»replication Stanza
The replication stanza specifies various parameters for tuning replication related values.
»replication Parameters
resolver_discover_servers(boolean: true)- This feature should probably always be turned on. It enables secondary cluster nodes to reach out to nodes in the primary cluster to request information on who the active node is. This fixes an issue where a secondary cluster may only know about a single node in the primary cluster, and if that node isn't the active node, the replication connection will not succeed. With resolving disabled, the only way around this is to go through an onerous UpdatePrimary workflow, or do step-downs on the primary cluster until the one node known by the secondary cluster is active.logshipper_buffer_length(integer: 16384)- The maximum number of entries that the log shipper buffer can hold. If this is not specified, a default of 16K will be used.logshipper_buffer_size(string: "")- The maximum size that the log shipper buffer can be. This size can be expressed as an integer, indicating the number of bytes, or as a capacity string. Capacity strings are a number, followed by an optional space, followed by a suffix indicating the capacity. Valid suffixes are:kb, kib, mb, mib, gb, gib, tb, tib, which correspond to kilobytes, kibibytes, megabytes, mebibytes, etc. Full details on multiple-byte units in both powers of 2 and 10 can be found on Wikipedia. The suffixes are not case-sensitive. If this is not specified, Vault will attempt to read the total amount of memory available on the host it's running on. If that succeeds, Vault will use 10% of the value found. If Vault is unable to read the amount of host memory, this defaults to 1GB.allow_forwarding_via_header(boolean: flase)- Enable forwarding options for client controlled consistency, i.e.X-Vault-Inconsistent: forward-active-nodeandX-Vault-Forward: active-node.best_effort_wal_wait_duration(duration: "2s")- Legacy mitigation to try to prevent stale reads: when a write is made via RPC, i.e. the node handling the request didn't do the write itself, wait up to this much time to see the resulting WAL present locally before returning a response to the client.
