feat(raft): InstallSnapshot wire protocol (backward-compatible)

This commit is contained in:
2026-07-31 02:12:51 +03:00
parent f416fe930e
commit cb9cd7415d
2 changed files with 129 additions and 0 deletions
+33
View File
@@ -83,6 +83,8 @@ type
rmkRequestVoteReply rmkRequestVoteReply
rmkAppendEntries rmkAppendEntries
rmkAppendEntriesReply rmkAppendEntriesReply
rmkInstallSnapshot
rmkInstallSnapshotReply
RaftMessage* = object RaftMessage* = object
kind*: RaftMessageKind kind*: RaftMessageKind
@@ -99,6 +101,12 @@ type
# Reply # Reply
success*: bool success*: bool
matchIdx*: uint64 matchIdx*: uint64
# InstallSnapshot (prevLogIndex/prevLogTerm reuse: snapshot base index/term;
# reply uses success/matchIdx as usual)
snapId*: uint64 # snapshot generation, matches leader's base at build time
snapOffset*: uint64 # byte offset of this chunk within the archive
snapData*: seq[byte] # chunk payload (<= snapChunkBytes)
snapDone*: bool # last chunk
RaftCluster* = ref object RaftCluster* = ref object
nodes*: Table[string, RaftNode] nodes*: Table[string, RaftNode]
@@ -694,6 +702,14 @@ proc serialize*(msg: RaftMessage): seq[byte] =
stream.write(msg.leaderCommit) stream.write(msg.leaderCommit)
stream.write(char(if msg.success: 1 else: 0)) stream.write(char(if msg.success: 1 else: 0))
stream.write(msg.matchIdx) stream.write(msg.matchIdx)
# InstallSnapshot trailing fields (appended for wire backward compatibility;
# pre-v1.3 peers stop reading at matchIdx and ignore these bytes)
stream.write(msg.snapId)
stream.write(msg.snapOffset)
stream.write(uint32(msg.snapData.len))
if msg.snapData.len > 0:
stream.writeData(addr msg.snapData[0], msg.snapData.len)
stream.write(char(if msg.snapDone: 1 else: 0))
let strData = stream.data let strData = stream.data
result = newSeq[byte](strData.len) result = newSeq[byte](strData.len)
for i in 0 ..< strData.len: for i in 0 ..< strData.len:
@@ -722,6 +738,19 @@ proc deserializeRaftMessage*(data: seq[byte]): RaftMessage =
result.leaderCommit = stream.readUint64() result.leaderCommit = stream.readUint64()
result.success = stream.readChar() != '\0' result.success = stream.readChar() != '\0'
result.matchIdx = stream.readUint64() result.matchIdx = stream.readUint64()
# Optional trailing InstallSnapshot fields (absent in pre-v1.3 buffers)
if not stream.atEnd:
result.snapId = stream.readUint64()
if not stream.atEnd:
result.snapOffset = stream.readUint64()
if not stream.atEnd:
let dataLen = int(stream.readUint32())
result.snapData = newSeq[byte](dataLen)
if dataLen > 0:
if stream.readData(addr result.snapData[0], dataLen) != dataLen:
raise newException(IOError, "Incomplete snapshot data read from stream")
if not stream.atEnd:
result.snapDone = stream.readChar() != '\0'
stream.close() stream.close()
# --------------------------------------------------------------------------- # ---------------------------------------------------------------------------
@@ -810,6 +839,10 @@ proc processMessage*(net: RaftNetwork, msg: RaftMessage) {.async.} =
await net.send(msg.senderId, reply) await net.send(msg.senderId, reply)
of rmkAppendEntriesReply: of rmkAppendEntriesReply:
net.node.handleAppendReply(msg.senderId, msg) net.node.handleAppendReply(msg.senderId, msg)
of rmkInstallSnapshot, rmkInstallSnapshotReply:
# Wire protocol only (v1.3); snapshot transfer behavior lands in a
# follow-up task. Ignore until then.
discard
proc recvExact*(client: AsyncSocket, size: int): Future[string] {.async.} = proc recvExact*(client: AsyncSocket, size: int): Future[string] {.async.} =
## Reads exactly `size` bytes from `client`. A short return means the peer ## Reads exactly `size` bytes from `client`. A short return means the peer
+96
View File
@@ -8,6 +8,7 @@ import std/asyncnet
import std/monotimes import std/monotimes
import std/base64 import std/base64
import std/json import std/json
import std/streams
import barabadb/core/types import barabadb/core/types
import barabadb/core/mvcc import barabadb/core/mvcc
@@ -2607,6 +2608,101 @@ suite "Raft Network Transport":
check replyMsg.kind == rmkRequestVoteReply check replyMsg.kind == rmkRequestVoteReply
check replyMsg.success check replyMsg.success
suite "Raft InstallSnapshot Protocol":
test "InstallSnapshot fields survive serialize/deserialize round-trip":
let msg = RaftMessage(
kind: rmkInstallSnapshot,
term: 9,
senderId: "leader-1",
prevLogIndex: 42, # snapshot base index
prevLogTerm: 7, # snapshot base term
snapId: 3,
snapOffset: 4096,
snapData: @[byte 1, 2, 3, 250, 0, 17],
snapDone: true)
let decoded = deserializeRaftMessage(serialize(msg))
check decoded.kind == rmkInstallSnapshot
check decoded.term == 9
check decoded.senderId == "leader-1"
check decoded.prevLogIndex == 42
check decoded.prevLogTerm == 7
check decoded.snapId == 3
check decoded.snapOffset == 4096
check decoded.snapData == @[byte 1, 2, 3, 250, 0, 17]
check decoded.snapDone
test "InstallSnapshotReply fields survive serialize/deserialize round-trip":
let msg = RaftMessage(
kind: rmkInstallSnapshotReply,
term: 9,
senderId: "follower-2",
success: true,
matchIdx: 42,
snapId: 3,
snapOffset: 8192,
snapData: @[],
snapDone: false)
let decoded = deserializeRaftMessage(serialize(msg))
check decoded.kind == rmkInstallSnapshotReply
check decoded.term == 9
check decoded.senderId == "follower-2"
check decoded.success
check decoded.matchIdx == 42
check decoded.snapId == 3
check decoded.snapOffset == 8192
check decoded.snapData.len == 0
check not decoded.snapDone
test "old wire layout (no snapshot fields) deserializes with zero defaults":
# Manually serialize a message in the pre-InstallSnapshot layout:
# magic, version, kind, term, senderId, lastLogIndex, lastLogTerm,
# prevLogIndex, prevLogTerm, entries, leaderCommit, success, matchIdx.
let s = newStringStream()
s.write("RAFT")
s.write(1'u32) # RaftProtoVersion
s.write(uint32(ord(rmkAppendEntries)))
s.write(5'u64) # term
let sender = "old-leader"
s.write(uint32(sender.len))
s.writeData(sender[0].unsafeAddr, sender.len)
s.write(11'u64) # lastLogIndex
s.write(4'u64) # lastLogTerm
s.write(10'u64) # prevLogIndex
s.write(4'u64) # prevLogTerm
s.write(0'u32) # entries count
s.write(10'u64) # leaderCommit
s.write(char(1)) # success
s.write(10'u64) # matchIdx
let strData = s.data
var buf = newSeq[byte](strData.len)
for i in 0 ..< strData.len:
buf[i] = byte(strData[i])
s.close()
let decoded = deserializeRaftMessage(buf)
check decoded.kind == rmkAppendEntries
check decoded.term == 5
check decoded.senderId == "old-leader"
check decoded.matchIdx == 10
check decoded.snapId == 0
check decoded.snapOffset == 0
check decoded.snapData.len == 0
check not decoded.snapDone
test "old message kinds still round-trip unchanged":
let msg = RaftMessage(kind: rmkRequestVote, term: 2, senderId: "cand",
lastLogIndex: 5, lastLogTerm: 1)
let decoded = deserializeRaftMessage(serialize(msg))
check decoded.kind == rmkRequestVote
check decoded.term == 2
check decoded.senderId == "cand"
check decoded.lastLogIndex == 5
check decoded.lastLogTerm == 1
check decoded.snapId == 0
check decoded.snapOffset == 0
check decoded.snapData.len == 0
check not decoded.snapDone
suite "Raft TLS Transport": suite "Raft TLS Transport":
test "2-node election over TLS": test "2-node election over TLS":
let certDir = getTempDir() / "baradb_test_raft_tls" let certDir = getTempDir() / "baradb_test_raft_tls"