storage_proxy: improve run_fenceable_write comment

This commit is contained in:
Petr Gusev
2025-10-28 16:58:35 +01:00
parent 58d100a0cb
commit c60223f009

View File

@@ -667,10 +667,12 @@ private:
// Returns fencing_token based on effective_replication_map.
static fencing_token get_fence(const locator::effective_replication_map& erm);
// Fencing tokens are checked twice: once before a replica-local storage operation
// and once after it completes. The second check ensures that if the coordinator
// was fenced out during execution, the replica does not count this operation
// toward the target CL.
// Fencing tokens are checked twice: once before performing a replica-local storage
// operation, and again after it completes. The second check ensures that if the
// request coordinator is fenced out while the replica is still executing the request — that is,
// when the topology coordinator distributes a new topology version across the cluster
// and increments the replicas fence version — the request coordinator does not count this
// operation toward the target consistency level (CL).
//
// For writes, this creates a cleanup concern: if the coordinator is fenced out
// mid-operation, the writes partial effects must be cleaned up before the range