frostfs-sdk-go/netmap/selector.go
Evgenii Stratonikov 1ac947af44 [#236] netmap: Reuse slice for weights in ContainerNodes()
```
goos: linux
goarch: amd64
pkg: git.frostfs.info/TrueCloudLab/frostfs-sdk-go/netmap
cpu: 11th Gen Intel(R) Core(TM) i5-1135G7 @ 2.40GHz
                                                              │    alloc     │               weights               │
                                                              │    sec/op    │    sec/op     vs base               │
Netmap_ContainerNodes/REP_2-8                                   8.677µ ±  6%   8.384µ ± 10%       ~ (p=0.247 n=10)
Netmap_ContainerNodes/REP_2_IN_X_CBF_2_SELECT_2_FROM_*_AS_X-8   7.946µ ± 14%   7.998µ ±  6%       ~ (p=0.481 n=10)
geomean                                                         8.303µ         8.189µ        -1.38%

                                                              │    alloc     │               weights               │
                                                              │     B/op     │     B/op      vs base               │
Netmap_ContainerNodes/REP_2-8                                   7.734Ki ± 0%   7.617Ki ± 0%  -1.52% (p=0.000 n=10)
Netmap_ContainerNodes/REP_2_IN_X_CBF_2_SELECT_2_FROM_*_AS_X-8   7.156Ki ± 0%   7.039Ki ± 0%  -1.64% (p=0.000 n=10)
geomean                                                         7.440Ki        7.322Ki       -1.58%

                                                              │   alloc    │              weights               │
                                                              │ allocs/op  │ allocs/op   vs base                │
Netmap_ContainerNodes/REP_2-8                                   92.00 ± 0%   77.00 ± 0%  -16.30% (p=0.000 n=10)
Netmap_ContainerNodes/REP_2_IN_X_CBF_2_SELECT_2_FROM_*_AS_X-8   92.00 ± 0%   77.00 ± 0%  -16.30% (p=0.000 n=10)
geomean                                                         92.00        77.00       -16.30%
```

Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
2024-07-12 14:22:20 +03:00

182 lines
4.6 KiB
Go

package netmap
import (
"cmp"
"fmt"
"slices"
"git.frostfs.info/TrueCloudLab/frostfs-api-go/v2/netmap"
"git.frostfs.info/TrueCloudLab/hrw"
)
// processSelectors processes selectors and returns error is any of them is invalid.
func (c *context) processSelectors(p PlacementPolicy) error {
for i := range p.selectors {
fName := p.selectors[i].GetFilter()
if fName != mainFilterName {
_, ok := c.processedFilters[p.selectors[i].GetFilter()]
if !ok {
return fmt.Errorf("%w: SELECT FROM '%s'", errFilterNotFound, fName)
}
}
sName := p.selectors[i].GetName()
c.processedSelectors[sName] = &p.selectors[i]
result, err := c.getSelection(p.selectors[i])
if err != nil {
return err
}
c.selections[sName] = result
}
return nil
}
// calcNodesCount returns number of buckets and minimum number of nodes in every bucket
// for the given selector.
func calcNodesCount(s netmap.Selector) (int, int) {
switch s.GetClause() {
case netmap.Same:
return 1, int(s.GetCount())
default:
return int(s.GetCount()), 1
}
}
// calcBucketWeight computes weight for a node bucket.
func calcBucketWeight(ns nodes, a aggregator, wf weightFunc) float64 {
for i := range ns {
a.Add(wf(ns[i]))
}
return a.Compute()
}
// getSelection returns nodes grouped by s.attribute.
// Last argument specifies if more buckets can be used to fulfill CBF.
func (c *context) getSelection(s netmap.Selector) ([]nodes, error) {
bucketCount, nodesInBucket := calcNodesCount(s)
buckets := c.getSelectionBase(s)
if c.strict && len(buckets) < bucketCount {
return nil, fmt.Errorf("%w: '%s'", errNotEnoughNodes, s.GetName())
}
// We need deterministic output in case there is no pivot.
// If pivot is set, buckets are sorted by HRW.
// However, because initial order influences HRW order for buckets with equal weights,
// we also need to have deterministic input to HRW sorting routine.
if len(c.hrwSeed) == 0 {
if s.GetAttribute() == "" {
slices.SortFunc(buckets, func(b1, b2 nodeAttrPair) int {
return cmp.Compare(b1.nodes[0].Hash(), b2.nodes[0].Hash())
})
} else {
slices.SortFunc(buckets, func(b1, b2 nodeAttrPair) int {
return cmp.Compare(b1.attr, b2.attr)
})
}
}
maxNodesInBucket := nodesInBucket * int(c.cbf)
res := make([]nodes, 0, len(buckets))
fallback := make([]nodes, 0, len(buckets))
for i := range buckets {
ns := buckets[i].nodes
if len(ns) >= maxNodesInBucket {
res = append(res, ns[:maxNodesInBucket])
} else if len(ns) >= nodesInBucket {
fallback = append(fallback, ns)
}
}
if len(res) < bucketCount {
// Fallback to using minimum allowed backup factor (1).
res = append(res, fallback...)
if c.strict && len(res) < bucketCount {
return nil, fmt.Errorf("%w: '%s'", errNotEnoughNodes, s.GetName())
}
}
if len(c.hrwSeed) != 0 {
weights := make([]float64, len(res))
a := new(meanIQRAgg)
for i := range res {
a.clear()
weights[i] = calcBucketWeight(res[i], a, c.weightFunc)
}
hrw.SortHasherSliceByWeightValue(res, weights, c.hrwSeedHash)
}
if len(res) < bucketCount {
if len(res) == 0 {
return nil, errNotEnoughNodes
}
bucketCount = len(res)
}
if s.GetAttribute() == "" {
res, fallback = res[:bucketCount], res[bucketCount:]
for i := range fallback {
index := i % bucketCount
if len(res[index]) >= maxNodesInBucket {
break
}
res[index] = append(res[index], fallback[i]...)
}
}
return res[:bucketCount], nil
}
type nodeAttrPair struct {
attr string
nodes nodes
}
// getSelectionBase returns nodes grouped by selector attribute.
// It it guaranteed that each pair will contain at least one node.
func (c *context) getSelectionBase(s netmap.Selector) []nodeAttrPair {
fName := s.GetFilter()
f := c.processedFilters[fName]
isMain := fName == mainFilterName
result := []nodeAttrPair{}
nodeMap := map[string][]NodeInfo{}
attr := s.GetAttribute()
for i := range c.netMap.nodes {
if c.usedNodes[c.netMap.nodes[i].hash] {
continue
}
if isMain || c.match(f, c.netMap.nodes[i]) {
if attr == "" {
// Default attribute is transparent identifier which is different for every node.
result = append(result, nodeAttrPair{attr: "", nodes: nodes{c.netMap.nodes[i]}})
} else {
v := c.netMap.nodes[i].Attribute(attr)
nodeMap[v] = append(nodeMap[v], c.netMap.nodes[i])
}
}
}
if attr != "" {
for k, ns := range nodeMap {
result = append(result, nodeAttrPair{attr: k, nodes: ns})
}
}
if len(c.hrwSeed) != 0 {
var ws []float64
for i := range result {
ws = result[i].nodes.appendWeightsTo(c.weightFunc, ws[:0])
hrw.SortHasherSliceByWeightValue(result[i].nodes, ws, c.hrwSeedHash)
}
}
return result
}