filer: serialize same-path mutations with a local lock

CreateEntry is a FindEntry-then-write with no lock, so concurrent creates to
the same path race: OExcl can admit two creators, and a conditional write has
no atomic check-then-act. Add a per-path exclusive lock (util.LockTable, which
evicts idle keys so it stays bounded) in the CreateEntry handler so the read
and the write are atomic on this filer. Once callers route a key's writes to
its owner filer, this local lock is the authoritative serialization point.

AppendToEntry moves from the distributed lock to the same per-path lock.
This commit is contained in:
Chris Lu committed 2026-05-22 22:24:16 -07:00
1 parent d1665750e1
commit 56c51e7f50
4 files changed
+83 -5

No files matched your search

+8
View File
@@ -124,6 +124,13 @@ type FilerServer struct {
// mountPeerRegistry backs the MountRegister / MountList RPCs for peer
// chunk sharing (tier 1). Always populated.
mountPeerRegistry *filer.MountPeerRegistry
// entryLockTable serializes mutations to the same entry path on this filer.
// It is the local serialization point for read-modify-write operations
// (conditional create, append) once writers for a key are routed to this
// node, replacing the distributed lock for that purpose. Idle keys are
// evicted automatically, so the table stays bounded.
entryLockTable *util.LockTable[util.FullPath]
}
func NewFilerServer(defaultMux, readonlyMux *http.ServeMux, option *FilerOption) (fs *FilerServer, err error) {
@@ -162,6 +169,7 @@ func NewFilerServer(defaultMux, readonlyMux *http.ServeMux, option *FilerOption)
inFlightDataLimitCond: sync.NewCond(new(sync.Mutex)),
recentCopyRequests: make(map[string]recentCopyRequest),
CredentialManager: option.CredentialManager,
entryLockTable: util.NewLockTable[util.FullPath](),
}
fs.mountPeerRegistry = filer.NewMountPeerRegistry()
go fs.runMountPeerRegistrySweeper()