infra for v3 acceptor -> proposer msgs

This commit is contained in:
Arseny Sher
2025-01-22 12:54:50 +01:00
parent c19a8b69f2
commit 10a7878230
4 changed files with 130 additions and 99 deletions
+79 -61
View File
@@ -1774,7 +1774,7 @@ HandleSafekeeperResponse(WalProposer *wp, Safekeeper *fromsk)
static void
MembershipConfigurationSerialize(MembershipConfiguration *mconf, StringInfo buf)
{
int i;
uint32 i;
pq_sendint32(buf, mconf->generation);
@@ -1808,8 +1808,8 @@ PAMessageSerialize(WalProposer *wp, ProposerAcceptorMessage *msg, StringInfo buf
resetStringInfo(buf);
/* removeme tag check after converting all msgs */
if (proto_version == 3 && msg->tag == 'g')
//removeme tag check after converting all msgs
{
/*
* v2 sends structs for some messages as is, so commonly send tag only
@@ -1838,8 +1838,9 @@ PAMessageSerialize(WalProposer *wp, ProposerAcceptorMessage *msg, StringInfo buf
return;
}
if (proto_version == 2 || proto_version == 3) // TODO remove proto_version == 3 after converting all msgs
//removeme tag check after converting all msgs
if (proto_version == 2 || proto_version == 3)
//TODO remove proto_version == 3 after converting all msgs
/* removeme tag check after converting all msgs */
{
switch (msg->tag)
{
@@ -1938,82 +1939,99 @@ AsyncReadMessage(Safekeeper *sk, AcceptorProposerMessage *anymsg)
char *buf;
int buf_size;
uint64 tag;
uint8 tag;
StringInfoData s;
if (!(AsyncRead(sk, &buf, &buf_size)))
return false;
sk->latestMsgReceivedAt = wp->api.get_current_timestamp(wp);
/* parse it */
s.data = buf;
s.len = buf_size;
s.cursor = 0;
tag = pq_getmsgint64_le(&s);
if (tag != anymsg->tag)
/* removeme tag check after converting all msgs */
if (wp->config->proto_version == 3 && anymsg->tag == 'g')
{
wp_log(WARNING, "unexpected message tag %c from node %s:%s in state %s", (char) tag, sk->host,
sk->port, FormatSafekeeperState(sk));
ResetConnection(sk);
return false;
tag = pq_getmsgint(&s, 1);
if (tag != anymsg->tag)
{
wp_log(WARNING, "unexpected message tag %c from node %s:%s in state %s", (char) tag, sk->host,
sk->port, FormatSafekeeperState(sk));
ResetConnection(sk);
return false;
}
}
sk->latestMsgReceivedAt = wp->api.get_current_timestamp(wp);
switch (tag)
/* removeme tag check after converting all msgs */
else if (wp->config->proto_version == 2 || wp->config->proto_version == 3)
{
case 'g':
{
AcceptorGreeting *msg = (AcceptorGreeting *) anymsg;
msg->term = pq_getmsgint64_le(&s);
msg->nodeId = pq_getmsgint64_le(&s);
pq_getmsgend(&s);
return true;
}
case 'v':
{
VoteResponse *msg = (VoteResponse *) anymsg;
msg->term = pq_getmsgint64_le(&s);
msg->voteGiven = pq_getmsgint64_le(&s);
msg->flushLsn = pq_getmsgint64_le(&s);
msg->truncateLsn = pq_getmsgint64_le(&s);
msg->termHistory.n_entries = pq_getmsgint32_le(&s);
msg->termHistory.entries = palloc(sizeof(TermSwitchEntry) * msg->termHistory.n_entries);
for (int i = 0; i < msg->termHistory.n_entries; i++)
tag = pq_getmsgint64_le(&s);
if (tag != anymsg->tag)
{
wp_log(WARNING, "unexpected message tag %c from node %s:%s in state %s", (char) tag, sk->host,
sk->port, FormatSafekeeperState(sk));
ResetConnection(sk);
return false;
}
switch (tag)
{
case 'g':
{
msg->termHistory.entries[i].term = pq_getmsgint64_le(&s);
msg->termHistory.entries[i].lsn = pq_getmsgint64_le(&s);
AcceptorGreeting *msg = (AcceptorGreeting *) anymsg;
msg->term = pq_getmsgint64_le(&s);
msg->nodeId = pq_getmsgint64_le(&s);
pq_getmsgend(&s);
return true;
}
msg->timelineStartLsn = pq_getmsgint64_le(&s);
pq_getmsgend(&s);
return true;
}
case 'a':
{
AppendResponse *msg = (AppendResponse *) anymsg;
case 'v':
{
VoteResponse *msg = (VoteResponse *) anymsg;
msg->term = pq_getmsgint64_le(&s);
msg->flushLsn = pq_getmsgint64_le(&s);
msg->commitLsn = pq_getmsgint64_le(&s);
msg->hs.ts = pq_getmsgint64_le(&s);
msg->hs.xmin.value = pq_getmsgint64_le(&s);
msg->hs.catalog_xmin.value = pq_getmsgint64_le(&s);
if (s.len > s.cursor)
ParsePageserverFeedbackMessage(wp, &s, &msg->ps_feedback);
else
msg->ps_feedback.present = false;
pq_getmsgend(&s);
return true;
}
msg->term = pq_getmsgint64_le(&s);
msg->voteGiven = pq_getmsgint64_le(&s);
msg->flushLsn = pq_getmsgint64_le(&s);
msg->truncateLsn = pq_getmsgint64_le(&s);
msg->termHistory.n_entries = pq_getmsgint32_le(&s);
msg->termHistory.entries = palloc(sizeof(TermSwitchEntry) * msg->termHistory.n_entries);
for (int i = 0; i < msg->termHistory.n_entries; i++)
{
msg->termHistory.entries[i].term = pq_getmsgint64_le(&s);
msg->termHistory.entries[i].lsn = pq_getmsgint64_le(&s);
}
msg->timelineStartLsn = pq_getmsgint64_le(&s);
pq_getmsgend(&s);
return true;
}
default:
{
Assert(false);
return false;
}
case 'a':
{
AppendResponse *msg = (AppendResponse *) anymsg;
msg->term = pq_getmsgint64_le(&s);
msg->flushLsn = pq_getmsgint64_le(&s);
msg->commitLsn = pq_getmsgint64_le(&s);
msg->hs.ts = pq_getmsgint64_le(&s);
msg->hs.xmin.value = pq_getmsgint64_le(&s);
msg->hs.catalog_xmin.value = pq_getmsgint64_le(&s);
if (s.len > s.cursor)
ParsePageserverFeedbackMessage(wp, &s, &msg->ps_feedback);
else
msg->ps_feedback.present = false;
pq_getmsgend(&s);
return true;
}
default:
{
wp_log(FATAL, "unexpected message tag %c", (char) tag);
return false;
}
}
}
wp_log(FATAL, "unsupported proto_version %d", wp->config->proto_version);
}
/*
+1 -1
View File
@@ -219,7 +219,7 @@ typedef struct ProposerGreetingV2
typedef struct AcceptorProposerMessage
{
uint64 tag;
uint8 tag;
} AcceptorProposerMessage;
/*
+3 -2
View File
@@ -281,7 +281,7 @@ impl SafekeeperPostgresHandler {
tokio::select! {
// todo: add read|write .context to these errors
r = network_reader.run(msg_tx, msg_rx, reply_tx, timeline, next_msg) => r,
r = network_write(pgb, reply_rx, pageserver_feedback_rx) => r,
r = network_write(pgb, reply_rx, pageserver_feedback_rx, proto_version) => r,
_ = timeline_cancel.cancelled() => {
return Err(CopyStreamHandlerEnd::Cancelled);
}
@@ -459,6 +459,7 @@ async fn network_write<IO: AsyncRead + AsyncWrite + Unpin>(
pgb_writer: &mut PostgresBackend<IO>,
mut reply_rx: Receiver<AcceptorProposerMessage>,
mut pageserver_feedback_rx: tokio::sync::broadcast::Receiver<PageserverFeedback>,
proto_version: u32,
) -> Result<(), CopyStreamHandlerEnd> {
let mut buf = BytesMut::with_capacity(128);
@@ -496,7 +497,7 @@ async fn network_write<IO: AsyncRead + AsyncWrite + Unpin>(
};
buf.clear();
msg.serialize(&mut buf)?;
msg.serialize(&mut buf, proto_version)?;
pgb_writer.write_message(&BeMessage::CopyData(&buf)).await?;
}
}
+47 -35
View File
@@ -34,7 +34,8 @@ use utils::{
lsn::Lsn,
};
pub const SK_PROTOCOL_VERSION: u32 = 2;
pub const SK_PROTO_VERSION_2: u32 = 2;
pub const SK_PROTO_VERSION_3: u32 = 3;
pub const UNKNOWN_SERVER_VERSION: u32 = 0;
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq, PartialOrd, Ord)]
@@ -447,7 +448,7 @@ impl ProposerAcceptorMessage {
pub fn parse(mut msg_bytes: Bytes, proto_version: u32) -> Result<ProposerAcceptorMessage> {
// TODO remove after converting all msgs
let t = msg_bytes[0] as char;
if proto_version == 3 && t == 'g' {
if proto_version == SK_PROTO_VERSION_3 && t == 'g' {
if msg_bytes.is_empty() {
bail!("ProposerAcceptorMessage is not complete: missing tag");
}
@@ -483,7 +484,7 @@ impl ProposerAcceptorMessage {
_ => bail!("unknown proposer-acceptor message tag: {}", tag),
}
// TODO remove proto_version == 3 after converting all msgs
} else if proto_version == 2 || proto_version == 3 {
} else if proto_version == SK_PROTO_VERSION_2 || proto_version == SK_PROTO_VERSION_3 {
// xxx using Reader is inefficient but easy to work with bincode
let mut stream = msg_bytes.reader();
// u64 is here to avoid padding; it will be removed once we stop packing C structs into the wire as is
@@ -624,44 +625,55 @@ pub enum AcceptorProposerMessage {
impl AcceptorProposerMessage {
/// Serialize acceptor -> proposer message.
pub fn serialize(&self, buf: &mut BytesMut) -> Result<()> {
match self {
AcceptorProposerMessage::Greeting(msg) => {
buf.put_u64_le('g' as u64);
buf.put_u64_le(msg.term);
buf.put_u64_le(msg.node_id.0);
pub fn serialize(&self, buf: &mut BytesMut, proto_version: u32) -> Result<()> {
// TODO remove after converting all msgs
if proto_version == SK_PROTO_VERSION_3
&& (matches!(self, AcceptorProposerMessage::Greeting(_)))
{
match self {
_ => bail!("not implemented"),
}
AcceptorProposerMessage::VoteResponse(msg) => {
buf.put_u64_le('v' as u64);
buf.put_u64_le(msg.term);
buf.put_u64_le(msg.vote_given);
buf.put_u64_le(msg.flush_lsn.into());
buf.put_u64_le(msg.truncate_lsn.into());
buf.put_u32_le(msg.term_history.0.len() as u32);
for e in &msg.term_history.0 {
buf.put_u64_le(e.term);
buf.put_u64_le(e.lsn.into());
// TODO remove 3 after converting all msgs
} else if proto_version == SK_PROTO_VERSION_2 || proto_version == SK_PROTO_VERSION_3 {
match self {
AcceptorProposerMessage::Greeting(msg) => {
buf.put_u64_le('g' as u64);
buf.put_u64_le(msg.term);
buf.put_u64_le(msg.node_id.0);
}
buf.put_u64_le(msg.timeline_start_lsn.into());
}
AcceptorProposerMessage::AppendResponse(msg) => {
buf.put_u64_le('a' as u64);
buf.put_u64_le(msg.term);
buf.put_u64_le(msg.flush_lsn.into());
buf.put_u64_le(msg.commit_lsn.into());
buf.put_i64_le(msg.hs_feedback.ts);
buf.put_u64_le(msg.hs_feedback.xmin);
buf.put_u64_le(msg.hs_feedback.catalog_xmin);
AcceptorProposerMessage::VoteResponse(msg) => {
buf.put_u64_le('v' as u64);
buf.put_u64_le(msg.term);
buf.put_u64_le(msg.vote_given);
buf.put_u64_le(msg.flush_lsn.into());
buf.put_u64_le(msg.truncate_lsn.into());
buf.put_u32_le(msg.term_history.0.len() as u32);
for e in &msg.term_history.0 {
buf.put_u64_le(e.term);
buf.put_u64_le(e.lsn.into());
}
buf.put_u64_le(msg.timeline_start_lsn.into());
}
AcceptorProposerMessage::AppendResponse(msg) => {
buf.put_u64_le('a' as u64);
buf.put_u64_le(msg.term);
buf.put_u64_le(msg.flush_lsn.into());
buf.put_u64_le(msg.commit_lsn.into());
buf.put_i64_le(msg.hs_feedback.ts);
buf.put_u64_le(msg.hs_feedback.xmin);
buf.put_u64_le(msg.hs_feedback.catalog_xmin);
// AsyncReadMessage in walproposer.c will not try to decode pageserver_feedback
// if it is not present.
if let Some(ref msg) = msg.pageserver_feedback {
msg.serialize(buf);
// AsyncReadMessage in walproposer.c will not try to decode pageserver_feedback
// if it is not present.
if let Some(ref msg) = msg.pageserver_feedback {
msg.serialize(buf);
}
}
}
Ok(())
} else {
bail!("unsupported protocol version {}", proto_version);
}
Ok(())
}
}