).

Build	Format	Lint	Breaking	Updated (UTC)
✅ passed	✅ passed	✅ passed	✅ passed	Oct 12, 2026, 4:12 PM

[seidroid]

This is a clean backport of #4321 to release/v6.7. pruneDescending now closes its scan iterator before compactPrunedRange runs, so the iterator no longer pins Pebble's readState and stops obsolete sstables from being deleted during a long compaction. The guarded double-close is correct, the deleted-key bounds are cloned so closing the iterator early is safe, and the new seilog logger matches how other sei-db packages create theirs. I found no issues.

Findings: 0 blocking | 0 non-blocking | 0 posted inline

Blockers

• None at the file/PR level.

Non-blocking

• None at the file/PR level.

[codecov]

Codecov Report

❌ Patch coverage is 84.61538% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 60.80%. Comparing base (210b31b) to head (d1aa21b).

Files with missing lines	Patch %	Lines
sei-db/db_engine/pebbledb/mvcc/db.go	84.61%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@               Coverage Diff                @@
##           release/v6.7    #4508      +/-   ##
================================================
- Coverage         61.84%   60.80%   -1.04%     
================================================
  Files              2168     2068     -100     
  Lines            188739   176790   -11949     
================================================
- Hits             116728   107500    -9228     
+ Misses            61157    59406    -1751     
+ Partials          10854     9884     -970     

Flag	Coverage Δ
sei-chain-pr	64.74% <84.61%> (?)
sei-db	69.80% <ø> (ø)
sei-db-state-db	?

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
sei-db/db_engine/pebbledb/mvcc/db.go	67.43% <84.61%> (+0.21%)	⬆️

... and 100 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[alexander-sei]

@seidroid review

[seidroid]

Backport of #4321 to release/v6.7. pruneDescending now closes its scan iterator before the range compaction runs, so the iterator's pinned readState no longer keeps obsolete sstables on disk during compaction, and the compaction time is now logged. The change is correct, because the deleted-span keys are cloned before the iterator closes and the guarded close prevents a double close, so I approve; codex's reading found nothing, which matches mine, and it added no findings to keep or drop.

Pre-existing

Already true on the base branch, not introduced here.

• suggestion — pruneAscending in sei-db/db_engine/pebbledb/mvcc/db_ascending.go:121 still defers itr.Close() until after compactPrunedRange runs (line 232). Nodes on legacy ascending-encoded databases therefore still get the same pinned-sstable disk spike that this change fixes for descending databases. The upstream fix (#4321) and main have the same gap.

1 nit, not posted on the code

• sei-db/db_engine/pebbledb/mvcc/db.go:1052 — A failed compaction is logged at Info level with only an err field. Logging failures at Error/Warn would make them easier to find and alert on.

_{seidroid review · decision approve · session 7292a75a5c3f4e06a249756ac21ac64d · turn resp_claude_aa68e640ab1b227d8cb80bf89e4283a7 · item a078edaa8fcc56e0abdf738f10363ab9}

Findings: 0 blocking | 0 non-blocking | 0 posted inline | 1 pre-existing

[alexander-sei]

@seidroid review

[seidroid]

Nothing material has changed since my last review: the diff is the same db.go change (blob 421fe4dcf8..df65836d87), which closes the pruneDescending scan iterator before range compaction so obsolete sstables stop being pinned, and the updated merge tree reads the same around it. It is still correct and the earlier findings still stand unchanged; codex's reading found nothing, which matches mine, and it added no findings to keep or drop.

Pre-existing

Already true on the base branch, not introduced here.

• suggestion — pruneAscending in sei-db/db_engine/pebbledb/mvcc/db_ascending.go:121 still defers itr.Close() until after compactPrunedRange runs (line 232). Nodes on legacy ascending-encoded databases therefore still get the same pinned-sstable disk spike that this change fixes for descending databases. The upstream fix (#4321) and main have the same gap.

1 nit, not posted on the code

• sei-db/db_engine/pebbledb/mvcc/db.go:1052 — A failed compaction is logged at Info level with only an err field. Logging failures at Error/Warn would make them easier to find and alert on.

_{seidroid review · decision approve · session 7292a75a5c3f4e06a249756ac21ac64d · turn resp_claude_a09970af0eb40829d53380487cc36e8d · item 945d399541095a9a953d563a4e01a32b}

Findings: 0 blocking | 0 non-blocking | 0 posted inline | 1 pre-existing

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 480fc8c. Configure here.}

[cursor]

Ascending prune still pins read state

High Severity

pruneAscending still closes its full-database iterator only in a defer, so the iterator stays open across compactPrunedRange. That pins Pebble's read state for the whole compaction, so obsolete sstables cannot be deleted and prune tombstones cannot be elided. Legacy databases stay on this path, so the disk spike remains for them.

 

^{Triggered by learned rule: sei-db pebbledb: ascending and descending MVCC iterators must stay aligned}

^{Reviewed by Cursor Bugbot for commit 480fc8c. Configure here.}