Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

perf: Speedup cachekv iterator on large deletions & IBC v2 upgrade logic #10741

Merged
merged 6 commits into from
Dec 12, 2021
Merged
Show file tree
Hide file tree
Changes from 5 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -149,6 +149,7 @@ Ref: https://keepachangelog.com/en/1.0.0/
* [\#10468](https://github.com/cosmos/cosmos-sdk/pull/10468) Allow futureOps to queue additional operations in simulations
* [\#10625](https://github.com/cosmos/cosmos-sdk/pull/10625) Add `--fee-payer` CLI flag
* (cli) [\#10683](https://github.com/cosmos/cosmos-sdk/pull/10683) In CLI, allow 1 SIGN_MODE_DIRECT signer in transactions with multiple signers.
* (store) [\#10741](https://github.com/cosmos/cosmos-sdk/pull/10741) Significantly speedup iterator creation after delete heavy workloads. Significantly improves IBC migration times.

### Bug Fixes

Expand Down
20 changes: 13 additions & 7 deletions store/cachekv/memiterator.go
Original file line number Diff line number Diff line change
@@ -1,6 +1,8 @@
package cachekv

import (
"bytes"

dbm "github.com/tendermint/tm-db"

"github.com/cosmos/cosmos-sdk/store/types"
Expand All @@ -12,6 +14,7 @@ import (
type memIterator struct {
types.Iterator

lastKey []byte
deleted map[string]struct{}
}

Expand All @@ -29,22 +32,25 @@ func newMemIterator(start, end []byte, items *dbm.MemDB, deleted map[string]stru
panic(err)
}

newDeleted := make(map[string]struct{})
for k, v := range deleted {
newDeleted[k] = v
}

return &memIterator{
Iterator: iter,

deleted: newDeleted,
lastKey: nil,
deleted: deleted,
}
}

func (mi *memIterator) Value() []byte {
key := mi.Iterator.Key()
if _, ok := mi.deleted[string(key)]; ok {
// We need to handle the case where deleted is modified and includes our current key
// We handle this by maintaining a lastKey object in the iterator.
// If the current key is the same as the last key (and last key is not nil / the start)
// then we are calling value on the same thing as last time.
// Therefore we don't check the mi.deleted to see if this key is included in there.
reCallingOnOldLastKey := (mi.lastKey != nil) && bytes.Equal(key, mi.lastKey)
if _, ok := mi.deleted[string(key)]; ok && !reCallingOnOldLastKey {
return nil
}
mi.lastKey = key
return mi.Iterator.Value()
}
51 changes: 51 additions & 0 deletions store/cachekv/store_bench_test.go
Original file line number Diff line number Diff line change
Expand Up @@ -78,6 +78,53 @@ func benchmarkRandomSet(b *testing.B, keysize int) {
for _, k := range keys {
kvstore.Set(k, value)
}

iter := kvstore.Iterator(keys[0], keys[b.N])
defer iter.Close()

for _ = iter.Key(); iter.Valid(); iter.Next() {
// deadcode elimination stub
sink = iter
}
}

// Benchmark creating an iterator on a parent with D entries,
// that are all deleted in the cacheKV store.
// We essentially are benchmarking the cacheKV iterator creation & iteration times
// with the number of entries deleted in the parent.
func benchmarkIteratorOnParentWithManyDeletes(b *testing.B, numDeletes int) {
mem := dbadapter.Store{DB: dbm.NewMemDB()}

// Use a singleton for value, to not waste time computing it
value := randSlice(32)
// Use simple values for keys, pick a random start,
// and take next D keys sequentially after.
startKey := randSlice(32)
keys := generateSequentialKeys(startKey, numDeletes)
// setup parent db with D keys.
for _, k := range keys {
mem.Set(k, value)
}
kvstore := cachekv.NewStore(mem)
// Delete all keys from the cache KV store.
// The keys[1:] is to keep at least one entry in parent, due to a bug in the SDK iterator design.
// Essentially the iterator will never be valid, in that it should never run.
// However, this is incompatible with the for loop structure the SDK uses, hence
// causes a panic. Thus we do keys[1:].
Comment on lines +110 to +113
Copy link
Contributor Author

@ValarDragon ValarDragon Dec 12, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This bug is not related to this PR, I just noticed it in constructing the benchmark / working with the relevant logic.

This can be spot checked if folks want by just checking out the first commit in this PR, and removing the [1:].

This should probably moved to its own issue where we discuss the iterator structure, but its relatively low priority given all the other module / app writing improvements in the works 🙏

for _, k := range keys[1:] {
kvstore.Delete(k)
}

b.ReportAllocs()
b.ResetTimer()

iter := kvstore.Iterator(keys[0], keys[b.N])
defer iter.Close()

for _ = iter.Key(); iter.Valid(); iter.Next() {
// deadcode elimination stub
sink = iter
}
}

func BenchmarkBlankParentIteratorNextKeySize32(b *testing.B) {
Expand All @@ -91,3 +138,7 @@ func BenchmarkBlankParentAppendKeySize32(b *testing.B) {
func BenchmarkSetKeySize32(b *testing.B) {
benchmarkRandomSet(b, 32)
}

func BenchmarkIteratorOnParentWith1MDeletes(b *testing.B) {
benchmarkIteratorOnParentWithManyDeletes(b, 1_000_000)
}