From a93a18678fc6a58c0f341be23134857b5cc3861e Mon Sep 17 00:00:00 2001
From: Steven Allen <steven@stebalien.com>
Date: Tue, 26 Jan 2021 17:08:06 -0800
Subject: [PATCH] stmgr: only persist the migration cache on success

Otherwise, we may end up referencing blocks we never wrote to disk.
---
 chain/stmgr/forks.go | 18 +++++++++++++++++-
 chain/stmgr/stmgr.go |  2 +-
 go.mod               |  2 +-
 go.sum               |  4 ++--
 4 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/chain/stmgr/forks.go b/chain/stmgr/forks.go
index 268db11ed32..5018653c76e 100644
--- a/chain/stmgr/forks.go
+++ b/chain/stmgr/forks.go
@@ -229,10 +229,17 @@ func (sm *StateManager) handleStateForks(ctx context.Context, root cid.Cid, heig
 	var err error
 	u := sm.stateMigrations[height]
 	if u != nil && u.upgrade != nil {
-		retCid, err = u.upgrade(ctx, sm, u.cache, cb, root, height, ts)
+		// Yes, we clone the cache, even for the final upgrade epoch. Why? Reverts. We may
+		// have to migrate multiple times.
+		tmpCache := u.cache.Clone()
+		retCid, err = u.upgrade(ctx, sm, tmpCache, cb, root, height, ts)
 		if err != nil {
 			return cid.Undef, err
 		}
+		// Yes, we update the cache, even for the final upgrade epoch. Why? Reverts. This
+		// can save us a _lot_ of time because very few actors will have changed if we
+		// do a small revert then need to re-run the migration.
+		u.cache.Update(tmpCache)
 	}
 
 	return retCid, nil
@@ -279,11 +286,20 @@ func (sm *StateManager) preMigrationWorker(ctx context.Context) {
 					wg.Add(1)
 					go func() {
 						defer wg.Done()
+
+						// Clone the cache so we don't actually _update_ it
+						// till we're done. Otherwise, if we fail, the next
+						// migration to use the cache may assume that
+						// certain blocks exist, even if they don't.
+						tmpCache := cache.Clone()
 						err := migrationFunc(preCtx, sm, cache, ts.ParentState(), ts.Height(), ts)
 						if err != nil {
 							log.Errorw("failed to run pre-migration",
 								"error", err)
+							return
 						}
+						// Finally, if everything worked, update the cache.
+						cache.Update(tmpCache)
 					}()
 				},
 			})
diff --git a/chain/stmgr/stmgr.go b/chain/stmgr/stmgr.go
index b0018ddc01f..8598ac23d29 100644
--- a/chain/stmgr/stmgr.go
+++ b/chain/stmgr/stmgr.go
@@ -66,7 +66,7 @@ type versionSpec struct {
 type migration struct {
 	upgrade       MigrationFunc
 	preMigrations []PreMigration
-	cache         MigrationCache
+	cache         *nv10.MemMigrationCache
 }
 
 type StateManager struct {
diff --git a/go.mod b/go.mod
index 5a062f34d49..0a4c4dfc3b5 100644
--- a/go.mod
+++ b/go.mod
@@ -44,7 +44,7 @@ require (
 	github.com/filecoin-project/go-storedcounter v0.0.0-20200421200003-1c99c62e8a5b
 	github.com/filecoin-project/specs-actors v0.9.13
 	github.com/filecoin-project/specs-actors/v2 v2.3.4
-	github.com/filecoin-project/specs-actors/v3 v3.0.1-0.20210126204401-0e4fdfa27dea
+	github.com/filecoin-project/specs-actors/v3 v3.0.1-0.20210127005641-3320300e4add
 	github.com/filecoin-project/specs-storage v0.1.1-0.20201105051918-5188d9774506
 	github.com/filecoin-project/test-vectors/schema v0.0.5
 	github.com/gbrlsnchs/jwt/v3 v3.0.0-beta.1
diff --git a/go.sum b/go.sum
index 6f5f1699d03..7debbf8dc35 100644
--- a/go.sum
+++ b/go.sum
@@ -307,8 +307,8 @@ github.com/filecoin-project/specs-actors/v2 v2.3.2 h1:2Vcf4CGa29kRh4JJ02m+FbvD/p
 github.com/filecoin-project/specs-actors/v2 v2.3.2/go.mod h1:UuJQLoTx/HPvvWeqlIFmC/ywlOLHNe8SNQ3OunFbu2Y=
 github.com/filecoin-project/specs-actors/v2 v2.3.4 h1:NZK2oMCcA71wNsUzDBmLQyRMzcCnX9tDGvwZ53G67j8=
 github.com/filecoin-project/specs-actors/v2 v2.3.4/go.mod h1:UuJQLoTx/HPvvWeqlIFmC/ywlOLHNe8SNQ3OunFbu2Y=
-github.com/filecoin-project/specs-actors/v3 v3.0.1-0.20210126204401-0e4fdfa27dea h1:TPMvEAUP4bUTDu7LaKkk0zKhtQHEPuOI5C4cEZls2ho=
-github.com/filecoin-project/specs-actors/v3 v3.0.1-0.20210126204401-0e4fdfa27dea/go.mod h1:aVf248CfjfyCmVel4UuFAA3u+9UQjqtqHpgfYv+M+9U=
+github.com/filecoin-project/specs-actors/v3 v3.0.1-0.20210127005641-3320300e4add h1:LvS5fTSY5HgibaXbFEscox2SfiuBlS8RrVgBdGbozOE=
+github.com/filecoin-project/specs-actors/v3 v3.0.1-0.20210127005641-3320300e4add/go.mod h1:aVf248CfjfyCmVel4UuFAA3u+9UQjqtqHpgfYv+M+9U=
 github.com/filecoin-project/specs-storage v0.1.1-0.20201105051918-5188d9774506 h1:Ur/l2+6qN+lQiqjozWWc5p9UDaAMDZKTlDS98oRnlIw=
 github.com/filecoin-project/specs-storage v0.1.1-0.20201105051918-5188d9774506/go.mod h1:nJRRM7Aa9XVvygr3W9k6xGF46RWzr2zxF/iGoAIfA/g=
 github.com/filecoin-project/test-vectors/schema v0.0.5 h1:w3zHQhzM4pYxJDl21avXjOKBLF8egrvwUwjpT8TquDg=