Golang engine.MVCCIterate函数代码示例

// Iterate walks through the abort cache, invoking the given callback for
// each unmarshaled entry with the key, the transaction ID and the decoded
// entry.
// TODO(tschottdorf): should not use a pointer to UUID.
func (sc *AbortCache) Iterate(
	ctx context.Context, e engine.Reader, f func([]byte, roachpb.AbortCacheEntry),
) {
	_, _ = engine.MVCCIterate(ctx, e, sc.min(), sc.max(), hlc.ZeroTimestamp,
		true /* consistent */, nil /* txn */, false, /* !reverse */
		func(kv roachpb.KeyValue) (bool, error) {
			var entry roachpb.AbortCacheEntry
			if _, err := keys.DecodeAbortCacheKey(kv.Key, nil); err != nil {
				panic(err) // TODO(tschottdorf): ReplicaCorruptionError
			}
			if err := kv.Value.GetProto(&entry); err != nil {
				panic(err) // TODO(tschottdorf): ReplicaCorruptionError
			}
			f(kv.Key, entry)
			return false, nil
		})
}

func iterateEntries(
	ctx context.Context,
	e engine.Reader,
	rangeID roachpb.RangeID,
	lo,
	hi uint64,
	scanFunc func(roachpb.KeyValue) (bool, error),
) error {
	_, err := engine.MVCCIterate(
		ctx, e,
		keys.RaftLogKey(rangeID, lo),
		keys.RaftLogKey(rangeID, hi),
		hlc.ZeroTimestamp,
		true,  /* consistent */
		nil,   /* txn */
		false, /* !reverse */
		scanFunc,
	)
	return err
}

//.........这里部分代码省略.........
	for strKey, test := range testCases {
		baseKey := roachpb.Key(strKey)
		txnClock := hlc.NewClock(hlc.NewManualClock(int64(test.orig)).UnixNano)
		txn := newTransaction("txn1", baseKey, 1, enginepb.SERIALIZABLE, txnClock)
		txn.Status = test.status
		txn.Intents = testIntents
		if test.hb > 0 {
			txn.LastHeartbeat = &hlc.Timestamp{WallTime: int64(test.hb)}
		}
		// Set a high Timestamp to make sure it does not matter. Only
		// OrigTimestamp (and heartbeat) are used for GC decisions.
		txn.Timestamp.Forward(hlc.MaxTimestamp)
		txns[strKey] = *txn
		for _, addrKey := range []roachpb.Key{baseKey, outsideKey} {
			key := keys.TransactionKey(addrKey, txn.ID)
			if err := engine.MVCCPutProto(context.Background(), tc.engine, nil, key, hlc.ZeroTimestamp, nil, txn); err != nil {
				t.Fatal(err)
			}
		}
		entry := roachpb.AbortCacheEntry{Key: txn.Key, Timestamp: txn.LastActive()}
		if err := tc.rng.abortCache.Put(context.Background(), tc.engine, nil, txn.ID, &entry); err != nil {
			t.Fatal(err)
		}
	}

	// Run GC.
	gcQ := newGCQueue(tc.store, tc.gossip)
	cfg, ok := tc.gossip.GetSystemConfig()
	if !ok {
		t.Fatal("config not set")
	}

	if err := gcQ.process(context.Background(), tc.clock.Now(), tc.rng, cfg); err != nil {
		t.Fatal(err)
	}

	util.SucceedsSoon(t, func() error {
		for strKey, sp := range testCases {
			txn := &roachpb.Transaction{}
			key := keys.TransactionKey(roachpb.Key(strKey), txns[strKey].ID)
			ok, err := engine.MVCCGetProto(context.Background(), tc.engine, key, hlc.ZeroTimestamp, true, nil, txn)
			if err != nil {
				return err
			}
			if expGC := (sp.newStatus == -1); expGC {
				if expGC != !ok {
					return fmt.Errorf("%s: expected gc: %t, but found %s\n%s", strKey, expGC, txn, roachpb.Key(strKey))
				}
			} else if sp.newStatus != txn.Status {
				return fmt.Errorf("%s: expected status %s, but found %s", strKey, sp.newStatus, txn.Status)
			}
			var expIntents []roachpb.Span
			if sp.expResolve {
				expIntents = testIntents
			}
			if !reflect.DeepEqual(resolved[strKey], expIntents) {
				return fmt.Errorf("%s: unexpected intent resolutions:\nexpected: %s\nobserved: %s",
					strKey, expIntents, resolved[strKey])
			}
			entry := &roachpb.AbortCacheEntry{}
			abortExists, err := tc.rng.abortCache.Get(context.Background(), tc.store.Engine(), txns[strKey].ID, entry)
			if err != nil {
				t.Fatal(err)
			}
			if abortExists == sp.expAbortGC {
				return fmt.Errorf("%s: expected abort cache gc: %t, found %+v", strKey, sp.expAbortGC, entry)
			}
		}
		return nil
	})

	outsideTxnPrefix := keys.TransactionKey(outsideKey, uuid.EmptyUUID)
	outsideTxnPrefixEnd := keys.TransactionKey(outsideKey.Next(), uuid.EmptyUUID)
	var count int
	if _, err := engine.MVCCIterate(context.Background(), tc.store.Engine(), outsideTxnPrefix, outsideTxnPrefixEnd, hlc.ZeroTimestamp,
		true, nil, false, func(roachpb.KeyValue) (bool, error) {
			count++
			return false, nil
		}); err != nil {
		t.Fatal(err)
	}
	if exp := len(testCases); exp != count {
		t.Fatalf("expected the %d external transaction entries to remain untouched, "+
			"but only %d are left", exp, count)
	}

	batch := tc.engine.NewSnapshot()
	defer batch.Close()
	tc.rng.assertState(batch) // check that in-mem and on-disk state were updated

	tc.rng.mu.Lock()
	txnSpanThreshold := tc.rng.mu.state.TxnSpanGCThreshold
	tc.rng.mu.Unlock()

	// Verify that the new TxnSpanGCThreshold has reached the Replica.
	if expWT := int64(gcTxnAndAC); txnSpanThreshold.WallTime != expWT {
		t.Fatalf("expected TxnSpanGCThreshold.Walltime %d, got timestamp %s",
			expWT, txnSpanThreshold)
	}
}

// processTransactionTable scans the transaction table and updates txnMap with
// those transactions which are old and either PENDING or with intents
// registered. In the first case we want to push the transaction so that it is
// aborted, and in the second case we may have to resolve the intents success-
// fully before GCing the entry. The transaction records which can be gc'ed are
// returned separately and are not added to txnMap nor intentSpanMap.
func processTransactionTable(
	ctx context.Context,
	snap engine.Reader,
	desc *roachpb.RangeDescriptor,
	txnMap map[uuid.UUID]*roachpb.Transaction,
	cutoff hlc.Timestamp,
	infoMu *lockableGCInfo,
	resolveIntents resolveFunc,
) ([]roachpb.GCRequest_GCKey, error) {
	infoMu.Lock()
	defer infoMu.Unlock()

	var gcKeys []roachpb.GCRequest_GCKey
	handleOne := func(kv roachpb.KeyValue) error {
		var txn roachpb.Transaction
		if err := kv.Value.GetProto(&txn); err != nil {
			return err
		}
		infoMu.TransactionSpanTotal++
		if !txn.LastActive().Less(cutoff) {
			return nil
		}

		txnID := *txn.ID

		// The transaction record should be considered for removal.
		switch txn.Status {
		case roachpb.PENDING:
			// Marked as running, so we need to push it to abort it but won't
			// try to GC it in this cycle (for convenience).
			// TODO(tschottdorf): refactor so that we can GC PENDING entries
			// in the same cycle, but keeping the calls to pushTxn in a central
			// location (keeping it easy to batch them up in the future).
			infoMu.TransactionSpanGCPending++
			txnMap[txnID] = &txn
			return nil
		case roachpb.ABORTED:
			// If we remove this transaction, it effectively still counts as
			// ABORTED (by design). So this can be GC'ed even if we can't
			// resolve the intents.
			// Note: Most aborted transaction weren't aborted by their client,
			// but instead by the coordinator - those will not have any intents
			// persisted, though they still might exist in the system.
			infoMu.TransactionSpanGCAborted++
			func() {
				infoMu.Unlock() // intentional
				defer infoMu.Lock()
				if err := resolveIntents(roachpb.AsIntents(txn.Intents, &txn),
					true /* wait */, false /* !poison */); err != nil {
					log.Warningf(ctx, "failed to resolve intents of aborted txn on gc: %s", err)
				}
			}()
		case roachpb.COMMITTED:
			// It's committed, so it doesn't need a push but we can only
			// GC it after its intents are resolved.
			if err := func() error {
				infoMu.Unlock() // intentional
				defer infoMu.Lock()
				return resolveIntents(roachpb.AsIntents(txn.Intents, &txn), true /* wait */, false /* !poison */)
			}(); err != nil {
				log.Warningf(ctx, "unable to resolve intents of committed txn on gc: %s", err)
				// Returning the error here would abort the whole GC run, and
				// we don't want that. Instead, we simply don't GC this entry.
				return nil
			}
			infoMu.TransactionSpanGCCommitted++
		default:
			panic(fmt.Sprintf("invalid transaction state: %s", txn))
		}
		gcKeys = append(gcKeys, roachpb.GCRequest_GCKey{Key: kv.Key}) // zero timestamp
		return nil
	}

	startKey := keys.TransactionKey(desc.StartKey.AsRawKey(), uuid.UUID{})
	endKey := keys.TransactionKey(desc.EndKey.AsRawKey(), uuid.UUID{})

	_, err := engine.MVCCIterate(ctx, snap, startKey, endKey,
		hlc.ZeroTimestamp, true /* consistent */, nil, /* txn */
		false /* !reverse */, func(kv roachpb.KeyValue) (bool, error) {
			return false, handleOne(kv)
		})
	return gcKeys, err
}

// TestStoreRangeMergeMetadataCleanup tests that all metadata of a
// subsumed range is cleaned up on merge.
func TestStoreRangeMergeMetadataCleanup(t *testing.T) {
	defer leaktest.AfterTest(t)()
	storeCfg := storage.TestStoreConfig(nil)
	storeCfg.TestingKnobs.DisableSplitQueue = true
	store, stopper := createTestStoreWithConfig(t, storeCfg)
	defer stopper.Stop()

	scan := func(f func(roachpb.KeyValue) (bool, error)) {
		if _, err := engine.MVCCIterate(context.Background(), store.Engine(), roachpb.KeyMin, roachpb.KeyMax, hlc.ZeroTimestamp, true, nil, false, f); err != nil {
			t.Fatal(err)
		}
	}
	content := roachpb.Key("testing!")

	// Write some values left of the proposed split key.
	pArgs := putArgs([]byte("aaa"), content)
	if _, err := client.SendWrapped(context.Background(), rg1(store), &pArgs); err != nil {
		t.Fatal(err)
	}

	// Collect all the keys.
	preKeys := make(map[string]struct{})
	scan(func(kv roachpb.KeyValue) (bool, error) {
		preKeys[string(kv.Key)] = struct{}{}
		return false, nil
	})

	// Split the range.
	_, bDesc, err := createSplitRanges(store)
	if err != nil {
		t.Fatal(err)
	}

	// Write some values right of the split key.
	pArgs = putArgs([]byte("ccc"), content)
	if _, err := client.SendWrappedWith(context.Background(), rg1(store), roachpb.Header{
		RangeID: bDesc.RangeID,
	}, &pArgs); err != nil {
		t.Fatal(err)
	}

	// Merge the b range back into the a range.
	args := adminMergeArgs(roachpb.KeyMin)
	if _, err := client.SendWrapped(context.Background(), rg1(store), &args); err != nil {
		t.Fatal(err)
	}

	// Collect all the keys again.
	postKeys := make(map[string]struct{})
	scan(func(kv roachpb.KeyValue) (bool, error) {
		postKeys[string(kv.Key)] = struct{}{}
		return false, nil
	})

	// Compute the new keys.
	for k := range preKeys {
		delete(postKeys, k)
	}

	// Keep only the subsumed range's local keys.
	localRangeKeyPrefix := string(keys.MakeRangeIDPrefix(bDesc.RangeID))
	for k := range postKeys {
		if !strings.HasPrefix(k, localRangeKeyPrefix) {
			delete(postKeys, k)
		}
	}

	if numKeys := len(postKeys); numKeys > 0 {
		var buf bytes.Buffer
		fmt.Fprintf(&buf, "%d keys were not cleaned up:\n", numKeys)
		for k := range postKeys {
			fmt.Fprintf(&buf, "%q\n", k)
		}
		t.Fatal(buf.String())
	}
}

func runDebugCheckStoreCmd(cmd *cobra.Command, args []string) error {
	stopper := stop.NewStopper()
	defer stopper.Stop()

	if len(args) != 1 {
		return errors.New("one required argument: dir")
	}

	db, err := openStore(cmd, args[0], stopper)
	if err != nil {
		return err
	}

	// Iterate over the entire range-id-local space.
	start := roachpb.Key(keys.LocalRangeIDPrefix)
	end := start.PrefixEnd()

	replicaInfo := map[roachpb.RangeID]*replicaCheckInfo{}
	getReplicaInfo := func(rangeID roachpb.RangeID) *replicaCheckInfo {
		if info, ok := replicaInfo[rangeID]; ok {
			return info
		}
		replicaInfo[rangeID] = &replicaCheckInfo{}
		return replicaInfo[rangeID]
	}

	if _, err := engine.MVCCIterate(context.Background(), db, start, end, hlc.MaxTimestamp,
		false /* !consistent */, nil, /* txn */
		false /* !reverse */, func(kv roachpb.KeyValue) (bool, error) {
			rangeID, _, suffix, detail, err := keys.DecodeRangeIDKey(kv.Key)
			if err != nil {
				return false, err
			}

			switch {
			case bytes.Equal(suffix, keys.LocalRaftTruncatedStateSuffix):
				var trunc roachpb.RaftTruncatedState
				if err := kv.Value.GetProto(&trunc); err != nil {
					return false, err
				}
				getReplicaInfo(rangeID).truncatedIndex = trunc.Index
			case bytes.Equal(suffix, keys.LocalRaftAppliedIndexSuffix):
				idx, err := kv.Value.GetInt()
				if err != nil {
					return false, err
				}
				getReplicaInfo(rangeID).appliedIndex = uint64(idx)
			case bytes.Equal(suffix, keys.LocalRaftLogSuffix):
				_, index, err := encoding.DecodeUint64Ascending(detail)
				if err != nil {
					return false, err
				}
				ri := getReplicaInfo(rangeID)
				if ri.firstIndex == 0 {
					ri.firstIndex = index
					ri.lastIndex = index
				} else {
					if index != ri.lastIndex+1 {
						fmt.Printf("range %s: log index anomaly: %v followed by %v\n",
							rangeID, ri.lastIndex, index)
					}
					ri.lastIndex = index
				}
			}

			return false, nil
		}); err != nil {
		return err
	}

	for rangeID, info := range replicaInfo {
		if info.truncatedIndex != info.firstIndex-1 {
			fmt.Printf("range %s: truncated index %v should equal first index %v - 1\n",
				rangeID, info.truncatedIndex, info.firstIndex)
		}
		if info.appliedIndex < info.firstIndex || info.appliedIndex > info.lastIndex {
			fmt.Printf("range %s: applied index %v should be between first index %v and last index %v\n",
				rangeID, info.appliedIndex, info.firstIndex, info.lastIndex)
		}
	}

	return nil
}

func runDebugGCCmd(cmd *cobra.Command, args []string) error {
	stopper := stop.NewStopper()
	defer stopper.Stop()

	if len(args) != 1 {
		return errors.New("one argument required: dir")
	}

	var rangeID roachpb.RangeID
	if len(args) == 2 {
		var err error
		if rangeID, err = parseRangeID(args[1]); err != nil {
			return err
		}
	}

	db, err := openStore(cmd, args[0], stopper)
	if err != nil {
		return err
	}

	start := keys.RangeDescriptorKey(roachpb.RKeyMin)
	end := keys.RangeDescriptorKey(roachpb.RKeyMax)

	var descs []roachpb.RangeDescriptor

	if _, err := engine.MVCCIterate(context.Background(), db, start, end, hlc.MaxTimestamp,
		false /* !consistent */, nil, /* txn */
		false /* !reverse */, func(kv roachpb.KeyValue) (bool, error) {
			var desc roachpb.RangeDescriptor
			_, suffix, _, err := keys.DecodeRangeKey(kv.Key)
			if err != nil {
				return false, err
			}
			if !bytes.Equal(suffix, keys.LocalRangeDescriptorSuffix) {
				return false, nil
			}
			if err := kv.Value.GetProto(&desc); err != nil {
				return false, err
			}
			if desc.RangeID == rangeID || rangeID == 0 {
				descs = append(descs, desc)
			}
			return desc.RangeID == rangeID, nil
		}); err != nil {
		return err
	}

	if len(descs) == 0 {
		return fmt.Errorf("no range matching the criteria found")
	}

	for _, desc := range descs {
		snap := db.NewSnapshot()
		defer snap.Close()
		_, info, err := storage.RunGC(context.Background(), &desc, snap, hlc.Timestamp{WallTime: timeutil.Now().UnixNano()},
			config.GCPolicy{TTLSeconds: 24 * 60 * 60 /* 1 day */}, func(_ hlc.Timestamp, _ *roachpb.Transaction, _ roachpb.PushTxnType) {
			}, func(_ []roachpb.Intent, _, _ bool) error { return nil })
		if err != nil {
			return err
		}
		fmt.Printf("RangeID: %d [%s, %s):\n", desc.RangeID, desc.StartKey, desc.EndKey)
		_, _ = pretty.Println(info)
	}
	return nil
}