Use journal instead of CacheKV for giga snapshot/rollback by codchen · Pull Request #3295 · sei-protocol/sei-chain

codchen · 2026-04-22T04:41:18Z

Problem

giga/deps/xevm/state implemented Snapshot()/RevertToSnapshot() by stacking CacheMultiStore layers: each Snapshot() call wrapped the current context in a new cache, and RevertToSnapshot() restored an older context (dropping the outer cache) to undo KV-store writes. This meant N nested EVM snapshots within a single transaction produced N nested CacheMultiStore layers, with all the associated allocation and copy-on-write overhead. The transient/in-memory state (access lists, logs, refunds, transient storage, account status) was already journal-driven; only the persisted KV state relied on the CacheMultiStore trick.

Solution

Adopt the go-ethereum journal pattern for all state—both KV-stored and in-memory:

One CacheMultiStore per stateDB, created at NewDBImpl. It isolates the transaction's writes from the underlying store and is flushed once at Finalize(). No additional layers are created per snapshot.
Snapshot() = record journal length + revision ID. No context replacement, no cache allocation.
RevertToSnapshot() = binary-search for the revision, replay journal entries backward. Every KV mutation records the previous value in a new journal entry type; reverting calls the inverse keeper operation.

Event-manager isolation

Snapshot() pushes the current EventManager onto a stack and attaches a fresh one to the context. RevertToSnapshot() pops back to the snapshot's EM, discarding any events emitted inside the reverted range. Finalize() drains all surviving EMs (in order) into committedCtx.EventManager().

GetCommittedState

Previously read from snapshottedCtxs[0] (the context before the first stateDB-level CMS layer). Now reads from committedCtx, the original context saved at NewDBImpl time, which sits below the single stateDB CMS and therefore always reflects the pre-transaction committed state.

Other changes

EVMKeeper interface gains SetAddressMapping so deleteMappingChange can restore address associations on revert.
clearAccountState (used by Finalize) is unchanged and un-journaled; a new clearAccountStateJournaled wrapper (called from CreateAccount) captures prev state before clearing.
DeepCopy now uses maps.Copy instead of manual loops.

Tests (snapshot_test.go, 20 new cases)

codeinputmachine · 2026-04-22T04:52:16Z

Hey @codchen, Code Input detected this PR has a merge conflict. Some of the conflicts of this PR can be resolved with a semantic merge driver. Code Input can do that automatically: https://codeinput.com/r/EESR7EMMZd3 Let me know if you need more help with this conflict or how Code Input works.

github-actions · 2026-04-23T04:38:13Z

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

Build	Format	Lint	Breaking	Updated (UTC)
`✅ passed`	`✅ passed`	`✅ passed`	`✅ passed`	May 5, 2026, 3:47 AM

codecov · 2026-04-23T05:15:26Z

Codecov Report

❌ Patch coverage is 71.83099% with 40 lines in your changes missing coverage. Please review.
✅ Project coverage is 59.09%. Comparing base (4ad975b) to head (32842ce).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
giga/deps/xevm/state/statedb.go	44.23%	29 Missing ⚠️
giga/deps/xevm/state/journal.go	75.00%	4 Missing and 4 partials ⚠️
giga/deps/xevm/state/state.go	96.15%	1 Missing and 1 partial ⚠️
giga/deps/xevm/state/nonce.go	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3295      +/-   ##
==========================================
+ Coverage   59.07%   59.09%   +0.01%     
==========================================
  Files        2099     2099              
  Lines      172988   173050      +62     
==========================================
+ Hits       102195   102258      +63     
+ Misses      61922    61918       -4     
- Partials     8871     8874       +3

Flag	Coverage Δ
sei-chain-pr	`71.97% <71.83%> (?)`
sei-db	`70.41% <ø> (ø)`

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

Files with missing lines	Coverage Δ
giga/deps/xevm/state/balance.go	`61.53% <100.00%> (+0.86%)`	⬆️
giga/deps/xevm/state/code.go	`78.57% <100.00%> (+1.64%)`	⬆️
giga/deps/xevm/state/nonce.go	`75.00% <66.66%> (+8.33%)`	⬆️
giga/deps/xevm/state/state.go	`97.33% <96.15%> (+0.63%)`	⬆️
giga/deps/xevm/state/journal.go	`80.82% <75.00%> (-4.90%)`	⬇️
giga/deps/xevm/state/statedb.go	`42.58% <44.23%> (+4.60%)`	⬆️

🚀 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.

arajasek · 2026-04-27T13:08:17Z

+	denom := s.k.GetBaseDenom(s.ctx)
+	if e.isAdd {
+		// Was AddBalance: reverse by subtracting
+		_ = s.k.BankKeeper().SubUnlockedCoins(ctx, e.seiAddr, sdk.NewCoins(sdk.NewCoin(denom, e.usei)), true)


do we want to be propagating the errors here?

i think it isn't expected to err here. Maybe we should just panic out?

I think panicking might be the most reasonable thing, honestly.

made it panic

philipsu522 · 2026-04-28T15:54:24Z

+	s.ctx.MultiStore().(sdk.CacheMultiStore).Write()
+	s.ctx.GigaMultiStore().WriteGiga()
+
+	// Emit all surviving events (from snapshots + current) to the committed ctx's EventManager.


do we have tests before/after (even existing one) to ensure that the same events are emitted before & after?

philipsu522 · 2026-04-28T16:37:27Z

+	denom := s.k.GetBaseDenom(s.ctx)
+	if e.isAdd {
+		// Was AddBalance: reverse by subtracting
+		_ = s.k.BankKeeper().SubUnlockedCoins(ctx, e.seiAddr, sdk.NewCoins(sdk.NewCoin(denom, e.usei)), true)


philipsu522 · 2026-04-28T16:40:14Z

 	s.tempState = NewTemporaryState()
 	s.journal = []journalEntry{}
-	s.snapshottedCtxs = []sdk.Context{}
-	s.Snapshot()


we used to create a new Snapshot() after cleanup, do we want to enforce that invariant in the new journaled state?

if we don't want it, maybe we add a test enforcing that this isn't required after CleanupForTracer()

…er CleanupForTracer, add event/cleanup tests Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…op SSTOREs) Matches go-ethereum behaviour: if the slot already holds the target value, return early without a KV write or journal entry. Previously N identical SSTOREs to the same slot produced N journal entries. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

codchen added the non-app-hash-breaking label Apr 22, 2026

codchen force-pushed the tony/inter-block-cache-2 branch from 1608a35 to 48a18ad Compare April 23, 2026 04:37

codchen force-pushed the tony/inter-block-cache-2 branch from 48a18ad to 00974b3 Compare April 23, 2026 05:12

codchen force-pushed the tony/inter-block-cache-2 branch from 00974b3 to 9480ea7 Compare April 24, 2026 03:57

arajasek approved these changes Apr 27, 2026

View reviewed changes

codchen requested a review from philipsu522 April 28, 2026 13:37

philipsu522 reviewed Apr 28, 2026

View reviewed changes

codchen force-pushed the tony/inter-block-cache-2 branch from 9480ea7 to 8185f50 Compare May 4, 2026 04:16

philipsu522 approved these changes May 4, 2026

View reviewed changes

codchen and others added 3 commits May 5, 2026 11:26

Use journal instead of CacheKV for giga snapshot/rollback

5f1e54b

gofmt

3de922e

address review comments: panic on balance revert errors, snapshot aft…

bca2f2e

…er CleanupForTracer, add event/cleanup tests Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

codchen force-pushed the tony/inter-block-cache-2 branch from 8185f50 to bca2f2e Compare May 5, 2026 03:27

Conversation

codchen commented Apr 22, 2026

Problem

Solution

Event-manager isolation

GetCommittedState

Other changes

Tests (snapshot_test.go, 20 new cases)

Uh oh!

codeinputmachine commented Apr 22, 2026

Uh oh!

github-actions Bot commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

github-actions Bot commented Apr 23, 2026 •

edited

Loading

codecov Bot commented Apr 23, 2026 •

edited

Loading