This patch introduces support for a `tcp_syn_count` parameter in
libpq, allowing control over the number of SYN retransmissions when
initiating a connection.
The primary goal is to prevent the walreceiver from getting stuck
resending SYNs for an extended period, up to
`net.ipv4.tcp_syn_retries` (127 seconds by default), in the event of
network disruptions.
A specific scenario where this can occur is during a failover in Kubernetes:
* The primary node fails, a standby is promoted, and other standbys
attempt to reconnect to the service representing the new primary.
* The `primary_conninfo` of the standby points to a service, usually
managed via iptables rules.
* If the walreceiver's initial SYN is dropped due to outdated rules,
the connection may remain stranded until the system timeout is
reached.
* As a result, a second standby may reattach after a couple of
minutes. In the case of synchronous replication, this can block the
writes from the application.
In this scenario, `tcp_user_timeout` could close a connection retrying
the SYNs (even though it doesn't seem to do it from the documentation,
it works) the parameter will affect the entire connection.
`connect_timeout`, doesn't work with `PQconnectPoll`, so it won't
prevent the walreceiver from timing out.
Thank you,
Francesco
From 34eb35abdfdc660d90f104544f40331d659718a9 Mon Sep 17 00:00:00 2001
From: Francesco Canovai <francesco.canovai@enterprisedb.com>
Date: Fri, 31 Jan 2025 13:39:51 +0100
Subject: [PATCH] Support TCP_SYNCNT in libpq
Add a tcp_syn_count parameter to libpq.
This will allow the user to define the maximum amount of retransmission
that TCP will send before aborting the attempt to establish a
connection. This can be used only on systems where TCP_SYNCNT is
available.
---
doc/src/sgml/libpq.sgml | 15 ++++++++++
src/interfaces/libpq/fe-connect.c | 49 +++++++++++++++++++++++++++++++
src/interfaces/libpq/libpq-int.h | 1 +
3 files changed, 65 insertions(+)
diff --git a/doc/src/sgml/libpq.sgml b/doc/src/sgml/libpq.sgml
index 8fa0515c6a0..8029819a25e 100644
--- a/doc/src/sgml/libpq.sgml
+++ b/doc/src/sgml/libpq.sgml
@@ -1576,6 +1576,21 @@ postgresql://%2Fvar%2Flib%2Fpostgresql/dbname
</listitem>
</varlistentry>
+ <varlistentry id="libpq-tcp-syn-count" xreflabel="tcp_syn_count">
+ <term><literal>tcp_syn_count</literal></term>
+ <listitem>
+ <para>
+ Controls the maximum number of SYN retransmissions that TCP
+ will send before aborting the attempt to establish a connection.
+ A higher value increases the number of retries before giving up,
+ while a lower value results in faster failure detection.
+ This parameter is only effective on systems where
+ <symbol>TCP_SYNCNT</symbol> is available; on other systems,
+ it has no effect.
+ </para>
+ </listitem>
+ </varlistentry>
+
<varlistentry id="libpq-connect-replication" xreflabel="replication">
<term><literal>replication</literal></term>
<listitem>
diff --git a/src/interfaces/libpq/fe-connect.c b/src/interfaces/libpq/fe-connect.c
index d5051f5e820..889c1c771d4 100644
--- a/src/interfaces/libpq/fe-connect.c
+++ b/src/interfaces/libpq/fe-connect.c
@@ -267,6 +267,10 @@ static const internalPQconninfoOption PQconninfoOptions[] = {
"TCP-User-Timeout", "", 10, /* strlen(INT32_MAX) == 10 */
offsetof(struct pg_conn, pgtcp_user_timeout)},
+ {"tcp_syn_count", NULL, NULL, NULL,
+ "TCP-SYN-Count", "", 10, /* strlen(INT32_MAX) == 10 */
+ offsetof(struct pg_conn, tcp_syn_count)},
+
/*
* ssl options are allowed even without client SSL support because the
* client can still handle SSL modes "disable" and "allow". Other
@@ -2627,6 +2631,41 @@ setTCPUserTimeout(PGconn *conn)
return 1;
}
+/*
+* Set the maximum number of SYN retransmissions.
+*/
+static int
+setTCPSynCnt(PGconn *conn)
+{
+ int count;
+
+ if (conn->tcp_syn_count == NULL)
+ return 1;
+
+ if (!pqParseIntParam(conn->tcp_syn_count, &count, conn,
+ "tcp_syn_count"))
+ return 0;
+
+ if (count < 0)
+ count = 0;
+
+#ifdef TCP_SYNCNT
+ if (setsockopt(conn->sock, IPPROTO_TCP, TCP_SYNCNT,
+ (char *) &count, sizeof(count)) < 0)
+ {
+ char sebuf[PG_STRERROR_R_BUFLEN];
+
+ libpq_append_conn_error(conn, "%s(%s) failed: %s",
+ "setsockopt",
+ "TCP_SYNCNT",
+ SOCK_STRERROR(SOCK_ERRNO, sebuf, sizeof(sebuf)));
+ return 0;
+ }
+#endif
+
+ return 1;
+}
+
/* ----------
* pqConnectDBStart -
* Begin the process of making a connection to the backend.
@@ -3368,6 +3407,15 @@ keep_going: /* We will come back to here until there is
}
}
+ if (addr_cur->family != AF_UNIX)
+ {
+ if (!setTCPSynCnt(conn))
+ {
+ conn->try_next_addr = true;
+ goto keep_going;
+ }
+ }
+
/*----------
* We have three methods of blocking SIGPIPE during
* send() calls to this socket:
@@ -5008,6 +5056,7 @@ freePGconn(PGconn *conn)
free(conn->keepalives_idle);
free(conn->keepalives_interval);
free(conn->keepalives_count);
+ free(conn->tcp_syn_count);
free(conn->sslmode);
free(conn->sslnegotiation);
free(conn->sslcert);
diff --git a/src/interfaces/libpq/libpq-int.h b/src/interfaces/libpq/libpq-int.h
index f36f7f19d58..d25f440a940 100644
--- a/src/interfaces/libpq/libpq-int.h
+++ b/src/interfaces/libpq/libpq-int.h
@@ -408,6 +408,7 @@ struct pg_conn
* retransmits */
char *keepalives_count; /* maximum number of TCP keepalive
* retransmits */
+ char *tcp_syn_count; /* maximum number of TCP SYN retransmits */
char *sslmode; /* SSL mode (require,prefer,allow,disable) */
char *sslnegotiation; /* SSL initiation style (postgres,direct) */
char *sslcompression; /* SSL compression (0 or 1) */
--
2.48.1
