func TestObjectIDForKey(t *testing.T) {
defer leaktest.AfterTest(t)
testCases := []struct {
key roachpb.RKey
success bool
id uint32
}{
// Before the structured span.
{roachpb.RKeyMin, false, 0},
// Boundaries of structured span.
{roachpb.RKeyMax, false, 0},
// Valid, even if there are things after the ID.
{keys.MakeKey(keys.MakeTablePrefix(42), roachpb.RKey("\xff")), true, 42},
{keys.MakeTablePrefix(0), true, 0},
{keys.MakeTablePrefix(999), true, 999},
}
for tcNum, tc := range testCases {
id, success := config.ObjectIDForKey(tc.key)
if success != tc.success {
t.Errorf("#%d: expected success=%t", tcNum, tc.success)
continue
}
if id != tc.id {
t.Errorf("#%d: expected id=%d, got %d", tcNum, tc.id, id)
}
}
}
// TestStoreZoneUpdateAndRangeSplit verifies that modifying the zone
// configuration changes range max bytes and Range.maybeSplit() takes
// max bytes into account when deciding whether to enqueue a range for
// splitting. It further verifies that the range is in fact split on
// exceeding zone's RangeMaxBytes.
func TestStoreZoneUpdateAndRangeSplit(t *testing.T) {
defer leaktest.AfterTest(t)
t.Skip("#3762")
store, stopper := createTestStore(t)
config.TestingSetupZoneConfigHook(stopper)
defer stopper.Stop()
maxBytes := int64(1 << 16)
// Set max bytes.
descID := uint32(keys.MaxReservedDescID + 1)
config.TestingSetZoneConfig(descID, &config.ZoneConfig{RangeMaxBytes: maxBytes})
// Trigger gossip callback.
if err := store.Gossip().AddInfoProto(gossip.KeySystemConfig, &config.SystemConfig{}, 0); err != nil {
t.Fatal(err)
}
// Wait for the range to be split along table boundaries.
originalRange := store.LookupReplica(roachpb.RKey(keys.UserTableDataMin), nil)
var rng *storage.Replica
if err := util.IsTrueWithin(func() bool {
rng = store.LookupReplica(keys.MakeTablePrefix(descID), nil)
return rng.RangeID != originalRange.RangeID
}, splitTimeout); err != nil {
t.Fatalf("failed to notice range max bytes update: %s", err)
}
// Check range's max bytes settings.
if rng.GetMaxBytes() != maxBytes {
t.Fatalf("range max bytes mismatch, got: %d, expected: %d", rng.GetMaxBytes(), maxBytes)
}
// Make sure the second range goes to the end.
if !roachpb.RKeyMax.Equal(rng.Desc().EndKey) {
t.Fatalf("second range has split: %+v", rng.Desc())
}
// Look in the range after prefix we're writing to.
fillRange(store, rng.RangeID, keys.MakeTablePrefix(descID), maxBytes, t)
// Verify that the range is in fact split (give it a few seconds for very
// slow test machines).
var newRng *storage.Replica
util.SucceedsWithin(t, splitTimeout, func() error {
newRng = store.LookupReplica(keys.MakeTablePrefix(descID+1), nil)
if newRng.RangeID == rng.RangeID {
return util.Errorf("range has not yet split")
}
return nil
})
// Make sure the new range goes to the end.
if !roachpb.RKeyMax.Equal(newRng.Desc().EndKey) {
t.Fatalf("second range has split: %+v", rng.Desc())
}
}
// TestStoreZoneUpdateAndRangeSplit verifies that modifying the zone
// configuration changes range max bytes and Range.maybeSplit() takes
// max bytes into account when deciding whether to enqueue a range for
// splitting. It further verifies that the range is in fact split on
// exceeding zone's RangeMaxBytes.
func TestStoreZoneUpdateAndRangeSplit(t *testing.T) {
defer leaktest.AfterTest(t)
store, stopper := createTestStore(t)
config.TestingSetupZoneConfigHook(stopper)
defer stopper.Stop()
maxBytes := int64(1 << 16)
// Set max bytes.
config.TestingSetZoneConfig(1000, &config.ZoneConfig{RangeMaxBytes: maxBytes})
// Trigger gossip callback.
if err := store.Gossip().AddInfoProto(gossip.KeySystemConfig, &config.SystemConfig{}, 0); err != nil {
t.Fatal(err)
}
// Wait for the range to be split along table boundaries.
originalRange := store.LookupReplica(roachpb.RKeyMin, nil)
var rng *storage.Replica
if err := util.IsTrueWithin(func() bool {
rng = store.LookupReplica(keys.MakeTablePrefix(1000), nil)
return rng.Desc().RangeID != originalRange.Desc().RangeID
}, 50*time.Millisecond); err != nil {
t.Fatalf("failed to notice range max bytes update: %s", err)
}
// Check range's max bytes settings.
if rng.GetMaxBytes() != maxBytes {
t.Fatalf("range max bytes mismatch, got: %d, expected: %d", rng.GetMaxBytes(), maxBytes)
}
// Make sure the second range goes to the end.
if !roachpb.RKeyMax.Equal(rng.Desc().EndKey) {
t.Fatalf("second range has split: %+v", rng.Desc())
}
// Look in the range after prefix we're writing to.
fillRange(store, rng.Desc().RangeID, keys.MakeTablePrefix(1000), maxBytes, t)
// Verify that the range is in fact split (give it a second for very slow test machines).
var newRng *storage.Replica
if err := util.IsTrueWithin(func() bool {
newRng = store.LookupReplica(keys.MakeTablePrefix(2000), nil)
return newRng.Desc().RangeID != rng.Desc().RangeID
}, time.Second); err != nil {
t.Errorf("expected range to split within 1s")
}
// Make sure the new range goes to the end.
if !roachpb.RKeyMax.Equal(newRng.Desc().EndKey) {
t.Fatalf("second range has split: %+v", rng.Desc())
}
}
// TestStoreZoneUpdateAndRangeSplit verifies that modifying the zone
// configuration changes range max bytes and Range.maybeSplit() takes
// max bytes into account when deciding whether to enqueue a range for
// splitting. It further verifies that the range is in fact split on
// exceeding zone's RangeMaxBytes.
func TestStoreZoneUpdateAndRangeSplit(t *testing.T) {
defer leaktest.AfterTest(t)()
store, stopper, _ := createTestStore(t)
config.TestingSetupZoneConfigHook(stopper)
defer stopper.Stop()
maxBytes := int64(1 << 16)
// Set max bytes.
descID := uint32(keys.MaxReservedDescID + 1)
config.TestingSetZoneConfig(descID, &config.ZoneConfig{RangeMaxBytes: maxBytes})
// Trigger gossip callback.
if err := store.Gossip().AddInfoProto(gossip.KeySystemConfig, &config.SystemConfig{}, 0); err != nil {
t.Fatal(err)
}
tableBoundary := keys.MakeTablePrefix(descID)
{
var rng *storage.Replica
// Wait for the range to be split along table boundaries.
expectedRSpan := roachpb.RSpan{Key: roachpb.RKey(tableBoundary), EndKey: roachpb.RKeyMax}
util.SucceedsSoon(t, func() error {
rng = store.LookupReplica(tableBoundary, nil)
if actualRSpan := rng.Desc().RSpan(); !actualRSpan.Equal(expectedRSpan) {
return util.Errorf("expected range %s to span %s", rng, expectedRSpan)
}
return nil
})
// Check range's max bytes settings.
if actualMaxBytes := rng.GetMaxBytes(); actualMaxBytes != maxBytes {
t.Fatalf("range %s max bytes mismatch, got: %d, expected: %d", rng, actualMaxBytes, maxBytes)
}
// Look in the range after prefix we're writing to.
fillRange(store, rng.RangeID, tableBoundary, maxBytes, t)
}
// Verify that the range is in fact split.
util.SucceedsSoon(t, func() error {
rng := store.LookupReplica(keys.MakeTablePrefix(descID+1), nil)
rngDesc := rng.Desc()
rngStart, rngEnd := rngDesc.StartKey, rngDesc.EndKey
if rngStart.Equal(tableBoundary) || !rngEnd.Equal(roachpb.RKeyMax) {
return util.Errorf("range %s has not yet split", rng)
}
return nil
})
}
// GetLargestObjectID returns the largest object ID found in the config.
// This could be either a table or a database.
func (s *SystemConfig) GetLargestObjectID() (uint32, error) {
testingLock.Lock()
hook := TestingLargestIDHook
testingLock.Unlock()
if hook != nil {
return hook(), nil
}
if len(s.Values) == 0 {
return 0, fmt.Errorf("empty system values in config")
}
// Search for the first key after the descriptor table.
// We can't use GetValue as we don't mind if there is nothing after
// the descriptor table.
key := proto.Key(keys.MakeTablePrefix(keys.DescriptorTableID + 1))
index := sort.Search(len(s.Values), func(i int) bool {
return !s.Values[i].Key.Less(key)
})
if index == 0 {
return 0, fmt.Errorf("descriptor table not found in system config of %d values", len(s.Values))
}
// This is the last key of the descriptor table.
key = s.Values[index-1].Key
// Extract object ID from key.
// TODO(marc): move sql/keys.go to keys (or similar) and use a DecodeDescMetadataKey.
// We should also check proper encoding.
descriptorPrefix := keys.MakeTablePrefix(keys.DescriptorTableID)
remaining := bytes.TrimPrefix(key, descriptorPrefix)
// TrimPrefix returns the bytes unchanged if the prefix does not match.
if len(remaining) == len(key) {
return 0, fmt.Errorf("descriptor table not found in system config of %d values", len(s.Values))
}
// DescriptorTable.PrimaryIndex.ID
remaining, _, err := encoding.DecodeUvarint(remaining)
if err != nil {
return 0, err
}
// descID
_, id, err := encoding.DecodeUvarint(remaining)
if err != nil {
return 0, err
}
return uint32(id), nil
}
func sqlKV(tableID uint32, indexID, descriptorID uint64) roachpb.KeyValue {
k := keys.MakeTablePrefix(tableID)
k = encoding.EncodeUvarintAscending(k, indexID)
k = encoding.EncodeUvarintAscending(k, descriptorID)
k = encoding.EncodeUvarintAscending(k, 12345) // Column ID, but could be anything.
return kv(k, nil)
}
// RefreshLeases starts a goroutine that refreshes the lease manager
// leases for tables received in the latest system configuration via gossip.
func (m *LeaseManager) RefreshLeases(s *stop.Stopper, db *client.DB, gossip *gossip.Gossip) {
s.RunWorker(func() {
descKeyPrefix := keys.MakeTablePrefix(uint32(sqlbase.DescriptorTable.ID))
gossipUpdateC := gossip.RegisterSystemConfigChannel()
for {
select {
case <-gossipUpdateC:
cfg, _ := gossip.GetSystemConfig()
if m.testingKnobs.GossipUpdateEvent != nil {
m.testingKnobs.GossipUpdateEvent(cfg)
}
// Read all tables and their versions
if log.V(2) {
log.Info("received a new config; will refresh leases")
}
// Loop through the configuration to find all the tables.
for _, kv := range cfg.Values {
if !bytes.HasPrefix(kv.Key, descKeyPrefix) {
continue
}
// Attempt to unmarshal config into a table/database descriptor.
var descriptor sqlbase.Descriptor
if err := kv.Value.GetProto(&descriptor); err != nil {
log.Warningf("%s: unable to unmarshal descriptor %v", kv.Key, kv.Value)
continue
}
switch union := descriptor.Union.(type) {
case *sqlbase.Descriptor_Table:
table := union.Table
if err := table.Validate(); err != nil {
log.Errorf("%s: received invalid table descriptor: %v", kv.Key, table)
continue
}
if log.V(2) {
log.Infof("%s: refreshing lease table: %d (%s), version: %d",
kv.Key, table.ID, table.Name, table.Version)
}
// Try to refresh the table lease to one >= this version.
if t := m.findTableState(table.ID, false /* create */, nil); t != nil {
if err := t.purgeOldLeases(
db, table.Deleted(), table.Version, m.LeaseStore); err != nil {
log.Warningf("error purging leases for table %d(%s): %s",
table.ID, table.Name, err)
}
}
case *sqlbase.Descriptor_Database:
// Ignore.
}
}
if m.testingKnobs.TestingLeasesRefreshedEvent != nil {
m.testingKnobs.TestingLeasesRefreshedEvent(cfg)
}
case <-s.ShouldStop():
return
}
}
})
}
// Truncate deletes all rows from a table.
// Privileges: DROP on table.
// Notes: postgres requires TRUNCATE.
// mysql requires DROP (for mysql >= 5.1.16, DELETE before that).
func (p *planner) Truncate(n *parser.Truncate) (planNode, *roachpb.Error) {
b := client.Batch{}
for _, tableQualifiedName := range n.Tables {
tableDesc, pErr := p.getTableLease(tableQualifiedName)
if pErr != nil {
return nil, pErr
}
if err := p.checkPrivilege(&tableDesc, privilege.DROP); err != nil {
return nil, roachpb.NewError(err)
}
tablePrefix := keys.MakeTablePrefix(uint32(tableDesc.ID))
// Delete rows and indexes starting with the table's prefix.
tableStartKey := roachpb.Key(tablePrefix)
tableEndKey := tableStartKey.PrefixEnd()
if log.V(2) {
log.Infof("DelRange %s - %s", tableStartKey, tableEndKey)
}
b.DelRange(tableStartKey, tableEndKey, false)
}
if pErr := p.txn.Run(&b); pErr != nil {
return nil, pErr
}
return &emptyNode{}, nil
}
func (z *zeroSum) monkey(tableID uint32, d time.Duration) {
r := newRand()
zipf := z.accountDistribution(r)
for {
time.Sleep(time.Duration(rand.Float64() * float64(d)))
key := keys.MakeTablePrefix(tableID)
key = encoding.EncodeVarintAscending(key, int64(zipf.Uint64()))
key = keys.MakeRowSentinelKey(key)
switch r.Intn(2) {
case 0:
if err := z.split(z.randNode(r.Intn), key); err != nil {
if strings.Contains(err.Error(), "range is already split at key") ||
strings.Contains(err.Error(), "conflict updating range descriptors") {
continue
}
z.maybeLogError(err)
} else {
atomic.AddUint64(&z.stats.splits, 1)
}
case 1:
if transferred, err := z.transferLease(z.randNode(r.Intn), r, key); err != nil {
z.maybeLogError(err)
} else if transferred {
atomic.AddUint64(&z.stats.transfers, 1)
}
}
}
}
// TestStoreRangeSplitWithMaxBytesUpdate tests a scenario where a new
// zone config that updates the max bytes is set and triggers a range
// split.
func TestStoreRangeSplitWithMaxBytesUpdate(t *testing.T) {
defer leaktest.AfterTest(t)
store, stopper := createTestStore(t)
config.TestingSetupZoneConfigHook(stopper)
defer stopper.Stop()
origRng := store.LookupReplica(roachpb.RKeyMin, nil)
// Set max bytes.
maxBytes := int64(1 << 16)
config.TestingSetZoneConfig(1000, &config.ZoneConfig{RangeMaxBytes: maxBytes})
// Trigger gossip callback.
if err := store.Gossip().AddInfoProto(gossip.KeySystemConfig, &config.SystemConfig{}, 0); err != nil {
t.Fatal(err)
}
// Verify that the range is split and the new range has the correct max bytes.
util.SucceedsWithin(t, time.Second, func() error {
newRng := store.LookupReplica(keys.MakeTablePrefix(1000), nil)
if newRng.Desc().RangeID == origRng.Desc().RangeID {
return util.Errorf("expected new range created by split")
}
if newRng.GetMaxBytes() != maxBytes {
return util.Errorf("expected %d max bytes for the new range, but got %d",
maxBytes, newRng.GetMaxBytes())
}
return nil
})
}
// MakeZoneKey returns the key for 'id's entry in the system.zones table.
func MakeZoneKey(id ID) roachpb.Key {
k := keys.MakeTablePrefix(uint32(ZonesTable.ID))
k = encoding.EncodeUvarint(k, uint64(ZonesTable.PrimaryIndex.ID))
k = encoding.EncodeUvarint(k, uint64(id))
k = encoding.EncodeUvarint(k, uint64(ZonesTable.Columns[1].ID))
return k
}
// MakeDescMetadataKey returns the key for the descriptor.
func MakeDescMetadataKey(descID ID) roachpb.Key {
k := keys.MakeTablePrefix(uint32(DescriptorTable.ID))
k = encoding.EncodeUvarint(k, uint64(DescriptorTable.PrimaryIndex.ID))
k = encoding.EncodeUvarint(k, uint64(descID))
k = encoding.EncodeUvarint(k, uint64(DescriptorTable.Columns[1].ID))
return k
}
// Truncate deletes all rows from a table.
// Privileges: DROP on table.
// Notes: postgres requires TRUNCATE.
// mysql requires DROP (for mysql >= 5.1.16, DELETE before that).
func (p *planner) Truncate(n *parser.Truncate) (planNode, error) {
b := client.Batch{}
for _, tableQualifiedName := range n.Tables {
tableDesc, err := p.getTableDesc(tableQualifiedName)
if err != nil {
return nil, err
}
if err := p.checkPrivilege(tableDesc, privilege.DROP); err != nil {
return nil, err
}
tablePrefix := keys.MakeTablePrefix(uint32(tableDesc.ID))
// Delete rows and indexes starting with the table's prefix.
tableStartKey := proto.Key(tablePrefix)
tableEndKey := tableStartKey.PrefixEnd()
if log.V(2) {
log.Infof("DelRange %s - %s", prettyKey(tableStartKey, 0), prettyKey(tableEndKey, 0))
}
b.DelRange(tableStartKey, tableEndKey)
}
if err := p.txn.Run(&b); err != nil {
return nil, err
}
return &valuesNode{}, nil
}
// ComputeSplitKeys takes a start and end key and returns an array of keys
// at which to split the span [start, end).
// The only required splits are at each user table prefix.
func (s SystemConfig) ComputeSplitKeys(startKey, endKey roachpb.RKey) []roachpb.RKey {
testingLock.Lock()
tableSplitsDisabled := testingDisableTableSplits
testingLock.Unlock()
if tableSplitsDisabled {
return nil
}
tableStart := roachpb.RKey(keys.UserTableDataMin)
if !tableStart.Less(endKey) {
// This range is before the user tables span: no required splits.
return nil
}
startID, ok := ObjectIDForKey(startKey)
if !ok || startID <= keys.MaxReservedDescID {
// The start key is either:
// - not part of the structured data span
// - part of the system span
// In either case, start looking for splits at the first ID usable
// by the user data span.
startID = keys.MaxReservedDescID + 1
} else {
// The start key is either already a split key, or after the split
// key for its ID. We can skip straight to the next one.
startID++
}
// Find the largest object ID.
// We can't keep splitting until we reach endKey as it could be roachpb.KeyMax.
endID, err := s.GetLargestObjectID()
if err != nil {
log.Errorf("unable to determine largest object ID from system config: %s", err)
return nil
}
// Build key prefixes for sequential table IDs until we reach endKey.
var splitKeys []roachpb.RKey
var key roachpb.RKey
// endID could be smaller than startID if we don't have user tables.
for id := startID; id <= endID; id++ {
key = keys.MakeTablePrefix(id)
// Skip if the range starts on a split key.
if !startKey.Less(key) {
continue
}
// Handle the case where EndKey is already a table prefix.
if !key.Less(endKey) {
break
}
splitKeys = append(splitKeys, key)
}
return splitKeys
}
// RefreshLeases starts a goroutine that refreshes the lease manager
// leases for tables received in the latest system configuration via gossip.
func (m *LeaseManager) RefreshLeases(s *stop.Stopper, db *client.DB, gossip *gossip.Gossip) {
s.RunWorker(func() {
descKeyPrefix := keys.MakeTablePrefix(uint32(DescriptorTable.ID))
gossip.RegisterSystemConfigCallback(m.updateSystemConfig)
for {
select {
case <-m.newConfig:
// Read all tables and their versions
cfg := m.getSystemConfig()
if log.V(2) {
log.Info("received a new config %v", cfg)
}
// Loop through the configuration to find all the tables.
for _, kv := range cfg.Values {
if kv.Value.Tag != roachpb.ValueType_BYTES {
continue
}
if !bytes.HasPrefix(kv.Key, descKeyPrefix) {
continue
}
// Attempt to unmarshal config into a table/database descriptor.
var descriptor Descriptor
if err := kv.Value.GetProto(&descriptor); err != nil {
log.Warningf("unable to unmarshal descriptor %v", kv.Value)
continue
}
switch union := descriptor.Union.(type) {
case *Descriptor_Table:
table := union.Table
if err := table.Validate(); err != nil {
log.Errorf("received invalid table descriptor: %v", table)
continue
}
if log.V(2) {
log.Infof("refreshing lease table: %d, version: %d", table.ID, table.Version)
}
// Try to refresh the table lease to one >= this version.
if err := m.refreshLease(db, table.ID, table.Version); err != nil {
log.Warning(err)
}
case *Descriptor_Database:
// Ignore.
}
}
case <-s.ShouldStop():
return
}
}
})
}
// MakeNameMetadataKey returns the key for the name. Pass name == "" in order
// to generate the prefix key to use to scan over all of the names for the
// specified parentID.
func MakeNameMetadataKey(parentID ID, name string) proto.Key {
name = normalizeName(name)
k := keys.MakeTablePrefix(uint32(NamespaceTable.ID))
k = encoding.EncodeUvarint(k, uint64(NamespaceTable.PrimaryIndex.ID))
k = encoding.EncodeUvarint(k, uint64(parentID))
if name != "" {
k = encoding.EncodeBytes(k, []byte(name))
k = encoding.EncodeUvarint(k, uint64(NamespaceTable.Columns[2].ID))
}
return k
}
// checkTableSize checks that the number of key:value pairs stored
// in the table equals e.
func (mt mutationTest) checkTableSize(e int) {
// Check that there are no hidden values
tablePrefix := keys.MakeTablePrefix(uint32(mt.tableDesc.ID))
tableStartKey := roachpb.Key(tablePrefix)
tableEndKey := tableStartKey.PrefixEnd()
if kvs, err := mt.kvDB.Scan(tableStartKey, tableEndKey, 0); err != nil {
mt.Error(err)
} else if len(kvs) != e {
mt.Errorf("expected %d key value pairs, but got %d", e, len(kvs))
}
}
// MakeNameMetadataKey returns the key for the name. Pass name == "" in order
// to generate the prefix key to use to scan over all of the names for the
// specified parentID.
func MakeNameMetadataKey(parentID ID, name string) roachpb.Key {
name = NormalizeName(name)
k := keys.MakeTablePrefix(uint32(namespaceTable.ID))
k = encoding.EncodeUvarintAscending(k, uint64(namespaceTable.PrimaryIndex.ID))
k = encoding.EncodeUvarintAscending(k, uint64(parentID))
if name != "" {
k = encoding.EncodeBytesAscending(k, []byte(name))
k = keys.MakeColumnKey(k, uint32(namespaceTable.Columns[2].ID))
}
return k
}
// TestStoreSetRangesMaxBytes creates a set of ranges via splitting
// and then sets the config zone to a custom max bytes value to
// verify the ranges' max bytes are updated appropriately.
func TestStoreSetRangesMaxBytes(t *testing.T) {
defer leaktest.AfterTest(t)
store, _, stopper := createTestStore(t)
defer stopper.Stop()
testData := []struct {
rng *Replica
expMaxBytes int64
}{
{store.LookupReplica(roachpb.KeyMin, nil),
config.DefaultZoneConfig.RangeMaxBytes},
{splitTestRange(store, roachpb.KeyMin, keys.MakeTablePrefix(1000), t),
1 << 20},
{splitTestRange(store, keys.MakeTablePrefix(1000), keys.MakeTablePrefix(1001), t),
config.DefaultZoneConfig.RangeMaxBytes},
{splitTestRange(store, keys.MakeTablePrefix(1001), keys.MakeTablePrefix(1002), t),
2 << 20},
}
// Set zone configs.
config.TestingSetZoneConfig(1000, &config.ZoneConfig{RangeMaxBytes: 1 << 20})
config.TestingSetZoneConfig(1002, &config.ZoneConfig{RangeMaxBytes: 2 << 20})
// Despite faking the zone configs, we still need to have a gossip entry.
if err := store.Gossip().AddInfoProto(gossip.KeySystemConfig, &config.SystemConfig{}, 0); err != nil {
t.Fatal(err)
}
if err := util.IsTrueWithin(func() bool {
for _, test := range testData {
if test.rng.GetMaxBytes() != test.expMaxBytes {
return false
}
}
return true
}, 500*time.Millisecond); err != nil {
t.Errorf("range max bytes values did not change as expected: %s", err)
}
}
// truncateTable truncates the data of a table.
// It deletes a range of data for the table, which includes the PK and all
// indexes.
func truncateTable(tableDesc *sqlbase.TableDescriptor, txn *client.Txn) error {
tablePrefix := keys.MakeTablePrefix(uint32(tableDesc.ID))
// Delete rows and indexes starting with the table's prefix.
tableStartKey := roachpb.Key(tablePrefix)
tableEndKey := tableStartKey.PrefixEnd()
if log.V(2) {
log.Infof("DelRange %s - %s", tableStartKey, tableEndKey)
}
b := client.Batch{}
b.DelRange(tableStartKey, tableEndKey, false)
return txn.Run(&b)
}
// TestStoreRangeSplitInsideRow verifies an attempt to split a range inside of
// a table row will cause a split at a boundary between rows.
func TestStoreRangeSplitInsideRow(t *testing.T) {
defer leaktest.AfterTest(t)()
sCtx := storage.TestStoreContext()
sCtx.TestingKnobs.DisableSplitQueue = true
store, stopper, _ := createTestStoreWithContext(t, sCtx)
defer stopper.Stop()
// Manually create some the column keys corresponding to the table:
//
// CREATE TABLE t (id STRING PRIMARY KEY, col1 INT, col2 INT)
tableKey := keys.MakeTablePrefix(keys.MaxReservedDescID + 1)
rowKey := roachpb.Key(encoding.EncodeVarintAscending(append([]byte(nil), tableKey...), 1))
rowKey = encoding.EncodeStringAscending(encoding.EncodeVarintAscending(rowKey, 1), "a")
col1Key := keys.MakeFamilyKey(append([]byte(nil), rowKey...), 1)
col2Key := keys.MakeFamilyKey(append([]byte(nil), rowKey...), 2)
// We don't care about the value, so just store any old thing.
if err := store.DB().Put(col1Key, "column 1"); err != nil {
t.Fatal(err)
}
if err := store.DB().Put(col2Key, "column 2"); err != nil {
t.Fatal(err)
}
// Split between col1Key and col2Key by splitting before col2Key.
args := adminSplitArgs(col2Key, col2Key)
_, err := client.SendWrapped(rg1(store), nil, &args)
if err != nil {
t.Fatalf("%s: split unexpected error: %s", col1Key, err)
}
rng1 := store.LookupReplica(col1Key, nil)
rng2 := store.LookupReplica(col2Key, nil)
// Verify the two columns are still on the same range.
if !reflect.DeepEqual(rng1, rng2) {
t.Fatalf("%s: ranges differ: %+v vs %+v", roachpb.Key(col1Key), rng1, rng2)
}
// Verify we split on a row key.
if startKey := rng1.Desc().StartKey; !startKey.Equal(rowKey) {
t.Fatalf("%s: expected split on %s, but found %s",
roachpb.Key(col1Key), roachpb.Key(rowKey), startKey)
}
// Verify the previous range was split on a row key.
rng3 := store.LookupReplica(tableKey, nil)
if endKey := rng3.Desc().EndKey; !endKey.Equal(rowKey) {
t.Fatalf("%s: expected split on %s, but found %s",
roachpb.Key(col1Key), roachpb.Key(rowKey), endKey)
}
}
func allSQLDescriptors(txn *client.Txn) ([]sqlbase.Descriptor, error) {
startKey := roachpb.Key(keys.MakeTablePrefix(keys.DescriptorTableID))
endKey := startKey.PrefixEnd()
// TODO(dan): Iterate with some batch size.
rows, err := txn.Scan(startKey, endKey, 0)
if err != nil {
return nil, errors.Wrap(err, "unable to scan SQL descriptors")
}
sqlDescs := make([]sqlbase.Descriptor, len(rows))
for i, row := range rows {
if err := row.ValueProto(&sqlDescs[i]); err != nil {
return nil, errors.Wrapf(err, "%s: unable to unmarshal SQL descriptor", row.Key)
}
}
return sqlDescs, nil
}
// GetTableSpan gets the key span for a SQL table, including any indices.
func (ie InternalExecutor) GetTableSpan(user string, txn *client.Txn, dbName, tableName string) (roachpb.Span, error) {
// Lookup the table ID.
p := makeInternalPlanner(txn, user)
p.leaseMgr = ie.LeaseManager
tn := parser.TableName{DatabaseName: parser.Name(dbName), TableName: parser.Name(tableName)}
tableID, err := getTableID(p, &tn)
if err != nil {
return roachpb.Span{}, err
}
// Determine table data span.
tablePrefix := keys.MakeTablePrefix(uint32(tableID))
tableStartKey := roachpb.Key(tablePrefix)
tableEndKey := tableStartKey.PrefixEnd()
return roachpb.Span{Key: tableStartKey, EndKey: tableEndKey}, nil
}
// TestStoreRangeSplitAtIllegalKeys verifies a range cannot be split
// at illegal keys.
func TestStoreRangeSplitAtIllegalKeys(t *testing.T) {
defer leaktest.AfterTest(t)
store, stopper := createTestStore(t)
defer stopper.Stop()
for _, key := range []roachpb.Key{
keys.Meta1Prefix,
keys.MakeKey(keys.Meta1Prefix, []byte("a")),
keys.MakeKey(keys.Meta1Prefix, roachpb.RKeyMax),
keys.Meta2KeyMax,
keys.MakeTablePrefix(10 /* system descriptor ID */),
} {
args := adminSplitArgs(roachpb.KeyMin, key)
_, err := client.SendWrapped(rg1(store), nil, &args)
if err == nil {
t.Fatalf("%q: split succeeded unexpectedly", key)
}
}
}
// TestStoreRangeSplitAtIllegalKeys verifies a range cannot be split
// at illegal keys.
func TestStoreRangeSplitAtIllegalKeys(t *testing.T) {
defer leaktest.AfterTest(t)()
store, stopper, _ := createTestStore(t)
defer stopper.Stop()
for _, key := range []roachpb.Key{
keys.Meta1Prefix,
testutils.MakeKey(keys.Meta1Prefix, []byte("a")),
testutils.MakeKey(keys.Meta1Prefix, roachpb.RKeyMax),
keys.Meta2KeyMax,
keys.MakeTablePrefix(10 /* system descriptor ID */),
} {
args := adminSplitArgs(roachpb.KeyMin, key)
_, pErr := client.SendWrapped(rg1(store), nil, &args)
if !testutils.IsPError(pErr, "cannot split") {
t.Errorf("%q: unexpected split error %s", key, pErr)
}
}
}
// TestStoreRangeSplitAtIllegalKeys verifies a range cannot be split
// at illegal keys.
func TestStoreRangeSplitAtIllegalKeys(t *testing.T) {
defer leaktest.AfterTest(t)
store, stopper := createTestStore(t)
defer stopper.Stop()
for _, key := range []proto.Key{
keys.Meta1Prefix,
keys.MakeKey(keys.Meta1Prefix, []byte("a")),
keys.MakeKey(keys.Meta1Prefix, proto.KeyMax),
keys.Meta2KeyMax,
keys.MakeTablePrefix(10 /* system descriptor ID */),
} {
args := adminSplitArgs(proto.KeyMin, key, 1, store.StoreID())
_, err := store.ExecuteCmd(context.Background(), &args)
if err == nil {
t.Fatalf("%q: split succeeded unexpectedly", key)
}
}
}
// GetTableSpan gets the key span for a SQL table, including any indices.
func (ie InternalExecutor) GetTableSpan(user string, txn *client.Txn, dbName, tableName string) (roachpb.Span, error) {
// Lookup the table ID.
p := makePlanner()
p.setTxn(txn)
p.session.User = user
p.leaseMgr = ie.LeaseManager
qname := &parser.QualifiedName{Base: parser.Name(tableName)}
if err := qname.NormalizeTableName(dbName); err != nil {
return roachpb.Span{}, err
}
tableID, err := p.getTableID(qname)
if err != nil {
return roachpb.Span{}, err
}
// Determine table data span.
tablePrefix := keys.MakeTablePrefix(uint32(tableID))
tableStartKey := roachpb.Key(tablePrefix)
tableEndKey := tableStartKey.PrefixEnd()
return roachpb.Span{Key: tableStartKey, EndKey: tableEndKey}, nil
}
// TestSplitOnTableBoundaries verifies that ranges get split
// as new tables get created.
func TestSplitOnTableBoundaries(t *testing.T) {
defer leaktest.AfterTest(t)
s, sqlDB, kvDB := setupWithContext(t, getFastScanContext())
defer cleanup(s, sqlDB)
num, err := getNumRanges(kvDB)
if err != nil {
t.Fatalf("failed to retrieve range list: %s", err)
}
if num != 1 {
t.Fatalf("expected no splits, but found %d ranges", num)
}
if _, err := sqlDB.Exec(`CREATE DATABASE test`); err != nil {
t.Fatal(err)
}
// We split up to the largest allocated descriptor ID, be it a table
// or a database.
util.SucceedsWithin(t, rangeSplitTimeout, func() error {
num, err := getNumRanges(kvDB)
if err != nil {
return err
}
if e := 2; num != e {
return util.Errorf("expected %d splits, found %d", e, num)
}
return nil
})
// Verify the actual splits.
objectID := uint32(keys.MaxReservedDescID + 1)
splits := []roachpb.RKey{keys.MakeTablePrefix(objectID), roachpb.RKeyMax}
ranges, err := getRangeKeys(kvDB)
if err != nil {
t.Fatal(err)
}
if !rangesMatchSplits(ranges, splits) {
t.Fatalf("Found ranges: %v\nexpected: %v", ranges, splits)
}
// Let's create a table.
if _, err := sqlDB.Exec(`CREATE TABLE test.test (k INT PRIMARY KEY, v INT)`); err != nil {
t.Fatal(err)
}
util.SucceedsWithin(t, rangeSplitTimeout, func() error {
num, err := getNumRanges(kvDB)
if err != nil {
return err
}
if e := 3; num != e {
return util.Errorf("expected %d splits, found %d", e, num)
}
return nil
})
// Verify the actual splits.
splits = []roachpb.RKey{keys.MakeTablePrefix(objectID), keys.MakeTablePrefix(objectID + 1), roachpb.RKeyMax}
ranges, err = getRangeKeys(kvDB)
if err != nil {
t.Fatal(err)
}
if !rangesMatchSplits(ranges, splits) {
t.Fatalf("Found ranges: %v\nexpected: %v", ranges, splits)
}
}
// TestStoreRangeSystemSplits verifies that splits are based on the
// contents of the SystemDB span.
func TestStoreRangeSystemSplits(t *testing.T) {
defer leaktest.AfterTest(t)
store, stopper := createTestStore(t)
defer stopper.Stop()
initialSystemValues := sql.MakeMetadataSchema().GetInitialValues()
numInitialValues := len(initialSystemValues)
// Write the initial sql values to the system DB as well
// as the equivalent of table descriptors for X user tables.
// This does two things:
// - descriptor IDs are used to determine split keys
// - the write triggers a SystemConfig update and gossip.
// We should end up with splits at each user table prefix.
if err := store.DB().Txn(func(txn *client.Txn) error {
txn.SetSystemDBTrigger()
for i, kv := range initialSystemValues {
bytes, err := kv.Value.GetBytes()
if err != nil {
log.Info(err)
continue
}
if err := txn.Put(kv.Key, bytes); err != nil {
return err
}
descID := keys.MaxReservedDescID + i + 1
// We don't care about the values, just the keys.
k := sql.MakeDescMetadataKey(sql.ID(descID))
if err := txn.Put(k, bytes); err != nil {
return err
}
}
return nil
}); err != nil {
t.Fatal(err)
}
verifySplitsAtTablePrefixes := func(maxTableID int) {
// We expect splits at each of the user tables, but not at the system
// tables boundaries.
expKeys := make([]roachpb.Key, 0, maxTableID+1)
for i := 1; i <= maxTableID; i++ {
expKeys = append(expKeys,
keys.MakeKey(keys.Meta2Prefix, keys.MakeTablePrefix(keys.MaxReservedDescID+uint32(i))),
)
}
expKeys = append(expKeys, keys.MakeKey(keys.Meta2Prefix, roachpb.RKeyMax))
util.SucceedsWithinDepth(1, t, 5*time.Second, func() error {
rows, err := store.DB().Scan(keys.Meta2Prefix, keys.MetaMax, 0)
if err != nil {
return err
}
keys := make([]roachpb.Key, 0, len(expKeys))
for _, r := range rows {
keys = append(keys, r.Key)
}
if !reflect.DeepEqual(keys, expKeys) {
return util.Errorf("expected split keys:\n%v\nbut found:\n%v", expKeys, keys)
}
return nil
})
}
verifySplitsAtTablePrefixes(len(initialSystemValues))
numTotalValues := numInitialValues + 5
// Write another, disjoint descriptor for a user table.
if err := store.DB().Txn(func(txn *client.Txn) error {
txn.SetSystemDBTrigger()
// This time, only write the last table descriptor. Splits
// still occur for every intervening ID.
// We don't care about the values, just the keys.
k := sql.MakeDescMetadataKey(sql.ID(keys.MaxReservedDescID + numTotalValues))
return txn.Put(k, &sql.TableDescriptor{})
}); err != nil {
t.Fatal(err)
}
verifySplitsAtTablePrefixes(numTotalValues)
}
// TestStoreRangeSplitStats starts by splitting the system keys from user-space
// keys and verifying that the user space side of the split (which is empty),
// has all zeros for stats. It then writes random data to the user space side,
// splits it halfway and verifies the two splits have stats exactly equaling
// the pre-split.
func TestStoreRangeSplitStats(t *testing.T) {
defer leaktest.AfterTest(t)
store, stopper := createTestStore(t)
defer stopper.Stop()
// Split the range after the last table data key.
keyPrefix := keys.MakeTablePrefix(keys.MaxReservedDescID + 1)
keyPrefix = keys.MakeNonColumnKey(keyPrefix)
args := adminSplitArgs(roachpb.KeyMin, keyPrefix)
if _, err := client.SendWrapped(rg1(store), nil, &args); err != nil {
t.Fatal(err)
}
// Verify empty range has empty stats.
rng := store.LookupReplica(keyPrefix, nil)
// NOTE that this value is expected to change over time, depending on what
// we store in the sys-local keyspace. Update it accordingly for this test.
if err := verifyRangeStats(store.Engine(), rng.Desc().RangeID, engine.MVCCStats{}); err != nil {
t.Fatal(err)
}
// Write random data.
src := rand.New(rand.NewSource(0))
for i := 0; i < 100; i++ {
key := append([]byte(nil), keyPrefix...)
key = append(key, randutil.RandBytes(src, int(src.Int31n(1<<7)))...)
key = keys.MakeNonColumnKey(key)
val := randutil.RandBytes(src, int(src.Int31n(1<<8)))
pArgs := putArgs(key, val)
if _, err := client.SendWrappedWith(rg1(store), nil, roachpb.Header{
RangeID: rng.Desc().RangeID,
}, &pArgs); err != nil {
t.Fatal(err)
}
}
// Get the range stats now that we have data.
var ms engine.MVCCStats
if err := engine.MVCCGetRangeStats(store.Engine(), rng.Desc().RangeID, &ms); err != nil {
t.Fatal(err)
}
// Split the range at approximate halfway point ("Z" in string "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz").
midKey := append([]byte(nil), keyPrefix...)
midKey = append(midKey, []byte("Z")...)
midKey = keys.MakeNonColumnKey(midKey)
args = adminSplitArgs(keyPrefix, midKey)
if _, err := client.SendWrappedWith(rg1(store), nil, roachpb.Header{
RangeID: rng.Desc().RangeID,
}, &args); err != nil {
t.Fatal(err)
}
var msLeft, msRight engine.MVCCStats
if err := engine.MVCCGetRangeStats(store.Engine(), rng.Desc().RangeID, &msLeft); err != nil {
t.Fatal(err)
}
rngRight := store.LookupReplica(midKey, nil)
if err := engine.MVCCGetRangeStats(store.Engine(), rngRight.Desc().RangeID, &msRight); err != nil {
t.Fatal(err)
}
// The stats should be exactly equal when added.
expMS := engine.MVCCStats{
LiveBytes: msLeft.LiveBytes + msRight.LiveBytes,
KeyBytes: msLeft.KeyBytes + msRight.KeyBytes,
ValBytes: msLeft.ValBytes + msRight.ValBytes,
IntentBytes: msLeft.IntentBytes + msRight.IntentBytes,
LiveCount: msLeft.LiveCount + msRight.LiveCount,
KeyCount: msLeft.KeyCount + msRight.KeyCount,
ValCount: msLeft.ValCount + msRight.ValCount,
IntentCount: msLeft.IntentCount + msRight.IntentCount,
}
ms.SysBytes, ms.SysCount = 0, 0
if !reflect.DeepEqual(expMS, ms) {
t.Errorf("expected left and right ranges to equal original: %+v + %+v != %+v", msLeft, msRight, ms)
}
}
