// getCachedRangeDescriptorLocked is a helper function to retrieve the
// descriptor of the range which contains the given key, if present in the
// cache. It is assumed that the caller holds a read lock on rdc.rangeCacheMu.
func (rdc *rangeDescriptorCache) getCachedRangeDescriptorLocked(key proto.Key, inclusive bool) (
rangeCacheKey, *proto.RangeDescriptor) {
// The cache is indexed using the end-key of the range, but the
// end-key is non-inclusive by default.
var metaKey proto.Key
if !inclusive {
metaKey = keys.RangeMetaKey(key.Next())
} else {
metaKey = keys.RangeMetaKey(key)
}
k, v, ok := rdc.rangeCache.Ceil(rangeCacheKey(metaKey))
if !ok {
return nil, nil
}
metaEndKey := k.(rangeCacheKey)
rd := v.(*proto.RangeDescriptor)
// Check that key actually belongs to the range.
if !rd.ContainsKey(key) {
// The key is the EndKey and we're inclusive, so just return the range descriptor.
if inclusive && key.Equal(rd.EndKey) {
return metaEndKey, rd
}
return nil, nil
}
// The key is the StartKey, but we're inclusive and thus need to return the
// previous range descriptor, but it is not in the cache yet.
if inclusive && key.Equal(rd.StartKey) {
return nil, nil
}
return metaEndKey, rd
}
// Add the specified timestamp to the cache as covering the range of
// keys from start to end. If end is nil, the range covers the start
// key only. txnID is nil for no transaction. readOnly specifies
// whether the command adding this timestamp was read-only or not.
func (tc *TimestampCache) Add(start, end proto.Key, timestamp proto.Timestamp, txnID []byte, readOnly bool) {
// This gives us a memory-efficient end key if end is empty.
if len(end) == 0 {
end = start.Next()
start = end[:len(start)]
}
if tc.latest.Less(timestamp) {
tc.latest = timestamp
}
// Only add to the cache if the timestamp is more recent than the
// low water mark.
if tc.lowWater.Less(timestamp) {
// Check existing, overlapping entries. Remove superseded
// entries or return without adding this entry if necessary.
key := tc.cache.NewKey(start, end)
for _, o := range tc.cache.GetOverlaps(start, end) {
ce := o.Value.(cacheEntry)
if ce.readOnly != readOnly {
continue
}
if o.Key.Contains(key) && !ce.timestamp.Less(timestamp) {
return // don't add this key; there's already a cache entry with >= timestamp.
} else if key.Contains(o.Key) && !timestamp.Less(ce.timestamp) {
tc.cache.Del(o.Key) // delete existing key; this cache entry supersedes.
}
}
ce := cacheEntry{timestamp: timestamp, txnID: txnID, readOnly: readOnly}
tc.cache.Add(key, ce)
}
}
// Add adds a command to the queue which affects the specified key
// range. If end is empty, it is set to start.Next(), meaning the
// command affects a single key. The returned interface is the key for
// the command queue and must be re-supplied on subsequent invocation
// of Remove().
//
// Add should be invoked after waiting on already-executing,
// overlapping commands via the WaitGroup initialized through
// GetWait().
func (cq *CommandQueue) Add(start, end proto.Key, readOnly bool) interface{} {
if len(end) == 0 {
end = start.Next()
}
key := cq.cache.NewKey(start, end)
cq.cache.Add(key, &cmd{readOnly: readOnly})
return key
}
// MetaScanBounds returns the start and end keys of the range within which the
// desired meta record can be found by means of an engine scan. The given key
// must be a valid RangeMetaKey as defined by ValidateRangeMetaKey.
func MetaScanBounds(key proto.Key) (proto.Key, proto.Key) {
if key.Equal(proto.KeyMin) {
// Special case KeyMin: find the first entry in meta1.
return Meta1Prefix, Meta1Prefix.PrefixEnd()
}
if key.Equal(Meta1KeyMax) {
// Special case Meta1KeyMax: this is the last key in Meta1, we don't want
// to start at Next().
return key, Meta1Prefix.PrefixEnd()
}
// Otherwise find the first entry greater than the given key in the same meta prefix.
return key.Next(), proto.Key(key[:len(Meta1Prefix)]).PrefixEnd()
}
// GetWait initializes the supplied wait group with the number of
// executing commands which overlap the specified key range. If end is
// empty, end is set to start.Next(), meaning the command affects a
// single key. The caller should call wg.Wait() to wait for
// confirmation that all gating commands have completed or
// failed. readOnly is true if the requester is a read-only command;
// false for read-write.
func (cq *CommandQueue) GetWait(start, end proto.Key, readOnly bool, wg *sync.WaitGroup) {
// This gives us a memory-efficient end key if end is empty.
if len(end) == 0 {
end = start.Next()
start = end[:len(start)]
}
for _, c := range cq.cache.GetOverlaps(start, end) {
c := c.Value.(*cmd)
// Only add to the wait group if one of the commands isn't read-only.
if !readOnly || !c.readOnly {
c.pending = append(c.pending, wg)
wg.Add(1)
}
}
}
// MetaReverseScanBounds returns the range [start,end) within which the desired
// meta record can be found by means of a reverse engine scan. The given key
// must be a valid RangeMetaKey as defined by ValidateRangeMetaKey.
func MetaReverseScanBounds(key proto.Key) (proto.Key, proto.Key, error) {
if key.Equal(proto.KeyMin) || key.Equal(Meta1Prefix) {
return nil, nil, NewInvalidRangeMetaKeyError("KeyMin and Meta1Prefix can't be used as the key of reverse scan", key)
}
if key.Equal(Meta2Prefix) {
// Special case Meta2Prefix: this is the first key in Meta2, and the scan
// interval covers all of Meta1.
return Meta1Prefix, key.Next(), nil
}
// Otherwise find the first entry greater than the given key and find the last entry
// in the same prefix. For MVCCReverseScan the endKey is exclusive, if we want to find
// the range descriptor the given key specified,we need to set the key.Next() as the
// MVCCReverseScan`s endKey. For example:
// If we have ranges ["", "f") and ["f", "z"), then we'll have corresponding meta records
// at "f" and "z". If you're looking for the meta record for key "f", then you want the
// second record (exclusive in MVCCReverseScan), hence key.Next() below.
return key[:len(Meta1Prefix)], key.Next(), nil
}
// GetMax returns the maximum read and write timestamps which overlap
// the interval spanning from start to end. Cached timestamps matching
// the specified txnID are not considered. If no part of the specified
// range is overlapped by timestamps in the cache, the low water
// timestamp is returned for both read and write timestamps.
//
// The txn ID prevents restarts with a pattern like: read("a"),
// write("a"). The read adds a timestamp for "a". Then the write (for
// the same transaction) would get that as the max timestamp and be
// forced to increment it. This allows timestamps from the same txn
// to be ignored.
func (tc *TimestampCache) GetMax(start, end proto.Key, txnID []byte) (proto.Timestamp, proto.Timestamp) {
if len(end) == 0 {
end = start.Next()
}
maxR := tc.lowWater
maxW := tc.lowWater
for _, o := range tc.cache.GetOverlaps(start, end) {
ce := o.Value.(cacheEntry)
if ce.txnID == nil || txnID == nil || !proto.TxnIDEqual(txnID, ce.txnID) {
if ce.readOnly && maxR.Less(ce.timestamp) {
maxR = ce.timestamp
} else if !ce.readOnly && maxW.Less(ce.timestamp) {
maxW = ce.timestamp
}
}
}
return maxR, maxW
}
// getCachedRangeDescriptorLocked is a helper function to retrieve the
// descriptor of the range which contains the given key, if present in the
// cache. It is assumed that the caller holds a read lock on rdc.rangeCacheMu.
func (rdc *rangeDescriptorCache) getCachedRangeDescriptorLocked(key proto.Key) (
rangeCacheKey, *proto.RangeDescriptor) {
// The cache is indexed using the end-key of the range, but the
// end-key is non-inclusive. If inclusive is false, we access the
// cache using key.Next().
metaKey := keys.RangeMetaKey(key.Next())
k, v, ok := rdc.rangeCache.Ceil(rangeCacheKey(metaKey))
if !ok {
return nil, nil
}
metaEndKey := k.(rangeCacheKey)
rd := v.(*proto.RangeDescriptor)
// Check that key actually belongs to range
if !rd.ContainsKey(keys.KeyAddress(key)) {
return nil, nil
}
return metaEndKey, rd
}
func (db *testDescriptorDB) getDescriptor(key proto.Key) []proto.RangeDescriptor {
response := make([]proto.RangeDescriptor, 0, 3)
for i := 0; i < 3; i++ {
v := db.data.Ceil(testDescriptorNode{
&proto.RangeDescriptor{
EndKey: key.Next(),
},
})
if v == nil {
break
}
response = append(response, *(v.(testDescriptorNode).RangeDescriptor))
// Break to keep from skidding off the end of the available ranges.
if response[i].EndKey.Equal(proto.KeyMax) {
break
}
key = proto.Key(response[i].EndKey).Next()
}
return response
}
// MetaScanBounds returns the range [start,end) within which the desired meta
// record can be found by means of an engine scan. The given key must be a
// valid RangeMetaKey as defined by validateRangeMetaKey.
func MetaScanBounds(key proto.Key) (proto.Key, proto.Key, error) {
if err := validateRangeMetaKey(key); err != nil {
return nil, nil, err
}
if key.Equal(Meta2KeyMax) {
return nil, nil, NewInvalidRangeMetaKeyError("Meta2KeyMax can't be used as the key of scan", key)
}
if key.Equal(proto.KeyMin) {
// Special case KeyMin: find the first entry in meta1.
return Meta1Prefix, Meta1Prefix.PrefixEnd(), nil
}
if key.Equal(Meta1KeyMax) {
// Special case Meta1KeyMax: this is the last key in Meta1, we don't want
// to start at Next().
return key, Meta1Prefix.PrefixEnd(), nil
}
// Otherwise find the first entry greater than the given key in the same meta prefix.
return key.Next(), proto.Key(key[:len(Meta1Prefix)]).PrefixEnd(), nil
}
// addKeyRange adds the specified key range to the interval cache,
// taking care not to add this range if existing entries already
// completely cover the range.
func (tm *txnMetadata) addKeyRange(start, end proto.Key) {
// This gives us a memory-efficient end key if end is empty.
// The most common case for keys in the intents interval map
// is for single keys. However, the interval cache requires
// a non-empty interval, so we create two key slices which
// share the same underlying byte array.
if len(end) == 0 {
end = start.Next()
start = end[:len(start)]
}
key := tm.keys.NewKey(start, end)
for _, o := range tm.keys.GetOverlaps(start, end) {
if o.Key.Contains(key) {
return
} else if key.Contains(o.Key) {
tm.keys.Del(o.Key)
}
}
// Since no existing key range fully covered this range, add it now.
tm.keys.Add(key, nil)
}
// getCachedRangeDescriptorLocked is a helper function to retrieve the
// descriptor of the range which contains the given key, if present in the
// cache. It is assumed that the caller holds a read lock on rdc.rangeCacheMu.
func (rdc *rangeDescriptorCache) getCachedRangeDescriptorLocked(key proto.Key, isReverse bool) (
rangeCacheKey, *proto.RangeDescriptor) {
// The cache is indexed using the end-key of the range, but the
// end-key is non-inclusive.
var metaKey proto.Key
if !isReverse {
// If it is not reverse scan, we access the cache using key.Next().
metaKey = keys.RangeMetaKey(key.Next())
} else {
// Because reverse scan request is begining at end key(exclusive),so we
// access the cache using key directly.
metaKey = keys.RangeMetaKey(key)
}
k, v, ok := rdc.rangeCache.Ceil(rangeCacheKey(metaKey))
if !ok {
return nil, nil
}
metaEndKey := k.(rangeCacheKey)
rd := v.(*proto.RangeDescriptor)
// Check that key actually belongs to the range.
if !rd.ContainsKey(keys.KeyAddress(key)) {
// The key is the EndKey of the range in reverse scan, just return the range descriptor.
if isReverse && key.Equal(rd.EndKey) {
return metaEndKey, rd
}
return nil, nil
}
// The key is the StartKey of the range in reverse scan. We need to return the previous range
// descriptor, but it is not in the cache yet.
if isReverse && key.Equal(rd.StartKey) {
return nil, nil
}
return metaEndKey, rd
}