// setupMVCCData writes up to numVersions values at each of numKeys
// keys. The number of versions written for each key is chosen
// randomly according to a uniform distribution. Each successive
// version is written starting at 5ns and then in 5ns increments. This
// allows scans at various times, starting at t=5ns, and continuing to
// t=5ns*(numVersions+1). A version for each key will be read on every
// such scan, but the dynamics of the scan will change depending on
// the historical timestamp. Earlier timestamps mean scans which must
// skip more historical versions; later timestamps mean scans which
// skip fewer.
//
// The creation of the rocksdb database is time consuming, especially
// for larger numbers of versions. The database is persisted between
// runs and stored in the current directory as
// "mvcc_scan_<versions>_<keys>".
func setupMVCCScanData(numVersions, numKeys int, b *testing.B) (*RocksDB, *stop.Stopper) {
loc := fmt.Sprintf("mvcc_scan_%d_%d", numVersions, numKeys)
exists := true
if _, err := os.Stat(loc); os.IsNotExist(err) {
exists = false
}
log.Infof("creating mvcc data: %s", loc)
const cacheSize = 8 << 30 // 8 GB
stopper := stop.NewStopper()
rocksdb := NewRocksDB(roachpb.Attributes{Attrs: []string{"ssd"}}, loc, cacheSize, stopper)
if err := rocksdb.Open(); err != nil {
b.Fatalf("could not create new rocksdb db instance at %s: %v", loc, err)
}
if exists {
return rocksdb, stopper
}
rng, _ := randutil.NewPseudoRand()
keys := make([]roachpb.Key, numKeys)
nvs := make([]int, numKeys)
for t := 1; t <= numVersions; t++ {
walltime := int64(5 * t)
ts := makeTS(walltime, 0)
batch := rocksdb.NewBatch()
for i := 0; i < numKeys; i++ {
if t == 1 {
keys[i] = roachpb.Key(encoding.EncodeUvarint([]byte("key-"), uint64(i)))
nvs[i] = int(rand.Int31n(int32(numVersions)) + 1)
}
// Only write values if this iteration is less than the random
// number of versions chosen for this key.
if t <= nvs[i] {
value := roachpb.Value{Bytes: randutil.RandBytes(rng, 1024)}
value.InitChecksum(keys[i])
if err := MVCCPut(batch, nil, keys[i], ts, value, nil); err != nil {
b.Fatal(err)
}
}
}
if err := batch.Commit(); err != nil {
b.Fatal(err)
}
batch.Close()
}
rocksdb.CompactRange(nil, nil)
return rocksdb, stopper
}
// append the given entries to the raft log. Takes the previous values of
// r.mu.lastIndex and r.mu.raftLogSize, and returns new values. We do this
// rather than modifying them directly because these modifications need to be
// atomic with the commit of the batch.
func (r *Replica) append(
ctx context.Context,
batch engine.ReadWriter,
prevLastIndex uint64,
prevRaftLogSize int64,
entries []raftpb.Entry,
) (uint64, int64, error) {
if len(entries) == 0 {
return prevLastIndex, prevRaftLogSize, nil
}
var diff enginepb.MVCCStats
var value roachpb.Value
for i := range entries {
ent := &entries[i]
key := keys.RaftLogKey(r.RangeID, ent.Index)
if err := value.SetProto(ent); err != nil {
return 0, 0, err
}
value.InitChecksum(key)
var err error
if ent.Index > prevLastIndex {
err = engine.MVCCBlindPut(ctx, batch, &diff, key, hlc.ZeroTimestamp, value, nil /* txn */)
} else {
err = engine.MVCCPut(ctx, batch, &diff, key, hlc.ZeroTimestamp, value, nil /* txn */)
}
if err != nil {
return 0, 0, err
}
}
// Delete any previously appended log entries which never committed.
lastIndex := entries[len(entries)-1].Index
for i := lastIndex + 1; i <= prevLastIndex; i++ {
err := engine.MVCCDelete(ctx, batch, &diff, keys.RaftLogKey(r.RangeID, i),
hlc.ZeroTimestamp, nil /* txn */)
if err != nil {
return 0, 0, err
}
}
if err := setLastIndex(ctx, batch, r.RangeID, lastIndex); err != nil {
return 0, 0, err
}
raftLogSize := prevRaftLogSize + diff.SysBytes
return lastIndex, raftLogSize, nil
}