halephant-core 0.5.2

A proxy and pooler for Postgres
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
//! Transaction-mode proxy: the outer loop cycles between an idle
//! state (no server connection) and an active state (holding a
//! checked-out [`crate::connections::server::ServerConn`] for the
//! duration of a single transaction). On `ReadyForQuery(Idle)` the
//! connection is returned to the pool; on `Terminate` or disconnect
//! the loop exits.

use std::sync::Arc;

use futures_util::{SinkExt, StreamExt};
use tokio::net::TcpStream;
use tokio_util::codec::Framed;
use tracing::{Instrument, debug, field, info_span, trace, warn};

use crate::clients::{ClientGuard, ClientState};
use crate::config::cluster::pool::ListenMode;
use crate::config::otel::QueryText;
use crate::connections::server::{Reply, ServerConn};
use crate::listener::ClientNotifications;
use crate::o11y;
use crate::pool::PoolManager;
use crate::proto;
use crate::proxy::intercept::{
    handle_close_intercept, handle_deallocate_intercept, handle_listen_intercept,
};
use crate::proxy::prepared::{ClientPrepared, PreparedGuard, rewrite_outbound};
use crate::sql;

/// Track session-state-affecting commands on the server connection so
/// halephant's per-connection pool state stays in sync with what
/// PostgreSQL actually holds. Two trackers are kept in step here:
///
/// - `server.dirty_vars` — GUC variables the client has SET and that
///   `reset_connection` must RESET on checkin.
/// - `server.statements` — canonical names of prepared statements
///   registered on this backend via `ensure_prepared`.
///
/// Only tracks the simple query protocol (Query messages) — a SET
/// inside a Parse is not tracked because it hasn't been executed yet.
fn track_set_reset(msg: &proto::frontend::FrontendMessage, server: &mut ServerConn) {
    let proto::frontend::FrontendMessage::Query(query) = msg else {
        return;
    };
    // SET LOCAL falls into the wildcard arm below: its effect is
    // transaction-scoped and unwound automatically on COMMIT/ROLLBACK,
    // so there's nothing to track in dirty_vars.
    match sql::parse(query) {
        sql::Statement::Set {
            scope: sql::SetScope::Session,
            parameter,
            ..
        } => {
            server.dirty_vars.insert(parameter);
        }
        sql::Statement::Reset {
            target: sql::ResetTarget::All,
        } => {
            server.dirty_vars.clear();
        }
        sql::Statement::Reset {
            target: sql::ResetTarget::Parameter(name),
        } => {
            server.dirty_vars.remove(&name);
        }
        sql::Statement::Discard {
            target: sql::DiscardTarget::All,
        } => {
            server.statements.discard_all();
            server.dirty_vars.clear();
        }
        _ => {}
    }
}

/// Why we returned from the active forwarding state.
enum Release {
    /// Server sent ReadyForQuery(Idle) — transaction complete, connection can
    /// be returned to the pool. The client remains connected and may start
    /// another transaction.
    Idle,
    /// Client sent Terminate or disconnected.
    ClientDone,
}

/// Outer loop: alternate between idle (no server connection) and active
/// (forwarding a transaction).
#[allow(clippy::too_many_arguments)]
pub async fn forward(
    client: &mut Framed<TcpStream, proto::codec::FrontendCodec>,
    pools: &Arc<PoolManager>,
    client_guard: &ClientGuard,
    notifications: &mut Option<ClientNotifications>,
    listen_mode: ListenMode,
    database: &str,
    user: &str,
    read_only: bool,
    query_text: QueryText,
) -> anyhow::Result<()> {
    let mut client_prepared = PreparedGuard {
        inner: ClientPrepared::new(),
        pools: Arc::clone(pools),
    };
    forward_loop(
        client,
        pools,
        client_guard,
        notifications,
        listen_mode,
        database,
        user,
        read_only,
        &mut client_prepared.inner,
        query_text,
    )
    .await
}

#[allow(clippy::too_many_arguments)]
async fn forward_loop(
    client: &mut Framed<TcpStream, proto::codec::FrontendCodec>,
    pools: &Arc<PoolManager>,
    client_guard: &ClientGuard,
    notifications: &mut Option<ClientNotifications>,
    listen_mode: ListenMode,
    database: &str,
    user: &str,
    read_only: bool,
    client_prepared: &mut ClientPrepared,
    query_text: QueryText,
) -> anyhow::Result<()> {
    loop {
        // Idle — wait for the client to send something (or a notification).
        let first_msg = loop {
            tokio::select! {
                msg = client.next() => match msg.transpose()? {
                    Some(proto::frontend::FrontendMessage::Terminate) => {
                        trace!("client sent Terminate (idle)");
                        return Ok(());
                    }
                    Some(msg) => {
                        // We're in the outer idle loop with no server
                        // connection checked out — by definition the
                        // session has no active transaction, so any
                        // synthetic intercept response advertises
                        // `Idle` to the client.
                        let idle = proto::types::TransactionStatus::Idle;

                        // In multiplex mode, intercept LISTEN/UNLISTEN in idle state
                        // (no server connection needed).
                        if listen_mode == ListenMode::Multiplex
                            && let Some(true) =
                                handle_listen_intercept(&msg, notifications, client, idle).await?
                        {
                            continue; // stay idle
                        }

                        // Intercept Close for named statements — return synthetic
                        // CloseComplete without needing a server connection.
                        if let Some(true) = handle_close_intercept(&msg, client_prepared, client, pools).await? {
                            continue; // stay idle
                        }

                        // Intercept DEALLOCATE queries for the same
                        // reason — halephant renames client Parse names
                        // to canonical hashes, so a raw DEALLOCATE
                        // would not find the statement on the server.
                        if let Some(true) = handle_deallocate_intercept(&msg, client_prepared, client, pools, idle).await? {
                            continue; // stay idle
                        }

                        break msg;
                    }
                    None => return Ok(()),
                },
                Some(notif) = recv_notification(notifications) => {
                    client.send(notif).await?;
                }
            }
        };

        // Reject read-write overrides on read-only sessions.
        if read_only && is_read_write_override_msg(&first_msg) {
            reject_read_write(client, proto::types::TransactionStatus::Idle).await?;
            continue;
        }

        // Determine routing for this transaction. If the first message is
        // BEGIN READ ONLY (or START TRANSACTION READ ONLY), route to a replica
        // even if the session isn't globally read-only.
        let txn_read_only = read_only || is_read_only_query(&first_msg);

        let txn_span = info_span!(
            "proxy.transaction",
            db.namespace = database,
            user,
            server.address = field::Empty,
            otel.status_code = field::Empty,
            otel.status_description = field::Empty,
        );

        let release = async {
            let result: anyhow::Result<Release> = async {
                let mut guard = pools
                    .checkout(client_guard, database, user, txn_read_only)
                    .await?;
                client_guard.set_state(ClientState::InTransaction);
                tracing::Span::current().record("server.address", guard.node());

                // Reset this connection's injected-response FIFOs only. A
                // client that disconnected mid extended-protocol batch can
                // leave a ParseComplete/CloseComplete disposition whose reply
                // the backend skipped and will never send; clearing here
                // keeps it from misaligning this transaction's filter.
                //
                // This is NOT where stale prepared-cache entries are handled:
                // a rejected `Parse` rolls back its optimistic insert at the
                // `ErrorResponse` arm below (before checkin), so it never
                // reaches a later checkout. `reset_pending` is FIFO alignment
                // only, not a catch-all for error-induced staleness.
                guard.conn().statements.reset_pending();

                // In pin mode, check if the first message is LISTEN — pin immediately.
                let mut pinned = false;
                if matches!(classify(&first_msg), sql::Statement::Listen { .. }) {
                    pinned = listen_mode == ListenMode::Pin;
                }
                if pinned {
                    warn!("LISTEN detected, pinning connection to session mode");
                }

                let mut tracker = o11y::spans::StatementTracker::new(query_text);

                // Track SET/RESET and rewrite/send the first message.
                // A first message that is itself a `Sync` ends its
                // (empty) batch immediately, so the inner loop must wait
                // for that `Sync`'s `ReadyForQuery` before reading any
                // pipelined follow-on messages (which belong to the next
                // transaction and need a fresh checkout).
                tracker.on_client_msg(&first_msg, |name| {
                    resolve_stmt_query(name, client_prepared, pools)
                });
                track_set_reset(&first_msg, guard.conn());
                // Derive the batch boundary from what actually reaches the
                // backend, not the raw message: if `rewrite_outbound`
                // suppresses the `Sync` (returns `None`), no `ReadyForQuery`
                // is owed and the gate must stay open. Mirrors the inner-loop
                // check below and eliminates the coupling to the raw message.
                let mut first_ends_batch = false;
                if let Some(first_msg) =
                    rewrite_outbound(first_msg, client_prepared, guard.conn(), pools, client)
                        .await?
                {
                    first_ends_batch = matches!(first_msg, proto::frontend::FrontendMessage::Sync);
                    guard.conn().framed.send(first_msg).await?;
                }
                debug!("transaction started");

                // Forward until the transaction completes (ReadyForQuery Idle).
                let result = forward_until_idle(
                    client,
                    guard.conn(),
                    notifications,
                    listen_mode,
                    &mut pinned,
                    client_prepared,
                    pools,
                    txn_read_only,
                    &mut tracker,
                    first_ends_batch,
                )
                .await;

                match result {
                    Ok(Release::Idle) => {
                        debug!("transaction complete, connection returned to pool");
                        guard.checkin();
                        Ok(Release::Idle)
                    }
                    Ok(Release::ClientDone) => {
                        guard.checkin();
                        Ok(Release::ClientDone)
                    }
                    Err(e) => Err(e),
                }
            }
            .await;

            // Back to Idle regardless of result — InTransaction is only
            // set while we're actively holding a backend. On error the
            // checkin happened inside the inner result; on success the
            // explicit checkin above already ran.
            client_guard.set_state(ClientState::Idle);

            let span = tracing::Span::current();
            match &result {
                Ok(_) => {
                    span.record("otel.status_code", "OK");
                }
                Err(e) => {
                    span.record("otel.status_code", "ERROR");
                    span.record("otel.status_description", e.to_string().as_str());
                }
            }
            result
        }
        .instrument(txn_span)
        .await;

        match release {
            Ok(Release::Idle) => {}
            Ok(Release::ClientDone) => return Ok(()),
            Err(e) => return Err(e),
        }
    }
}

/// Inner loop: forward bidirectionally until ReadyForQuery(Idle) or the client
/// goes away.
#[allow(clippy::too_many_arguments)]
async fn forward_until_idle(
    client: &mut Framed<TcpStream, proto::codec::FrontendCodec>,
    server: &mut ServerConn,
    notifications: &mut Option<ClientNotifications>,
    listen_mode: ListenMode,
    pinned: &mut bool,
    client_prepared: &mut ClientPrepared,
    pools: &Arc<PoolManager>,
    txn_read_only: bool,
    tracker: &mut o11y::spans::StatementTracker,
    awaiting_ready: bool,
) -> anyhow::Result<Release> {
    // While `awaiting_ready` is set, a `Sync` has been forwarded and the
    // backend owes a `ReadyForQuery`; we must NOT read further client
    // messages, because a client that pipelines batches has the *next*
    // transaction's messages already buffered, and forwarding them onto
    // this (about-to-be-checked-in) backend would strand their responses
    // when this batch's `ReadyForQuery(Idle)` returns us to the pool.
    let mut awaiting_ready = awaiting_ready;
    // Whether a real `ReadyForQuery(InTransaction|Failed)` has confirmed we
    // are inside an explicit `BEGIN`…`COMMIT`. While held that way the
    // backend is ours until the transaction ends, so cross-`Sync` read-ahead
    // is safe and the gate is skipped to keep intra-transaction pipelining.
    // It must start `false` (not derive from `last_server_status`, whose
    // optimistic `InTransaction` default below would wrongly relax the gate
    // and let the first autocommit batch be read past).
    let mut in_explicit_txn = false;
    // Track the most recent transaction status from the server so
    // synthetic responses for intercepted client messages reflect
    // the actual session state. The default is `InTransaction`
    // because we entered this function after sending the first
    // message of a new transaction — even if no server response has
    // arrived yet (the client may have pipelined a BEGIN + LISTEN),
    // the expected post-first-message state is in-transaction. The
    // server's eventual ReadyForQuery overwrites this with the real
    // status, including `InFailedTransaction` when applicable.
    let mut last_server_status = proto::types::TransactionStatus::InTransaction;

    loop {
        tokio::select! {
            msg = client.next(), if !awaiting_ready => match msg.transpose()? {
                Some(proto::frontend::FrontendMessage::Terminate) => {
                    trace!("client sent Terminate (in transaction)");
                    return Ok(Release::ClientDone);
                }
                Some(msg) => {
                    // Multiplex: intercept LISTEN/UNLISTEN even mid-transaction.
                    if listen_mode == ListenMode::Multiplex
                        && let Some(true) = handle_listen_intercept(
                            &msg,
                            notifications,
                            client,
                            last_server_status,
                        )
                        .await?
                    {
                        continue;
                    }

                    // Reject read-write overrides on read-only transactions.
                    if txn_read_only && is_read_write_override_msg(&msg) {
                        reject_read_write(client, proto::types::TransactionStatus::InTransaction).await?;
                        continue;
                    }

                    // Intercept Close for named statements.
                    if let Some(true) = handle_close_intercept(&msg, client_prepared, client, pools).await? {
                        continue;
                    }

                    // Intercept DEALLOCATE queries.
                    if let Some(true) = handle_deallocate_intercept(
                        &msg,
                        client_prepared,
                        client,
                        pools,
                        last_server_status,
                    )
                    .await?
                    {
                        continue;
                    }

                    // Pin mode: detect LISTEN/UNLISTEN for pinning.
                    if listen_mode == ListenMode::Pin {
                        match classify(&msg) {
                            sql::Statement::Listen { .. } if !*pinned => {
                                warn!("LISTEN detected, pinning connection to session mode");
                                *pinned = true;
                            }
                            sql::Statement::Unlisten {
                                target: sql::UnlistenTarget::Star,
                            } if *pinned => {
                                debug!("UNLISTEN * detected, unpinning connection");
                                *pinned = false;
                            }
                            _ => {}
                        }
                    }

                    // Record statement span before rewriting.
                    tracker.on_client_msg(&msg, |name| resolve_stmt_query(name, client_prepared, pools));

                    // Track SET/RESET and rewrite prepared statement names.
                    track_set_reset(&msg, server);
                    if let Some(msg) = rewrite_outbound(msg, client_prepared, server, pools, client).await? {
                        // A `Sync` ends the current extended-protocol batch.
                        // In autocommit the next `ReadyForQuery(Idle)` checks
                        // this backend in, so stop reading the client until it
                        // arrives — otherwise a pipelining client's next batch
                        // is read ahead onto this backend and stranded. Inside
                        // an explicit transaction the backend is held to
                        // COMMIT, so read-ahead is safe and the gate is skipped.
                        if matches!(msg, proto::frontend::FrontendMessage::Sync)
                            && !in_explicit_txn
                        {
                            awaiting_ready = true;
                        }
                        trace!(?msg, "client -> server");
                        server.framed.send(msg).await?;
                    }
                }
                None => return Ok(Release::ClientDone),
            },
            msg = server.framed.next() => match msg.transpose()? {
                Some(proto::backend::BackendMessage::ReadyForQuery(status)) => {
                    // The batch's boundary arrived — resume reading the client.
                    awaiting_ready = false;
                    // A non-`Idle` status means we're held inside an explicit
                    // transaction, where the gate above can be relaxed.
                    in_explicit_txn = status != proto::types::TransactionStatus::Idle;
                    last_server_status = status;
                    server.last_tx_status = status;
                    tracker.drain();
                    client.send(proto::backend::BackendMessage::ReadyForQuery(status)).await?;
                    if status == proto::types::TransactionStatus::Idle && !*pinned {
                        return Ok(Release::Idle);
                    }
                }
                Some(proto::backend::BackendMessage::CommandComplete(ref tag)) => {
                    tracker.on_command_complete(tag);
                    client.send(proto::backend::BackendMessage::CommandComplete(tag.clone())).await?;
                }
                Some(proto::backend::BackendMessage::EmptyQueryResponse) => {
                    tracker.on_command_complete("");
                    client.send(proto::backend::BackendMessage::EmptyQueryResponse).await?;
                }
                Some(proto::backend::BackendMessage::ErrorResponse(ref err)) => {
                    tracker.on_error(err.message().unwrap_or("unknown"));
                    // The error aborts the extended-protocol batch: every
                    // message after it is skipped, so any Parse halephant
                    // recorded optimistically will never be confirmed. Roll
                    // those inserts back so the pooled connection doesn't
                    // claim statements the backend never prepared.
                    server.statements.roll_back_after_error();
                    client.send(proto::backend::BackendMessage::ErrorResponse(err.clone())).await?;
                }
                Some(proto::backend::BackendMessage::ParseComplete) => {
                    // A ParseComplete answers a Parse halephant either
                    // forwarded for the client (deliver it) or injected to
                    // re-prepare on the client's behalf (swallow it). The
                    // disposition was recorded in send order when the Parse
                    // went out.
                    match server.statements.next_parse_reply() {
                        Reply::Suppress => trace!("suppressing injected ParseComplete"),
                        Reply::Forward => {
                            client.send(proto::backend::BackendMessage::ParseComplete).await?;
                        }
                    }
                }
                Some(proto::backend::BackendMessage::CloseComplete) => {
                    // Likewise for CloseComplete: forward the client's portal
                    // closes, swallow halephant's LRU-eviction closes.
                    match server.statements.next_close_reply() {
                        Reply::Suppress => trace!("suppressing injected CloseComplete"),
                        Reply::Forward => {
                            client.send(proto::backend::BackendMessage::CloseComplete).await?;
                        }
                    }
                }
                Some(msg) => {
                    trace!(?msg, "server -> client");
                    client.send(msg).await?;
                }
                None => {
                    tracker.drain();
                    // Send a synthesized FATAL ErrorResponse so the
                    // client knows the transaction was lost, rather
                    // than just seeing EOF.
                    crate::messages::error::send_fatal(
                        client,
                        "08006",
                        "upstream server closed unexpectedly",
                    )
                    .await;
                    anyhow::bail!("upstream server closed unexpectedly");
                }
            },
            Some(notif) = recv_notification(notifications) => {
                client.send(notif).await?;
            },
        }
    }
}

// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------

/// Receive a notification from the client's subscription, or pend forever if
/// multiplexing is not active for this client.
async fn recv_notification(
    notifs: &mut Option<ClientNotifications>,
) -> Option<proto::backend::BackendMessage> {
    match notifs {
        Some(n) => n.recv().await,
        None => std::future::pending().await,
    }
}

/// Look up the query text for a named prepared statement via the client's
/// name mapping and the global statement store. Used as the fallback for
/// cross-transaction statement reuse.
fn resolve_stmt_query(
    client_name: &str,
    client_prepared: &ClientPrepared,
    pools: &PoolManager,
) -> Option<String> {
    let canon = client_prepared.resolve(client_name)?;
    let store = pools.stmt_store.lock();
    store.get(canon).map(|parse| parse.query.clone())
}

/// Parse the SQL body of a client message, if any.
fn classify(msg: &proto::frontend::FrontendMessage) -> sql::Statement {
    match msg {
        proto::frontend::FrontendMessage::Query(q) => sql::parse(q),
        proto::frontend::FrontendMessage::Parse(p) => sql::parse(&p.query),
        _ => sql::Statement::Other,
    }
}

/// Check if a message is a `BEGIN READ ONLY` (or `START TRANSACTION READ ONLY`).
fn is_read_only_query(msg: &proto::frontend::FrontendMessage) -> bool {
    matches!(
        classify(msg),
        sql::Statement::Begin {
            options: sql::TransactionOptions {
                read_only: Some(true),
            },
        }
    )
}

/// Check if a message attempts to switch to read-write mode.
fn is_read_write_override_msg(msg: &proto::frontend::FrontendMessage) -> bool {
    classify(msg).is_read_write_override()
}

/// Send an error to the client rejecting a read-write override on a
/// replica-routed connection. Thin wrapper around
/// [`crate::messages::error::send_error`] that hardcodes the
/// SQLSTATE and message for this specific rejection.
async fn reject_read_write(
    client: &mut Framed<TcpStream, proto::codec::FrontendCodec>,
    status: proto::types::TransactionStatus,
) -> anyhow::Result<()> {
    crate::messages::error::send_error(
        client,
        "25006",
        "cannot switch to read-write mode: connection is routed to a read replica",
        status,
    )
    .await
}