Skip to content

refactor(cache): Integrate the Radix LRU cache with the --experimental-enable-optimized-metadata-cache flag #4872

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.

Sign up for GitHub

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

Open
vedantdas-source wants to merge 2 commits into
base: radix-trie-cache
Choose a base branch
from
+54 −6
Open

refactor(cache): Integrate the Radix LRU cache with the --experimental-enable-optimized-metadata-cache flag #4872

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
df3bf69
used experimental flag to switch to radix cache when flag is set
vedantdas-source Jul 10, 2026
4420c9b
modified benchmarks to test radix cache
vedantdas-source Jul 10, 2026
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 cmd/mount.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,6 +108,7 @@ be interacting with the file system.`)
DummyIOCfg: newConfig.DummyIo,
IsTypeCacheDeprecated: newConfig.EnableTypeCacheDeprecation,
ImplicitDir: newConfig.ImplicitDirs,
EnableOptimizedMetadataCache: newConfig.MetadataCache.ExperimentalEnableOptimizedMetadataCache,
}
bm := gcsx.NewBucketManager(bucketCfg, storageHandle)

Expand Down
15 changes: 14 additions & 1 deletion internal/cache/lru/lru_memory_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,8 +93,21 @@ func TestMapCacheWorkloads(t *testing.T) {
alloc := getMemStats()
pureMem := alloc - baseMem
runtime.KeepAlive(pureCache)
runtime.KeepAlive(paths)

t.Logf("%-20s Heap Used: %10.2f MB\n", "MapLRU", float64(pureMem)/(1024*1024))

// 2. Pure RadixLRU
baseMemRadix := getMemStats()

radixCache := lru.NewRadixCache(capacity)
for _, p := range paths {
_, _ = radixCache.Insert(p, dummyValue{})
}
allocRadix := getMemStats()
pureMemRadix := allocRadix - baseMemRadix
runtime.KeepAlive(radixCache)
runtime.KeepAlive(paths)

t.Logf("%-20s Heap Used: %10.2f MB\n", "RadixLRU", float64(pureMemRadix)/(1024*1024))
}
}
15 changes: 13 additions & 2 deletions internal/cache/lru/lru_workload_benchmark_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,13 +88,14 @@ func benchmarkErasePrefix(b *testing.B, cache lru.Cache, prefixMap map[string][]
}

i := 0
b.ResetTimer()
Comment thread
vedantdas-source marked this conversation as resolved.

@vipnydav vipnydav Jul 16, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

b.Loop() resets the timer automatically on its first call. So we can remove it before b.Loop().

for b.Loop() {
b.StopTimer() // Timer must be running when b.Loop() evaluates, stop it for setup
prefix := prefixes[i%len(prefixes)]

@vipnydav vipnydav Jul 16, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Since prefix := prefixes[i%len(prefixes)] would take negligible time.

Can we do something like:

	i := 0
	for b.Loop() {
		// 2. TIMED: Erase prefix (Timer is RUNNING from b.Loop)
		prefix := prefixes[i%len(prefixes)]
		cache.EraseEntriesWithGivenPrefix(prefix)
		// 3. UNTIMED: Pause clock, restore erased keys, and restart timer for next iteration
		b.StopTimer()
		for _, key := range prefixMap[prefix] {
			cache.Insert(key, dummyValue{})
		}
		i++
		b.StartTimer() // Timer MUST be running when b.Loop() evaluates next!
	}


b.StartTimer()
cache.EraseEntriesWithGivenPrefix(prefix)
//reset the timer so that we only measure erasure time
b.StopTimer()
b.StopTimer() // Stop timer for cleanup

//restore the map
keysToRestore := prefixMap[prefix]
Expand All @@ -103,6 +104,7 @@ func benchmarkErasePrefix(b *testing.B, cache lru.Cache, prefixMap map[string][]
}

i++
b.StartTimer() // Restart timer before b.Loop() evaluates
}
}

Expand All @@ -116,14 +118,23 @@ func runBenchmarks(b *testing.B, name string, depth int) {
b.Run(name+"_MapLRU_Insert", func(b *testing.B) {
benchmarkInsert(b, lru.NewCache(capacity), keys)
})
b.Run(name+"_RadixLRU_Insert", func(b *testing.B) {
benchmarkInsert(b, lru.NewRadixCache(capacity), keys)
})

b.Run(name+"_MapLRU_Lookup", func(b *testing.B) {
benchmarkLookup(b, lru.NewCache(capacity), keys)
})
b.Run(name+"_RadixLRU_Lookup", func(b *testing.B) {
benchmarkLookup(b, lru.NewRadixCache(capacity), keys)
})

b.Run(name+"_MapLRU_ErasePrefix", func(b *testing.B) {
benchmarkErasePrefix(b, lru.NewCache(capacity), prefixMap, prefixes)
})
b.Run(name+"_RadixLRU_ErasePrefix", func(b *testing.B) {
benchmarkErasePrefix(b, lru.NewRadixCache(capacity), prefixMap, prefixes)
})
}

func BenchmarkLRU_Flat(b *testing.B) {
Expand Down
20 changes: 18 additions & 2 deletions internal/cache/metadata/statcache_memory_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,7 +79,7 @@ func TestStatCacheMemoryWorkloads(t *testing.T) {
for _, w := range workloads {
t.Logf("=== WORKLOAD: %s (%d files) ===", w, count)

// 1. StatCache
// 1. StatCache (Map)
paths := generatePaths(w, count)
baseMem := getMemStats()

Expand All @@ -93,8 +93,24 @@ func TestStatCacheMemoryWorkloads(t *testing.T) {
alloc := getMemStats()
pureMem := alloc - baseMem
runtime.KeepAlive(statCache)

t.Logf("%-20s Heap Used: %10.2f MB\n", "Map StatCache", float64(pureMem)/(1024*1024))

@vipnydav vipnydav Jul 16, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

See if this can be converted to Table driven test or is it better to create 2 independent tests.
Also use AAA.

// 2. StatCache (Radix)
baseMemRadix := getMemStats()

sharedRadixCache := lru.NewRadixCache(capacity)
statRadixCache := metadata.NewStatCacheBucketView(sharedRadixCache, "")

for _, p := range paths {
statRadixCache.Insert(&gcs.MinObject{Name: p}, expiration)
}

allocRadix := getMemStats()
pureMemRadix := allocRadix - baseMemRadix
runtime.KeepAlive(statRadixCache)
runtime.KeepAlive(paths)

t.Logf("%-20s Heap Used: %10.2f MB\n", "StatCache", float64(pureMem)/(1024*1024))
t.Logf("%-20s Heap Used: %10.2f MB\n", "Radix StatCache", float64(pureMemRadix)/(1024*1024))
}
}
9 changes: 8 additions & 1 deletion internal/gcsx/bucket_manager.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,6 +78,8 @@ type BucketConfig struct {
IsTypeCacheDeprecated bool

ImplicitDir bool

EnableOptimizedMetadataCache bool
}

// BucketManager manages the lifecycle of buckets.
Expand All @@ -103,7 +105,12 @@ type bucketManager struct {
func NewBucketManager(config BucketConfig, storageHandle storage.StorageHandle) BucketManager {
var c lru.Cache
if config.StatCacheMaxSizeMB > 0 {
c = lru.NewCache(util.MiBsToBytes(config.StatCacheMaxSizeMB))
cacheSize := util.MiBsToBytes(config.StatCacheMaxSizeMB)
if config.EnableOptimizedMetadataCache {
c = lru.NewRadixCache(cacheSize)
} else {
c = lru.NewCache(cacheSize)
}
}

bm := &bucketManager{
Expand Down
Loading