mirror of
https://github.com/featurebasedb/featurebase.git
synced 2026-08-28 10:54:59 +00:00
016765d8a2
This partially-implemented prototype of the ingest API is based on our programmatic ingest API reference. It has noticable limitations, most crucially that it doesn't handle multi-node clusters right now. However, it basically implements the expected semantics. There's some noticeable performance issues to do with the high overhead of sorting bits in order to import them efficiently, but this is fixable. We also add the hooks to the internal client, and make the finisher logic a bit smarter. Much of this code was originally by Nia Weiss, but it's been merged and restructured a bit to get things broken into logical commits.
245 lines
7.4 KiB
Go
245 lines
7.4 KiB
Go
// Copyright 2021 Molecula Corp.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
package ingest
|
|
|
|
// "math/bits"
|
|
|
|
// HERE THERE BE DRAGONS
|
|
|
|
// This is some sorting logic Nia was experimenting with, which we aren't currently
|
|
// using, but which beat stdlib sort by a factor-of-several on at least some test
|
|
// data, so we aren't deleting it just yet.
|
|
|
|
// groupIDPairsByShard destructively groups ID pairs by shard.
|
|
// The returned slices reference the original pairs slice.
|
|
// func groupIDPairsByShard(pairs []IDPair) map[uint64][]IDPair {
|
|
// if len(pairs) == 0 {
|
|
// return nil
|
|
// }
|
|
//
|
|
// // Sort pairs by shard (in-place radix sort).
|
|
// // This may also change the order of pairs within a shard, but that should not matter.
|
|
// for {
|
|
// // Find the highest bit which needs to be sorted.
|
|
// var diffMask uint64
|
|
// prev := pairs[0].RecordID
|
|
// for _, v := range pairs[1:] {
|
|
// if v.RecordID < prev {
|
|
// diffMask |= v.RecordID ^ prev
|
|
// }
|
|
//
|
|
// prev = v.RecordID
|
|
// }
|
|
// diffLen := bits.Len64(diffMask)
|
|
// if diffLen <= shardwidth.Exponent {
|
|
// // The pairs are sorted by shard.
|
|
// break
|
|
// }
|
|
//
|
|
// // Select a right bit shift index such that the highest unsorted bit moves to the 128's place.
|
|
// shift := uint(diffLen) - 8
|
|
//
|
|
// // Create a mask that can be used to group values by sorted bits.
|
|
// sortedMask := ^uint64(0) << bits.Len64(diffMask)
|
|
//
|
|
// for i := 0; i < len(pairs); {
|
|
// // Select a group of pairs to sort.
|
|
// // While doing so, count the pairs within each bucket.
|
|
// j := i
|
|
// var buckets [256]struct {
|
|
// start, end uint
|
|
// }
|
|
// for group := pairs[i].RecordID & sortedMask; i < len(pairs) && pairs[i].RecordID&sortedMask == group; i++ {
|
|
// buckets[uint8(pairs[i].RecordID>>uint64(shift))].end++
|
|
// }
|
|
// group := pairs[j:i]
|
|
//
|
|
// // Assign indices within the group to the buckets.
|
|
// {
|
|
// var start uint
|
|
// for i := range buckets {
|
|
// bucket := &buckets[i]
|
|
// bucket.start = start
|
|
// bucket.end += start
|
|
// start = bucket.end
|
|
// }
|
|
// }
|
|
//
|
|
// // Split the group into the buckets.
|
|
// for i, b := range buckets {
|
|
// // There is no need to update the state of the current bucket - we will never reference it again after this.
|
|
// i := uint8(i)
|
|
// for j := b.start; j < b.end; j++ {
|
|
// // This inner loop may run quite a few times for the first few buckets, but no element will be moved more than twice per byte.
|
|
// for uint8(group[j].RecordID>>shift) != i {
|
|
// // This pair is in the wrong bucket.
|
|
// // Swap it into the correct bucket.
|
|
// dstBucket := &buckets[uint8(group[j].RecordID>>shift)]
|
|
// k := dstBucket.start
|
|
// dstBucket.start++
|
|
// group[j], group[k] = group[k], group[j]
|
|
// }
|
|
// }
|
|
// }
|
|
// }
|
|
// }
|
|
//
|
|
// // Split the pairs by shard.
|
|
// shards := make(map[uint64][]IDPair)
|
|
// for i := 0; i < len(pairs); {
|
|
// // Select the shard.
|
|
// shard := pairs[i].RecordID >> shardwidth.Exponent
|
|
//
|
|
// // Find all pairs in the shard.
|
|
// j := i
|
|
// incr := 1
|
|
// for i+incr < len(pairs) && pairs[i+incr].RecordID>>shardwidth.Exponent == shard {
|
|
// i += incr
|
|
// incr *= 2
|
|
// }
|
|
// for ; incr > 0; incr /= 2 {
|
|
// if i+incr < len(pairs) && pairs[i+incr].RecordID>>shardwidth.Exponent == shard {
|
|
// i += incr
|
|
// }
|
|
// }
|
|
// // that found us the last thing in this shard, so...
|
|
// i++
|
|
//
|
|
// // Add the shard, referencing the original slice.
|
|
// // This sets the cap so that we dont accidentally overwrite other shards data.
|
|
// shards[shard] = pairs[j:i:i]
|
|
// }
|
|
//
|
|
// return shards
|
|
// }
|
|
|
|
// pairsToZigZag converts a set of record-value pairs to Pilosa's zig-zag format.
|
|
// This assumes that all pairs are within the same shard.
|
|
// func pairsToZigZag(pairs []IDPair) []uint64 {
|
|
// const recordMask = (1 << shardwidth.Exponent) - 1
|
|
//
|
|
// dst := make([]uint64, len(pairs))
|
|
// for i, p := range pairs {
|
|
// dst[i] = (p.ID << shardwidth.Exponent) | (p.RecordID & recordMask)
|
|
// }
|
|
//
|
|
// return dst
|
|
// }
|
|
|
|
// radixSort64 sorts the data with radix-sort.
|
|
// The "shift" is the highest differing bit position, rounded down to a multiple of 8.
|
|
// If there are duplicates, this may change the duplicate count for some values.
|
|
// func radixSort64(data []uint64, shift uint) {
|
|
// if len(data) < 2 {
|
|
// return
|
|
// }
|
|
// if shift <= 8 {
|
|
// // The data falls into a 16-bit span, so the remaining digits can be sorted simultaneously with a bitmask.
|
|
// maskSort(data)
|
|
// return
|
|
// }
|
|
//
|
|
// // Count the values within each bucket.
|
|
// var buckets [256]struct {
|
|
// start, end uint
|
|
// }
|
|
// for _, v := range data {
|
|
// buckets[uint8(v>>shift)].end++
|
|
// }
|
|
//
|
|
// // Assign indices within the group to the buckets.
|
|
// {
|
|
// var start uint
|
|
// for i := range buckets {
|
|
// bucket := &buckets[i]
|
|
// bucket.start = start
|
|
// bucket.end += start
|
|
// start = bucket.end
|
|
// }
|
|
// }
|
|
//
|
|
// // Split the data into the buckets.
|
|
// var start uint
|
|
// for i, b := range buckets {
|
|
// // Replace misplaced values until the contents of the bucket all have the correct digit.
|
|
// i := uint8(i)
|
|
// for j := b.start; j < b.end; j++ {
|
|
// // This inner loop may run quite a few times for the first few buckets, but it will never run more than once-per-element-per-byte.
|
|
// for uint8(data[j]>>shift) != i {
|
|
// // This pair is in the wrong bucket.
|
|
// // Swap it into the correct bucket.
|
|
// dstBucket := &buckets[uint8(data[j]>>shift)]
|
|
// k := dstBucket.start
|
|
// dstBucket.start++
|
|
// data[j], data[k] = data[k], data[j]
|
|
// }
|
|
// }
|
|
//
|
|
// // Sort the contents of the bucket.
|
|
// data := data[start:b.end]
|
|
// switch {
|
|
// case len(data) < 64:
|
|
// // Use insertion-sort because the data is too small for a more complex algorithm to be efficient.
|
|
// for i := 0; i < len(data); i++ {
|
|
// for j := i; j > 0 && data[j-1] > data[j]; j-- {
|
|
// data[j-1], data[j] = data[j], data[j-1]
|
|
// }
|
|
// }
|
|
//
|
|
// default:
|
|
// // Sort the next byte recursively.
|
|
// radixSort64(data, shift-8)
|
|
// }
|
|
// start = b.end
|
|
// }
|
|
// }
|
|
|
|
// maskSort sorts integers using a bitmask.
|
|
// The values must all fall within one 16-bit span.
|
|
// If there are duplicates, this may change the duplicate count for some values.
|
|
// func maskSort(data []uint64) {
|
|
// if len(data) < 2 {
|
|
// return
|
|
// }
|
|
//
|
|
// base := data[0] &^ ((1 << 16) - 1)
|
|
//
|
|
// // Dump everything into the mask.
|
|
// var mask [(1 << 16) / 64]uint64
|
|
// for _, v := range data {
|
|
// mask[uint16(v)/64] |= 1 << (v % 64)
|
|
// }
|
|
//
|
|
// // Scan through the set bits in the mask.
|
|
// k := 0
|
|
// for i, w := range mask {
|
|
// for w != 0 {
|
|
// j := bits.TrailingZeros64(w)
|
|
// w &^= 1 << j
|
|
// data[k] = 64*uint64(i) + uint64(j) + base
|
|
// k++
|
|
// }
|
|
// }
|
|
//
|
|
// if k < len(data) {
|
|
// // Copy the ending value to fill up the rest of the space.
|
|
// // This happens once for each duplicate value.
|
|
// endVal := data[k-1]
|
|
// for k < len(data) {
|
|
// data[k] = endVal
|
|
// k++
|
|
// }
|
|
// }
|
|
// }
|