stmgr: only persist the migration cache on success

Otherwise, we may end up referencing blocks we never wrote to disk.

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)
 					}()
 				},
 			})

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 {

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

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=