package replication
Get desktop application:
View/edit binary Protocol Buffers messages
coordinator -> node
rpc AddFollower (, )
replication.proto:30message
replication.proto:149string namespace = 1
int64 shard = 2
int64 term = 3
string follower_name = 4
optional follower_head_entry_id = 5
bool observer = 6
If true, this follower is an observer that does not participate in the replication quorum. Used for shard split catch-up.
optional int64 target_shard = 7
For observer followers during shard split: the shard ID that the receiving server should use for the replicated data. This allows the parent leader to replicate data to a child shard on the same server without conflicting with the parent's own controllers.
optional split_hash_range = 8
For observer followers during shard split: the hash range that the child shard should retain after receiving the parent's snapshot. The child filters its database and WAL entries to only keep keys whose hash falls within this range.
message
replication.proto:176(message has no fields)
rpc BecomeLeader (, )
replication.proto:29message
replication.proto:135string namespace = 1
int64 shard = 2
int64 term = 3
uint32 replication_factor = 4
map<string, > follower_maps = 5
repeated Feature features_supported = 6
features_supported contains the list of features that all members of the quorum support. This enables feature negotiation during leader election to support safe rolling upgrades where new and old nodes can coexist.
message
replication.proto:174(message has no fields)
rpc DeleteShard (, )
replication.proto:33message
replication.proto:213string namespace = 1
int64 shard = 2
int64 term = 3
message
replication.proto:219(message has no fields)
rpc FreezeShard (, )
replication.proto:48FreezeShard stops a leader from accepting new writes (or resumes it) without fencing it, so its observer/follower cursors keep streaming. Used during split cutover to quiesce the parent and let children drain the final tail before the parent is fenced.
message
replication.proto:233string namespace = 1
int64 shard = 2
int64 term = 3
bool frozen = 4
When true, the leader stops accepting new writes; when false, it resumes.
message
replication.proto:241int64 head_offset = 1
The leader's head offset at the moment it was frozen. While frozen the head no longer advances, so this is the final offset children must reach.
rpc GetInfo (, )
replication.proto:38Deprecated: use Handshake for node initialization and feature discovery. Keep GetInfo only so new coordinators can fall back to older dataservers during rolling upgrades, and remove it in the next major version.
message
replication.proto:66(message has no fields)
message
replication.proto:70repeated Feature features_supported = 1
rpc GetStatus (, )
replication.proto:32message
replication.proto:249int64 shard = 1
message
replication.proto:260int64 term = 1
ServingStatus status = 2
int64 head_offset = 3
int64 commit_offset = 4
optional shard_stats = 5
rpc Handshake (, )
replication.proto:34message
replication.proto:81string instance_id = 1
message
replication.proto:85HandshakeStatus status = 1
repeated Feature features_supported = 2
rpc NewTerm (, )
replication.proto:28message
replication.proto:123string namespace = 1
int64 shard = 2
int64 term = 3
optional options = 4
message
replication.proto:131optional head_entry_id = 1
rpc PushShardAssignments (stream , )
replication.proto:25message
replication.proto:90(message has no fields)
rpc RemoveObserver (, )
replication.proto:42Removes an observer follower from the parent leader without fencing. Used during split abort to cleanly stop data streaming to children.
message
replication.proto:223string namespace = 1
int64 shard = 2
parent shard ID
int64 term = 3
string follower_name = 4
int64 target_shard = 5
child shard ID — used to form the observer key
message
replication.proto:231(message has no fields)
node (leader) -> node (follower)
rpc Replicate (stream , stream )
replication.proto:54message
replication.proto:190int64 term = 1
optional entry = 2
int64 commit_offset = 3
optional uint32 previous_entry_crc = 4
bool cumulative_acks_supported = 5
The leader treats acks as cumulative (an ack confirms every entry up to its offset): the follower can then acknowledge a whole sync round with a single Ack message instead of one per entry. Followers must keep sending one ack per entry when the leader does not advertise the support.
message
replication.proto:205Acknowledges the entries up to `offset` (when the leader advertises cumulative_acks_supported; a single entry otherwise).
int64 offset = 1
rpc SendSnapshot (stream , )
replication.proto:55message
replication.proto:104int64 term = 1
string name = 2
bytes content = 3
int32 chunk_index = 4
int32 chunk_count = 5
message
replication.proto:209int64 ack_offset = 1
rpc Truncate (, )
replication.proto:53message
replication.proto:178string namespace = 1
int64 shard = 2
int64 term = 3
optional head_entry_id = 4
message
replication.proto:186optional head_entry_id = 1
message
replication.proto:92Used in: , , , ,
int64 term = 1
int64 offset = 2
Feature represents a capability that can be negotiated during leader election. This enables safe rolling upgrades where new and old nodes can coexist.
Used in: , , , ,
FEATURE_UNKNOWN = 0
FEATURE_DB_CHECKSUM = 1
Used in:
HANDSHAKE_STATUS_UNKNOWN = 0
HANDSHAKE_STATUS_BOUND = 1
HANDSHAKE_STATUS_ALREADY_BOUND = 2
HANDSHAKE_STATUS_MISMATCH = 3
Used in:
UNKNOWN = 0
NATURAL = 1
HIERARCHICAL = 2
message
replication.proto:97Used in:
int64 term = 1
int64 offset = 2
bytes value = 3
fixed64 timestamp = 4
message
replication.proto:118Used in:
bool enable_notifications = 1
KeySortingType key_sorting = 2
Used in:
NOT_MEMBER = 0
FENCED = 1
FOLLOWER = 2
LEADER = 3
message
replication.proto:270Used in:
uint64 db_size_bytes = 1
uint64 read_ops_total = 2
uint64 write_ops_total = 3