Files
seaweedFS/weed/cluster/lock_manager/distributed_lock_manager_test.go
Chris Lu 75a6a34528 dlm: resilient distributed locks via consistent hashing + backup replication (#8860)
* dlm: replace modulo hashing with consistent hash ring

Introduce HashRing with virtual nodes (CRC32-based consistent hashing)
to replace the modulo-based hashKeyToServer. When a filer node is
removed, only keys that hashed to that node are remapped to the next
server on the ring, leaving all other mappings stable. This is the
foundation for backup replication — the successor on the ring is
always the natural takeover node.

* dlm: add Generation and IsBackup fields to Lock

Lock now carries IsBackup (whether this node holds the lock as a backup
replica) and Generation (a monotonic fencing token that increments on
each fresh acquisition, stays the same on renewal). Add helper methods:
AllLocks, PromoteLock, DemoteLock, InsertBackupLock, RemoveLock, GetLock.

* dlm: add ReplicateLock RPC and generation/is_backup proto fields

Add generation field to LockResponse for fencing tokens.
Add generation and is_backup fields to Lock message.
Add ReplicateLock RPC for primary-to-backup lock replication.
Add ReplicateLockRequest/ReplicateLockResponse messages.

* dlm: add async backup replication to DistributedLockManager

Route lock/unlock via consistent hash ring's GetPrimaryAndBackup().
After a successful lock or unlock on the primary, asynchronously
replicate the operation to the backup server via ReplicateFunc
callback. Single-server deployments skip replication.

* dlm: add ReplicateLock handler and backup-aware topology changes

Add ReplicateLock gRPC handler for primary-to-backup replication.
Revise OnDlmChangeSnapshot to handle three cases on topology change:
- Promote backup locks when this node becomes primary
- Demote primary locks when this node becomes backup
- Transfer locks when this node is neither primary nor backup
Wire up SetupDlmReplication during filer server initialization.

* dlm: expose generation fencing token in lock client

LiveLock now captures the generation from LockResponse and exposes it
via Generation() method. Consumers can use this as a fencing token to
detect stale lock holders.

* dlm: update empty folder cleaner to use consistent hash ring

Replace local modulo-based hashKeyToServer with LockRing.GetPrimary()
which uses the shared consistent hash ring for folder ownership.

* dlm: add unit tests for consistent hash ring

Test basic operations, consistency on server removal (only keys from
removed server move), backup-is-successor property (backup becomes
new primary when primary is removed), and key distribution balance.

* dlm: add integration tests for lock replication failure scenarios

Test cases:
- Primary crash with backup promotion (backup has valid token)
- Backup crash with primary continuing
- Both primary and backup crash (lock lost, re-acquirable)
- Rolling restart across all nodes
- Generation fencing token increments on new acquisition
- Replication failure (primary still works independently)
- Unlock replicates deletion to backup
- Lock survives server addition (topology change)
- Consistent hashing minimal disruption (only removed server's keys move)

* dlm: address PR review findings

1. Causal replication ordering: Add per-lock sequence number (Seq) that
   increments on every mutation. Backup rejects incoming mutations with
   seq <= current seq, preventing stale async replications from
   overwriting newer state. Unlock replication also carries seq and is
   rejected if stale.

2. Demote-after-handoff: OnDlmChangeSnapshot now transfers the lock to
   the new primary first and only demotes to backup after a successful
   TransferLocks RPC. If the transfer fails, the lock stays as primary
   on this node.

3. SetSnapshot candidateServers leak: Replace the candidateServers map
   entirely instead of appending, so removed servers don't linger.

4. TransferLocks preserves Generation and Seq: InsertLock now accepts
   generation and seq parameters. After accepting a transferred lock,
   the receiving node re-replicates to its backup.

5. Rolling restart test: Add re-replication step after promotion and
   assert survivedCount > 0. Add TestDLM_StaleReplicationRejected.

6. Mixed-version upgrade note: Add comment on HashRing documenting that
   all filer nodes must be upgraded together.

* dlm: serve renewals locally during transfer window on node join

When a new node joins and steals hash ranges from surviving nodes,
there's a window between ring update and lock transfer where the
client gets redirected to a node that doesn't have the lock yet.

Fix: if the ring says primary != self but we still hold the lock
locally (non-backup, matching token), serve the renewal/unlock here
rather than redirecting. The lock will be transferred by
OnDlmChangeSnapshot, and subsequent requests will go to the new
primary once the transfer completes.

Add tests:
- TestDLM_NodeDropAndJoin_OwnershipDisruption: measures disruption
  when a node drops and a new one joins (14/100 surviving-node locks
  disrupted, all handled by transfer logic)
- TestDLM_RenewalDuringTransferWindow: verifies renewal succeeds on
  old primary during the transfer window

* dlm: master-managed lock ring with stabilization batching

The master now owns the lock ring membership. Instead of filers
independently reacting to individual ClusterNodeUpdate add/remove
events, the master:

1. Tracks filer membership in LockRingManager
2. Batches rapid changes with a 1-second stabilization timer
   (e.g., a node drop + join within 1 second → single ring update)
3. Broadcasts the complete ring snapshot atomically via the new
   LockRingUpdate message in KeepConnectedResponse

Filers receive the ring as a complete snapshot and apply it via
SetSnapshot, ensuring all filers converge to the same ring state
without intermediate churn.

This eliminates the double-churn problem where a rapid drop+join
would fire two separate ring mutations, each triggering lock
transfers and disrupting ownership on surviving nodes.

* dlm: track ring version, reject stale updates, remove dead code

SetSnapshot now takes a version parameter from the master. Stale
updates (version < current) are rejected, preventing reordered
messages from overwriting a newer ring state. Version 0 is always
accepted for bootstrap.

Remove AddServer/RemoveServer from LockRing — the ring is now
exclusively managed by the master via SetSnapshot. Remove the
candidateServers map that was only used by those methods.

* dlm: fix SelectLocks data race, advance generation on backup insert

- SelectLocks: change RLock to Lock since the function deletes map
  entries, which is a write operation and causes a data race under RLock.
- InsertBackupLock: advance nextGeneration to at least the incoming
  generation so that after failover promotion, new lock acquisitions
  get a generation strictly greater than any replicated lock.
- Bump replication failure log from V(1) to Warningf for production
  visibility.

* dlm: fix SetSnapshot race, test reliability, timer edge cases

- SetSnapshot: hold LockRing lock through both version update and
  Ring.SetServers() so they're atomic. Prevents a concurrent caller
  from seeing the new version but applying stale servers.
- Transfer window test: search for a key that actually moves primary
  when filer4 joins, instead of relying on a fixed key that may not.
- renewLock redirect: pass the existing token to the new primary
  instead of empty string, so redirected renewals work correctly.
- scheduleBroadcast: check timer.Stop() return value. If the timer
  already fired, the callback picks up latest state.
- FlushPending: only broadcast if timer.Stop() returns true (timer
  was still pending). If false, the callback is already running.
- Fix test comment: "idempotent" → "accepted, state-changing".

* dlm: use wall-clock nanoseconds for lock ring version

The lock ring version was an in-memory counter that reset to 0 on
master restart. A filer that had seen version 5 would reject version 1
from the restarted master.

Fix: use time.Now().UnixNano() as the version. This survives master
restarts without persistence — the restarted master produces a
version greater than any pre-restart value.

* dlm: treat expired lock owners as missing

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* dlm: reject stale lock transfers

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* dlm: order replication by generation

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* dlm: bootstrap lock ring on reconnect

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-03-30 23:29:56 -07:00

711 lines
24 KiB
Go

package lock_manager
import (
"fmt"
"sync"
"testing"
"time"
"github.com/seaweedfs/seaweedfs/weed/pb"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// testCluster simulates a cluster of filer nodes with DLMs.
// It wires up ReplicateFn so that replication calls arrive at the
// correct peer's DLM, enabling end-to-end backup testing without gRPC.
type testCluster struct {
mu sync.Mutex
nodes map[pb.ServerAddress]*DistributedLockManager
}
func newTestCluster(hosts ...pb.ServerAddress) *testCluster {
c := &testCluster{nodes: make(map[pb.ServerAddress]*DistributedLockManager)}
servers := make([]pb.ServerAddress, len(hosts))
copy(servers, hosts)
for _, host := range hosts {
dlm := NewDistributedLockManager(host)
dlm.LockRing.SetSnapshot(servers, 0)
c.nodes[host] = dlm
}
// Wire up replication: each node's ReplicateFn calls the backup's DLM directly
for _, dlm := range c.nodes {
d := dlm // capture
d.ReplicateFn = func(server pb.ServerAddress, key string, expiredAtNs int64, token string, owner string, generation int64, seq int64, isUnlock bool) {
c.mu.Lock()
target, ok := c.nodes[server]
c.mu.Unlock()
if !ok {
return // server is down
}
if isUnlock {
target.RemoveBackupLockIfSeq(key, generation, seq)
} else {
target.InsertBackupLock(key, expiredAtNs, token, owner, generation, seq)
}
}
}
return c
}
func (c *testCluster) removeNode(host pb.ServerAddress) {
c.mu.Lock()
delete(c.nodes, host)
c.mu.Unlock()
// Update all remaining nodes' rings
remaining := c.getServers()
for _, dlm := range c.getNodes() {
dlm.LockRing.SetSnapshot(remaining, 0)
}
}
func (c *testCluster) addNode(host pb.ServerAddress) {
c.mu.Lock()
dlm := NewDistributedLockManager(host)
c.nodes[host] = dlm
c.mu.Unlock()
// Wire up replication
dlm.ReplicateFn = func(server pb.ServerAddress, key string, expiredAtNs int64, token string, owner string, generation int64, seq int64, isUnlock bool) {
c.mu.Lock()
target, ok := c.nodes[server]
c.mu.Unlock()
if !ok {
return
}
if isUnlock {
target.RemoveBackupLockIfSeq(key, generation, seq)
} else {
target.InsertBackupLock(key, expiredAtNs, token, owner, generation, seq)
}
}
servers := c.getServers()
for _, n := range c.getNodes() {
n.LockRing.SetSnapshot(servers, 0)
}
}
func (c *testCluster) getNodes() map[pb.ServerAddress]*DistributedLockManager {
c.mu.Lock()
defer c.mu.Unlock()
cp := make(map[pb.ServerAddress]*DistributedLockManager, len(c.nodes))
for k, v := range c.nodes {
cp[k] = v
}
return cp
}
func (c *testCluster) getServers() []pb.ServerAddress {
c.mu.Lock()
defer c.mu.Unlock()
var servers []pb.ServerAddress
for s := range c.nodes {
servers = append(servers, s)
}
return servers
}
func (c *testCluster) get(host pb.ServerAddress) *DistributedLockManager {
c.mu.Lock()
defer c.mu.Unlock()
return c.nodes[host]
}
// acquireLock tries to acquire a lock on the correct primary node.
// It follows redirects (movedTo) like a real client would.
func (c *testCluster) acquireLock(key, owner string, ttl time.Duration) (renewToken string, generation int64, primaryHost pb.ServerAddress, err error) {
// Try any node first (simulates client connecting to seed filer)
for _, dlm := range c.getNodes() {
expiry := time.Now().Add(ttl).UnixNano()
var movedTo pb.ServerAddress
var lockErr error
_, renewToken, generation, movedTo, lockErr = dlm.LockWithTimeout(key, expiry, "", owner)
if movedTo != "" && movedTo != dlm.Host {
// Follow redirect
target := c.get(movedTo)
if target == nil {
err = fmt.Errorf("primary %s is down", movedTo)
return
}
_, renewToken, generation, _, lockErr = target.LockWithTimeout(key, expiry, "", owner)
if lockErr != nil {
err = lockErr
return
}
primaryHost = movedTo
// Wait briefly for async replication to complete
time.Sleep(10 * time.Millisecond)
return
}
if lockErr != nil {
err = lockErr
return
}
primaryHost = dlm.Host
time.Sleep(10 * time.Millisecond)
return
}
err = fmt.Errorf("no nodes available")
return
}
// renewLock renews a lock on the primary node
func (c *testCluster) renewLock(key, owner, token string, ttl time.Duration, primaryHost pb.ServerAddress) (newToken string, generation int64, err error) {
target := c.get(primaryHost)
if target == nil {
err = fmt.Errorf("primary %s is down", primaryHost)
return
}
expiry := time.Now().Add(ttl).UnixNano()
var movedTo pb.ServerAddress
var lockErr error
_, newToken, generation, movedTo, lockErr = target.LockWithTimeout(key, expiry, token, owner)
if movedTo != "" && movedTo != primaryHost {
target = c.get(movedTo)
if target == nil {
err = fmt.Errorf("new primary %s is down", movedTo)
return
}
// Pass the existing token so the redirected renewal can match
// if the lock was already transferred to the new primary.
_, newToken, generation, _, lockErr = target.LockWithTimeout(key, expiry, token, owner)
}
err = lockErr
time.Sleep(10 * time.Millisecond)
return
}
// --- Test Cases ---
func TestDLM_PrimaryCrash_BackupPromotes(t *testing.T) {
// Scenario: Lock is acquired, primary crashes, backup should have the lock
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
key := "test-lock-primary-crash"
renewToken, _, primaryHost, err := cluster.acquireLock(key, "owner1", 30*time.Second)
require.NoError(t, err)
require.NotEmpty(t, renewToken)
// Find the backup for this key
_, backup := cluster.get(primaryHost).LockRing.GetPrimaryAndBackup(key)
require.NotEmpty(t, backup, "should have a backup server")
// Verify backup has the lock
backupDlm := cluster.get(backup)
backupLock, found := backupDlm.GetLock(key)
require.True(t, found, "backup should have the lock")
assert.True(t, backupLock.IsBackup, "lock on backup should be marked as backup")
assert.Equal(t, renewToken, backupLock.Token, "backup should have the same token")
// Crash the primary
cluster.removeNode(primaryHost)
// Simulate topology change: promote backup locks
for _, dlm := range cluster.getNodes() {
locks := dlm.AllLocks()
for _, lock := range locks {
newPrimary, _ := dlm.LockRing.GetPrimaryAndBackup(lock.Key)
if newPrimary == dlm.Host && lock.IsBackup {
dlm.PromoteLock(lock.Key)
}
}
}
// The backup should now be the primary
newPrimary := backupDlm.LockRing.GetPrimary(key)
assert.Equal(t, backup, newPrimary, "backup should be the new primary")
// The promoted lock should work — verify it's no longer a backup
promotedLock, found := backupDlm.GetLock(key)
require.True(t, found, "lock should still exist after promotion")
assert.False(t, promotedLock.IsBackup, "lock should be promoted to primary")
// Client should be able to renew with the same token on the new primary
newToken, _, err := cluster.renewLock(key, "owner1", renewToken, 30*time.Second, backup)
require.NoError(t, err)
assert.NotEmpty(t, newToken, "renewal on new primary should succeed")
}
func TestDLM_BackupCrash_PrimaryContinues(t *testing.T) {
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
key := "test-lock-backup-crash"
renewToken, _, primaryHost, err := cluster.acquireLock(key, "owner1", 30*time.Second)
require.NoError(t, err)
_, backup := cluster.get(primaryHost).LockRing.GetPrimaryAndBackup(key)
// Crash the backup
cluster.removeNode(backup)
// Primary should still work — renew the lock
newToken, _, err := cluster.renewLock(key, "owner1", renewToken, 30*time.Second, primaryHost)
require.NoError(t, err)
assert.NotEmpty(t, newToken, "primary should continue working after backup crash")
// Verify primary is still the primary for this key
newPrimary := cluster.get(primaryHost).LockRing.GetPrimary(key)
assert.Equal(t, primaryHost, newPrimary)
}
func TestDLM_BothPrimaryAndBackupCrash(t *testing.T) {
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
key := "test-lock-both-crash"
_, _, primaryHost, err := cluster.acquireLock(key, "owner1", 30*time.Second)
require.NoError(t, err)
_, backup := cluster.get(primaryHost).LockRing.GetPrimaryAndBackup(key)
// Crash both
cluster.removeNode(primaryHost)
cluster.removeNode(backup)
// The lock is lost — the surviving node should be able to acquire it fresh
newToken, _, _, err := cluster.acquireLock(key, "owner2", 30*time.Second)
require.NoError(t, err)
assert.NotEmpty(t, newToken, "new owner should acquire lock after both crash")
}
func TestDLM_RollingRestart(t *testing.T) {
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
// Acquire multiple locks
type lockState struct {
key, owner, token string
generation int64
primary pb.ServerAddress
}
locks := make([]lockState, 5)
for i := range locks {
key := fmt.Sprintf("rolling-lock-%d", i)
token, gen, primary, err := cluster.acquireLock(key, fmt.Sprintf("owner-%d", i), 30*time.Second)
require.NoError(t, err)
locks[i] = lockState{key: key, owner: fmt.Sprintf("owner-%d", i), token: token, generation: gen, primary: primary}
}
// Rolling restart: remove and re-add each node one at a time.
// After removing a node, promote backups and re-replicate to new backups
// to maintain the invariant that each lock has a backup copy.
for _, host := range hosts {
cluster.removeNode(host)
// Simulate full OnDlmChangeSnapshot: promote backups and re-replicate
for _, dlm := range cluster.getNodes() {
for _, lock := range dlm.AllLocks() {
newPrimary, _ := dlm.LockRing.GetPrimaryAndBackup(lock.Key)
if newPrimary == dlm.Host && lock.IsBackup {
dlm.PromoteLock(lock.Key)
}
}
// Re-replicate all primary locks to their new backups
for _, lock := range dlm.AllLocks() {
newPrimary, _ := dlm.LockRing.GetPrimaryAndBackup(lock.Key)
if newPrimary == dlm.Host && !lock.IsBackup {
dlm.replicateToBackup(lock.Key, lock.ExpiredAtNs, lock.Token, lock.Owner, lock.Generation, lock.Seq, false)
}
}
}
time.Sleep(10 * time.Millisecond)
// Re-add the node
cluster.addNode(host)
time.Sleep(10 * time.Millisecond)
}
// After rolling restart, locks should survive via backup promotion
survivedCount := 0
for _, ls := range locks {
for _, dlm := range cluster.getNodes() {
lock, found := dlm.GetLock(ls.key)
if found && !lock.IsBackup {
survivedCount++
break
}
}
}
t.Logf("Locks survived rolling restart: %d / %d", survivedCount, len(locks))
require.Greater(t, survivedCount, 0, "at least some locks should survive a rolling restart via backup promotion")
}
func TestDLM_GenerationIncrementsOnNewAcquisition(t *testing.T) {
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888"}
cluster := newTestCluster(hosts...)
key := "gen-test-lock"
// Acquire lock — generation should be > 0
token1, gen1, primary, err := cluster.acquireLock(key, "owner1", 2*time.Second)
require.NoError(t, err)
assert.Greater(t, gen1, int64(0))
// Renew — generation should stay the same
token2, gen2, err := cluster.renewLock(key, "owner1", token1, 2*time.Second, primary)
require.NoError(t, err)
assert.Equal(t, gen1, gen2, "generation should not change on renewal")
// Let lock expire
time.Sleep(3 * time.Second)
// Re-acquire — generation should increment
_, gen3, _, err := cluster.acquireLock(key, "owner2", 30*time.Second)
require.NoError(t, err)
assert.Greater(t, gen3, gen1, "generation should increment on new acquisition")
_ = token2
}
func TestDLM_ReplicationFailure_PrimaryStillWorks(t *testing.T) {
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
// Break replication by setting a no-op ReplicateFn on all nodes
for _, dlm := range cluster.getNodes() {
dlm.ReplicateFn = func(server pb.ServerAddress, key string, expiredAtNs int64, token string, owner string, generation int64, seq int64, isUnlock bool) {
// Simulate replication failure: do nothing
}
}
key := "repl-fail-lock"
renewToken, _, primaryHost, err := cluster.acquireLock(key, "owner1", 30*time.Second)
require.NoError(t, err)
// Primary should have the lock
primaryDlm := cluster.get(primaryHost)
lock, found := primaryDlm.GetLock(key)
require.True(t, found, "primary should have the lock")
assert.False(t, lock.IsBackup)
// Backup should NOT have it (replication failed)
_, backup := primaryDlm.LockRing.GetPrimaryAndBackup(key)
backupDlm := cluster.get(backup)
_, found = backupDlm.GetLock(key)
assert.False(t, found, "backup should not have the lock when replication fails")
// Primary should still be able to renew
newToken, _, err := cluster.renewLock(key, "owner1", renewToken, 30*time.Second, primaryHost)
require.NoError(t, err)
assert.NotEmpty(t, newToken)
}
func TestDLM_UnlockReplicatesToBackup(t *testing.T) {
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888"}
cluster := newTestCluster(hosts...)
key := "unlock-repl-lock"
renewToken, _, primaryHost, err := cluster.acquireLock(key, "owner1", 30*time.Second)
require.NoError(t, err)
_, backup := cluster.get(primaryHost).LockRing.GetPrimaryAndBackup(key)
// Verify backup has the lock
_, found := cluster.get(backup).GetLock(key)
require.True(t, found, "backup should have the lock")
// Unlock on primary
primaryDlm := cluster.get(primaryHost)
movedTo, err := primaryDlm.Unlock(key, renewToken)
require.NoError(t, err)
assert.Empty(t, movedTo)
// Wait for async replication
time.Sleep(20 * time.Millisecond)
// Backup should also have removed the lock
_, found = cluster.get(backup).GetLock(key)
assert.False(t, found, "backup should remove lock after unlock replication")
}
func TestDLM_TopologyChange_LockSurvivesServerAddition(t *testing.T) {
// Start with 2 servers, acquire lock, add a 3rd server
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888"}
cluster := newTestCluster(hosts...)
key := "topo-add-lock"
renewToken, _, primaryHost, err := cluster.acquireLock(key, "owner1", 30*time.Second)
require.NoError(t, err)
// Add a new server
cluster.addNode("filer3:8888")
time.Sleep(20 * time.Millisecond)
// The lock should still be accessible — either the same primary or on a new one
// Try to renew on the original primary first
newPrimary := cluster.get(primaryHost).LockRing.GetPrimary(key)
if newPrimary == primaryHost {
// Still on same primary
newToken, _, err := cluster.renewLock(key, "owner1", renewToken, 30*time.Second, primaryHost)
require.NoError(t, err)
assert.NotEmpty(t, newToken)
}
// If primary changed, the lock may need transfer — that's handled by OnDlmChangeSnapshot
// which is tested at the server level
}
func TestDLM_ConsistentHashing_MinimalDisruption(t *testing.T) {
// Verify that removing a server only affects locks on that server
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
// Acquire 50 locks
type lockInfo struct {
key, token string
primary pb.ServerAddress
}
locks := make([]lockInfo, 50)
for i := range locks {
key := fmt.Sprintf("min-disrupt-%d", i)
token, _, primary, err := cluster.acquireLock(key, "owner", 30*time.Second)
require.NoError(t, err)
locks[i] = lockInfo{key: key, token: token, primary: primary}
}
// Count locks per server before removal
countBefore := make(map[pb.ServerAddress]int)
for _, l := range locks {
countBefore[l.primary]++
}
t.Logf("Lock distribution before: %v", countBefore)
// Remove filer2
cluster.removeNode("filer2:8888")
// Count how many locks changed primary
changed := 0
for _, l := range locks {
// Check where the lock should be now
for _, dlm := range cluster.getNodes() {
newPrimary := dlm.LockRing.GetPrimary(l.key)
if newPrimary != l.primary {
changed++
}
break
}
}
// Only locks from filer2 should have changed
assert.Equal(t, countBefore["filer2:8888"], changed,
"only locks from removed server should change primary")
}
func TestDLM_NodeDropAndJoin_OwnershipDisruption(t *testing.T) {
// Scenario: 3 nodes, acquire locks, one drops and a NEW node joins quickly.
// The new node steals hash ranges from surviving nodes, not just from the
// departed node. This test measures the disruption.
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
// Acquire many locks
numLocks := 100
type lockInfo struct {
key, token string
primary pb.ServerAddress
}
locks := make([]lockInfo, numLocks)
for i := range locks {
key := fmt.Sprintf("churn-lock-%d", i)
token, _, primary, err := cluster.acquireLock(key, "owner", 30*time.Second)
require.NoError(t, err)
locks[i] = lockInfo{key: key, token: token, primary: primary}
}
// Record primary for each lock before the change
beforePrimary := make(map[string]pb.ServerAddress)
for _, l := range locks {
beforePrimary[l.key] = l.primary
}
// Drop filer3 and immediately add filer4
cluster.removeNode("filer3:8888")
// Promote backups on remaining nodes (simulates OnDlmChangeSnapshot)
for _, dlm := range cluster.getNodes() {
for _, lock := range dlm.AllLocks() {
p, _ := dlm.LockRing.GetPrimaryAndBackup(lock.Key)
if p == dlm.Host && lock.IsBackup {
dlm.PromoteLock(lock.Key)
}
}
// Re-replicate primary locks to new backups
for _, lock := range dlm.AllLocks() {
p, _ := dlm.LockRing.GetPrimaryAndBackup(lock.Key)
if p == dlm.Host && !lock.IsBackup {
dlm.replicateToBackup(lock.Key, lock.ExpiredAtNs, lock.Token, lock.Owner, lock.Generation, lock.Seq, false)
}
}
}
time.Sleep(10 * time.Millisecond)
// Now add filer4 (new node, empty)
cluster.addNode("filer4:8888")
time.Sleep(10 * time.Millisecond)
// Simulate OnDlmChangeSnapshot on all nodes after filer4 joins:
// transfer locks that now belong to filer4
for host, dlm := range cluster.getNodes() {
for _, lock := range dlm.AllLocks() {
p, _ := dlm.LockRing.GetPrimaryAndBackup(lock.Key)
if p != host && !lock.IsBackup {
// This lock should move to the new primary
target := cluster.get(p)
if target != nil {
target.InsertLock(lock.Key, lock.ExpiredAtNs, lock.Token, lock.Owner, lock.Generation, lock.Seq)
dlm.DemoteLock(lock.Key)
}
}
}
}
time.Sleep(10 * time.Millisecond)
// Count disruptions: locks whose primary changed to a node other than filer3's successor
disruptedFromSurvivors := 0
disruptedFromDeparted := 0
movedToFiler4 := 0
for _, l := range locks {
// What's the new primary?
var newPrimary pb.ServerAddress
for _, dlm := range cluster.getNodes() {
newPrimary = dlm.LockRing.GetPrimary(l.key)
break
}
oldPrimary := beforePrimary[l.key]
if newPrimary != oldPrimary {
if oldPrimary == "filer3:8888" {
disruptedFromDeparted++
} else {
disruptedFromSurvivors++
}
}
if newPrimary == "filer4:8888" {
movedToFiler4++
}
}
t.Logf("Locks disrupted from departed filer3: %d / %d", disruptedFromDeparted, numLocks)
t.Logf("Locks disrupted from surviving filer1/filer2: %d / %d", disruptedFromSurvivors, numLocks)
t.Logf("Locks now on new filer4: %d / %d", movedToFiler4, numLocks)
// The key concern: filer4 joining disrupts locks on surviving nodes
// With consistent hashing, new node steals ~1/N of each surviving node's keys
// Verify that the transfer logic above moved those locks to filer4
for _, l := range locks {
var newPrimary pb.ServerAddress
for _, dlm := range cluster.getNodes() {
newPrimary = dlm.LockRing.GetPrimary(l.key)
break
}
target := cluster.get(newPrimary)
require.NotNil(t, target, "primary %s should exist", newPrimary)
lock, found := target.GetLock(l.key)
if !found {
// Lock may have only a backup copy if transfer happened but
// the lock was on the departed node and wasn't re-replicated.
// Check all nodes for any copy.
anyFound := false
for _, dlm := range cluster.getNodes() {
if _, f := dlm.GetLock(l.key); f {
anyFound = true
break
}
}
if !anyFound {
t.Errorf("lock %s completely lost (primary should be %s)", l.key, newPrimary)
}
continue
}
assert.False(t, lock.IsBackup, "lock %s on primary %s should not be a backup", l.key, newPrimary)
}
}
func TestDLM_RenewalDuringTransferWindow(t *testing.T) {
// When a new node joins and steals a key range from a surviving node,
// there's a window between ring update and lock transfer. During this
// window, a client renewal should still succeed on the old primary
// (because it still holds the lock locally).
hosts := []pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888"}
cluster := newTestCluster(hosts...)
// Find a key that will move primary when filer4 is added.
// Try candidate keys until we find one whose primary changes.
var key, renewToken string
var primaryHost pb.ServerAddress
for i := 0; i < 1000; i++ {
candidate := fmt.Sprintf("transfer-window-lock-%d", i)
token, _, primary, err := cluster.acquireLock(candidate, "owner1", 30*time.Second)
require.NoError(t, err)
// Check if adding filer4 would move this key's primary
tmpRing := NewHashRing(DefaultVnodeCount)
tmpRing.SetServers([]pb.ServerAddress{"filer1:8888", "filer2:8888", "filer3:8888", "filer4:8888"})
newPrimary := tmpRing.GetPrimary(candidate)
if newPrimary != primary {
key = candidate
renewToken = token
primaryHost = primary
break
}
}
require.NotEmpty(t, key, "should find a key that moves primary when filer4 joins")
// Add filer4 — this changes the primary for our key per the ring
cluster.addNode("filer4:8888")
time.Sleep(10 * time.Millisecond)
newPrimary := cluster.get(primaryHost).LockRing.GetPrimary(key)
require.NotEqual(t, primaryHost, newPrimary, "key should have moved to a different primary")
// Renewal on the OLD primary should still succeed because it holds the lock locally
newToken, _, err := cluster.renewLock(key, "owner1", renewToken, 30*time.Second, primaryHost)
require.NoError(t, err, "renewal on old primary should succeed during transfer window")
assert.NotEmpty(t, newToken, "should get a new token from old primary")
t.Logf("Key %s: primary changed from %s to %s, but renewal on old primary succeeded", key, primaryHost, newPrimary)
}
func TestDLM_StaleReplicationRejected(t *testing.T) {
// Verify that a stale replication (lower seq) does not overwrite a newer one
lm := NewLockManager()
// Insert backup with seq=3
lm.InsertBackupLock("key1", time.Now().Add(30*time.Second).UnixNano(), "token-new", "owner1", 1, 3)
lock, found := lm.GetLock("key1")
require.True(t, found)
assert.Equal(t, "token-new", lock.Token)
assert.Equal(t, int64(3), lock.Seq)
// Try to overwrite with stale seq=2 — should be rejected
lm.InsertBackupLock("key1", time.Now().Add(30*time.Second).UnixNano(), "token-old", "owner1", 1, 2)
lock, found = lm.GetLock("key1")
require.True(t, found)
assert.Equal(t, "token-new", lock.Token, "stale replication should be rejected")
assert.Equal(t, int64(3), lock.Seq)
// Update with higher seq=4 — should succeed
lm.InsertBackupLock("key1", time.Now().Add(30*time.Second).UnixNano(), "token-newer", "owner1", 1, 4)
lock, found = lm.GetLock("key1")
require.True(t, found)
assert.Equal(t, "token-newer", lock.Token, "newer replication should be accepted")
assert.Equal(t, int64(4), lock.Seq)
// Stale unlock (seq=2) should not delete the lock
removed := lm.RemoveBackupLockIfSeq("key1", 1, 2)
assert.False(t, removed, "stale unlock should be rejected")
_, found = lm.GetLock("key1")
assert.True(t, found, "lock should still exist after stale unlock")
// Valid unlock (seq=5) should delete
removed = lm.RemoveBackupLockIfSeq("key1", 1, 5)
assert.True(t, removed, "valid unlock should be accepted")
_, found = lm.GetLock("key1")
assert.False(t, found, "lock should be removed after valid unlock")
}