Files
seaweedfs/weed/cluster/lock_manager/hash_ring.go
T
Chris Lu f037fc4dce s3: dial the object lock's primary filer directly (#9626)
* s3: dial the object lock's primary filer directly

The S3 object write lock builds a fresh short-lived lock per write, each
starting at the seed filer. When the seed isn't the key's hash-ring primary
the filer forwards the request to the primary, and in multi-cluster setups
that forward crosses clusters on every write.

Give the lock client a view of the filer lock ring, fed by the master's
LockRingUpdate broadcasts the gateway already receives, so it dials the
primary directly. The view tracks filer membership by version; a stale view
stays correct because the filer still forwards as a fallback.

Also send the initial ring snapshot to S3 clients, not just filers.

* s3: subscribe to lock-ring updates before starting the master loop

The master delivers the initial LockRingUpdate once, on connect. Registering the
callback after KeepConnectedToMaster started left a window where that first
update could arrive before the handler was set and be dropped, delaying the ring
view until the next membership change. Build the lock client and register the
callback in the masters block before launching the loop; the filers block reuses
that client (or creates a plain one when no masters are configured).

* lock_manager: build the hash ring in a deterministic server order

rebuildRing ranged over the server set (a map), whose iteration order is
randomized per process. On a vnode hash collision the last writer into
vnodeToServer wins, so two nodes holding the same server set could resolve the
collision to different servers and disagree on the primary for keys near that
slot. Now that the S3 gateway also computes PrimaryForKey, such a disagreement
would route the same key to different filers and defeat per-path serialization.

Iterate the servers in sorted order so the ring is identical on every node with
the same set, regardless of discovery order.

* lock_manager: skip redundant ring rebuilds, trim comments

SetRing now ignores a non-zero version at or below the current one once a ring
exists, so repeated LockRingUpdate broadcasts on reconnect no longer rebuild the
ring.

* s3: hold the lock-ring client on the server for route-by-key

Store the object-write lock client on S3ApiServer so handlers can resolve a
key's owner filer via PrimaryForKey.
2026-05-24 00:40:43 -07:00

207 lines
5.5 KiB
Go

package lock_manager
import (
"hash/crc32"
"sort"
"sync"
"github.com/seaweedfs/seaweedfs/weed/pb"
)
const DefaultVnodeCount = 50
// HashRing implements consistent hashing with virtual nodes.
// When a server is removed, only the keys that hashed to that server
// are remapped (to the next server on the ring), leaving all other
// key-to-server mappings stable.
//
// UPGRADE NOTE: This replaces the previous modulo-based hashing
// (hash % len(servers)). The two schemes compute different primaries
// for the same key, so all filer nodes in the cluster must be upgraded
// together (or via a rolling restart that completes within the lock TTL
// window of ~7 seconds) to avoid routing disagreements.
type HashRing struct {
mu sync.RWMutex
vnodeCount int
sortedHashes []uint32 // sorted ring positions
vnodeToServer map[uint32]pb.ServerAddress // ring position → server
servers map[pb.ServerAddress]struct{} // set of all servers
}
func NewHashRing(vnodeCount int) *HashRing {
if vnodeCount <= 0 {
vnodeCount = DefaultVnodeCount
}
return &HashRing{
vnodeCount: vnodeCount,
vnodeToServer: make(map[uint32]pb.ServerAddress),
servers: make(map[pb.ServerAddress]struct{}),
}
}
// AddServer adds a server with virtual nodes to the ring.
func (hr *HashRing) AddServer(server pb.ServerAddress) {
hr.mu.Lock()
defer hr.mu.Unlock()
if _, exists := hr.servers[server]; exists {
return
}
hr.servers[server] = struct{}{}
hr.rebuildRing()
}
// RemoveServer removes a server and its virtual nodes from the ring.
func (hr *HashRing) RemoveServer(server pb.ServerAddress) {
hr.mu.Lock()
defer hr.mu.Unlock()
if _, exists := hr.servers[server]; !exists {
return
}
delete(hr.servers, server)
hr.rebuildRing()
}
// SetServers replaces the entire server set.
func (hr *HashRing) SetServers(servers []pb.ServerAddress) {
hr.mu.Lock()
defer hr.mu.Unlock()
hr.servers = make(map[pb.ServerAddress]struct{}, len(servers))
for _, s := range servers {
hr.servers[s] = struct{}{}
}
hr.rebuildRing()
}
// GetPrimaryAndBackup returns the primary server for a key and its backup
// (the next distinct server clockwise on the ring).
// If there is only one server, backup is empty.
func (hr *HashRing) GetPrimaryAndBackup(key string) (primary, backup pb.ServerAddress) {
hr.mu.RLock()
defer hr.mu.RUnlock()
if len(hr.sortedHashes) == 0 {
return "", ""
}
hash := hashKey(key)
idx := hr.search(hash)
primary = hr.vnodeToServer[hr.sortedHashes[idx]]
// Walk clockwise to find a different server for backup
ringLen := len(hr.sortedHashes)
for i := 1; i < ringLen; i++ {
candidate := hr.vnodeToServer[hr.sortedHashes[(idx+i)%ringLen]]
if candidate != primary {
backup = candidate
return
}
}
// Only one server — no backup
return primary, ""
}
// GetPrimary returns just the primary server for a key.
func (hr *HashRing) GetPrimary(key string) pb.ServerAddress {
hr.mu.RLock()
defer hr.mu.RUnlock()
if len(hr.sortedHashes) == 0 {
return ""
}
hash := hashKey(key)
idx := hr.search(hash)
return hr.vnodeToServer[hr.sortedHashes[idx]]
}
// GetServers returns a sorted copy of all servers in the ring.
func (hr *HashRing) GetServers() []pb.ServerAddress {
hr.mu.RLock()
defer hr.mu.RUnlock()
servers := make([]pb.ServerAddress, 0, len(hr.servers))
for s := range hr.servers {
servers = append(servers, s)
}
sort.Slice(servers, func(i, j int) bool {
return servers[i] < servers[j]
})
return servers
}
// ServerCount returns the number of servers in the ring.
func (hr *HashRing) ServerCount() int {
hr.mu.RLock()
defer hr.mu.RUnlock()
return len(hr.servers)
}
// rebuildRing rebuilds the sorted hash ring from the current server set.
// Caller must hold hr.mu write lock.
func (hr *HashRing) rebuildRing() {
hr.vnodeToServer = make(map[uint32]pb.ServerAddress, len(hr.servers)*hr.vnodeCount)
hr.sortedHashes = make([]uint32, 0, len(hr.servers)*hr.vnodeCount)
// Sort so a vnode-hash collision resolves to the same server on every node;
// map iteration order alone is randomized per process.
servers := make([]pb.ServerAddress, 0, len(hr.servers))
for server := range hr.servers {
servers = append(servers, server)
}
sort.Slice(servers, func(i, j int) bool { return servers[i] < servers[j] })
for _, server := range servers {
for i := 0; i < hr.vnodeCount; i++ {
vnodeKey := vnodeKeyFor(server, i)
hash := hashKey(vnodeKey)
hr.vnodeToServer[hash] = server
hr.sortedHashes = append(hr.sortedHashes, hash)
}
}
sort.Slice(hr.sortedHashes, func(i, j int) bool {
return hr.sortedHashes[i] < hr.sortedHashes[j]
})
}
// search finds the first ring position >= hash.
func (hr *HashRing) search(hash uint32) int {
idx := sort.Search(len(hr.sortedHashes), func(i int) bool {
return hr.sortedHashes[i] >= hash
})
if idx >= len(hr.sortedHashes) {
idx = 0 // wrap around
}
return idx
}
func hashKey(key string) uint32 {
return crc32.ChecksumIEEE([]byte(key))
}
func vnodeKeyFor(server pb.ServerAddress, index int) string {
// Use a format that distributes well across the ring
buf := make([]byte, 0, len(server)+10)
buf = append(buf, []byte(server)...)
buf = append(buf, '#')
buf = appendInt(buf, index)
return string(buf)
}
func appendInt(buf []byte, n int) []byte {
if n == 0 {
return append(buf, '0')
}
// Simple int-to-string without importing strconv
digits := [20]byte{}
pos := len(digits)
for n > 0 {
pos--
digits[pos] = byte('0' + n%10)
n /= 10
}
return append(buf, digits[pos:]...)
}