package LibLogCabin.Raft.Protocol

Get desktop application:
View/edit binary Protocol Buffers messages

* AppendEntries RPC: replicate log entries to a follower.

(message has no fields)

• required uint64 server_id = 1

  * ID of leader (caller), so the follower can redirect clients.

• required uint64 term = 2

  * Caller's term.

• required uint64 prev_log_index = 3

  * ID of entry preceding entries, or ID of last log entry for heartbeat.

• required uint64 prev_log_term = 4

  * Term of prev_log_index.

• repeated Entry entries = 5

  * Log entries for follower to store, or empty for heartbeat.

• required uint64 commit_index = 6

  * Last committed entry that the follower has, so the follower can advance its state machine.

• required uint64 term = 1

  * Callee's term, for the caller to update itself.

• required bool success = 2

  * True if new entries were added to the log.

• optional uint64 last_log_index = 3

  * The recipient's last log index (after it's applied this RPC's changes to the log). This is used to speed up finding the correct value for nextIndex with a follower that is far behind the leader.

• optional Response.ServerCapabilities server_capabilities = 4

  * Sent back to inform leader of what code the recipient is running.

Used in: Response

• optional uint32 min_supported_state_machine_version = 1

  * The server's state machine can behave like all state machine versions between min_supported_state_machine_version and max_supported_state_machine_version, inclusive.

• optional uint32 max_supported_state_machine_version = 2

  * The server's state machine can behave like all state machine versions between min_supported_state_machine_version and max_supported_state_machine_version, inclusive.

* A stable or transitional configuration.

Used in: Entry, Storage.SnapshotMetadata.Header

• required SimpleConfiguration prev_configuration = 1

  * The servers in a stable configuration, or the old servers in a transitional configuration.

• optional SimpleConfiguration next_configuration = 2

  * Not present in a stable configuration, or the new servers in a transitional configuration.

* Log entry.

Used in: AppendEntries.Request

• required uint64 term = 1

  * The term in which the entry was first created.

• optional uint64 index = 5

  * The index for the entry. It's truly optional (if you receive an entry over the network, you shouldn't rely on this). It's used for followers to check that they've got the entry at the right index and also used internally in some storage backends such as SegmentedLog.

• required uint64 cluster_time = 6

  * Roughly the number of nanoseconds the cluster has had a leader. This gets passed onto state machines, where it's used for cleaning up client sessions. See ClusterClock for more details.

• optional EntryType type = 2

  * See EntryType.

  the following are mutually exclusive, depending on 'type'

• optional Configuration configuration = 3

  * A Configuration to be used internally by the consensus module.

• optional bytes data = 4

  * A command to be processed by the state machine.

* The type of payload stored in a log entry.

Used in: Entry

• UNKNOWN = 0

  * This must be the first value in the enum and is never assigned explicitly. If new values are added, old code will see them as set and equal to this value (though they will still serialize to the correct value).

• CONFIGURATION = 1

  * Stores a 'Configuration', to be used internally by the consensus module.

• DATA = 2

  * Stores a command to be processed by the state machine.

• NOOP = 3

  * No op.

* The type of "service-specific error" replies that this service returns.

No errors are defined. enum Code { }; required Code error_code = 1;

(message has no fields)

* InstallSnapshot RPC: replicate part of a snapshot file to a follower.

(message has no fields)

• required uint64 server_id = 1

  * ID of leader (caller), so the follower can redirect clients.

• required uint64 term = 3

  * Caller's term.

• required uint64 last_snapshot_index = 4

  * The snapshot covers log entries in the range [1, lastSnapshotIndex]. While this information can be found in the snapshot itself, a follower may want to know sooner (for example, to name the snapshot file or to assert that it's not overwriting something more important).

• required uint64 byte_offset = 5

  * The byte offset where 'data' belongs in the file. Followers can expect this to grow without gaps, but they should use this to drop duplicate request messages.

• required bytes data = 6

  * Raw bytes of the snapshot file. This should be big enough to achieve reasonable throughput without having to pipeline RPCs.

• required bool done = 7

  * Set to true if this is the last chunk of the file and the follower should now load the contents; false otherwise.

• optional uint32 version = 8

  * Explains which version of this RPC the leader (caller) supports. - Servers speaking version 1 of this RPC (corresponding to LogCabin release v1.0.0 and up to a few weeks after) did not set this field. - Version 2 introduced the bytes_stored field in responses. Before this, leaders assumed that InstallSnapshot always succeeded if the term matched.

• required uint64 term = 1

  * Callee's term, for the caller to update itself.

• optional uint64 bytes_stored = 2

  * The total number of bytes in the snapshot that the follower has stored (after applying the request). This was introduced to fix https://github.com/logcabin/logcabin/issues/174 . Before, followers could only receive snapshot chunks in sequence. If they restarted, they'd have discarded their previous chunks and they would just repeatedly PANIC. This is now used to signal to the leader how many bytes the follower actually has saved. The leader considers the snapshot transfer complete when bytes_stored equals the full size of the snapshot. The leader should use bytes_stored as the value for byte_offset in the next request (most importantly, when a follower reboots, it returns 0 here and the leader starts at offset 0 in the next request). Leaders that do not support InstallSnapshot version 2 entirely ignore this field.

* Identifies which RPC is being executed.

• REQUEST_VOTE = 1

  Keep these numbers dense.

• APPEND_ENTRIES = 2

• INSTALL_SNAPSHOT = 3

* RequestVote RPC: ask a server for its vote in an election and/or get a server's log information.

(message has no fields)

• required uint64 server_id = 1

  * ID of the caller, so that if it re-requests the vote from the callee, the server will respond granted.

• required uint64 term = 2

  * Caller's term.

• required uint64 last_log_term = 3

  * Term of last entry in caller's log. Used to compare log completeness.

• required uint64 last_log_index = 4

  * ID of last entry in caller's log. Used to compare log completeness.

• required uint64 term = 1

  * Callee's term, for caller to update itself.

• required bool granted = 2

  * True if the follower granted the candidate its vote, false otherwise.

• optional bool log_ok = 3

  * Set to true if the caller's log is as up-to-date as the recipient's. This isn't presently used, but it's easy to return and might be useful later. In the Pre-Vote extension to Raft described in Section 9.6 "Preventing disruptions when a server rejoins the cluster" of Diego Ongaro's PhD dissertation, a server needs to know if its log is sufficient to get elected before actually starting an election.

* A server in a configuration.

Used in: SimpleConfiguration

• required uint64 server_id = 1

  * The ID of the server.

• required string addresses = 2

  * The network address(es) of the server (comma-delimited).

* A stable configuraton, in which a simple majority constitutes a quorum.

Used in: Configuration

• repeated Server servers = 1