Make sendControlMessage to retry when interrupted (#13371)

`sendControlMessage()` method has no retry attempts, this refactor abstracted a `sendto` system call wrapper with retry enabled.

Co-authored-by: zwenlin <zwenlin@vmware.com>

Author: 2 people

contrib/interconnect/udp/ic_udpifc.c

@@ -848,6 +848,8 @@ static inline void logPkt(char *prefix, icpkthdr *pkt);
 static void aggregateStatistics(ChunkTransportStateEntry *pChunkEntry);
 
 static inline bool pollAcks(ChunkTransportState *transportStates, int fd, int timeout);
+
+static ssize_t sendtoWithRetry(int socket, const void *message, size_t length, int flags, const struct sockaddr *dest_addr, socklen_t dest_len, int retry, const char *errDetail);
 static TupleChunkListItem receiveChunksUDPIFCLoop(ChunkTransportState *pTransportStates, ChunkTransportStateEntry *pEntry,
 						int16 *srcRoute, MotionConn *conn, WaitEventSet *waitset, int nevent);
 
@@ -2224,9 +2226,6 @@ destroyConnHashTable(ConnHashTable *ht)
 /*
  * sendControlMessage
  * 		Helper function to send a control message.
- *
- * It is different from sendOnce which retries on interrupts...
- * Here, we leave it to retransmit logic to handle these cases.
  */
 static inline void
 sendControlMessage(icpkthdr *pkt, int fd, struct sockaddr *addr, socklen_t peerLen)
@@ -2247,13 +2246,10 @@ sendControlMessage(icpkthdr *pkt, int fd, struct sockaddr *addr, socklen_t peerL
 	if (gp_interconnect_full_crc)
 		addCRC(pkt);
 
-	n = sendto(fd, (const char *) pkt, pkt->len, 0, addr, peerLen);
-
-	/*
-	 * No need to handle EAGAIN here: no-space just means that we dropped the
-	 * packet: our ordinary retransmit mechanism will handle that case
-	 */
-
+	char errDetail[100];
+	snprintf(errDetail, sizeof(errDetail), "Send control message: got error with seq %u", pkt->seq);
+	/* Retry for infinite times since we have no retransmit mechanism for control message */
+	n = sendtoWithRetry(fd, (const char *) pkt, pkt->len, 0, addr, peerLen, -1, errDetail);
 	if (n < pkt->len)
 		write_log("sendcontrolmessage: got error %d errno %d seq %d", n, errno, pkt->seq);
 }
@@ -5346,43 +5342,41 @@ prepareXmit(MotionConn *mConn)
 }
 
 /*
- * sendOnce
- * 		Send a packet.
+ * sendtoWithRetry
+ * 		Retry sendto logic and send the packets.
  */
-static void
-sendOnce(ChunkTransportState *transportStates, ChunkTransportStateEntry *pChunkEntry, ICBuffer *buf, MotionConn *mConn)
+static ssize_t
+sendtoWithRetry(int socket, const void *message, size_t length,
+           int flags, const struct sockaddr *dest_addr,
+           socklen_t dest_len, int retry, const char *errDetail)
 {
 	int32		n;
-	ChunkTransportStateEntryUDP *pEntry = NULL;
-	MotionConnUDP *conn = NULL;
-
-	pEntry = CONTAINER_OF(pChunkEntry, ChunkTransportStateEntryUDP, entry);
-	Assert(pEntry);
-
-	conn = CONTAINER_OF(mConn, MotionConnUDP, mConn);
-
-#ifdef USE_ASSERT_CHECKING
-	if (testmode_inject_fault(gp_udpic_dropxmit_percent))
-	{
-#ifdef AMS_VERBOSE_LOGGING
-		write_log("THROW PKT with seq %d srcpid %d despid %d", buf->pkt->seq, buf->pkt->srcPid, buf->pkt->dstPid);
-#endif
-		return;
-	}
-#endif
+	int count = 0;
 
 xmit_retry:
-	n = sendto(pEntry->txfd, buf->pkt, buf->pkt->len, 0,
-			   (struct sockaddr *) &conn->peer, conn->peer_len);
+	/*
+	 * If given retry count is positive, retry up to the limited times.
+	 * Otherwise, retry for unlimited times until succeed.
+	 */
+	if (retry > 0 && ++count > retry)
+		return n;
+	n = sendto(socket, message, length, flags, dest_addr, dest_len);
 	if (n < 0)
 	{
 		int			save_errno = errno;
 
 		if (errno == EINTR)
 			goto xmit_retry;
 
-		if (errno == EAGAIN)	/* no space ? not an error. */
-			return;
+		/*
+		 * EAGAIN: no space ? not an error.
+		 *
+		 * EFAULT: In Linux system call, it only happens when copying a socket
+		 * address into kernel space failed, which is less likely to happen,
+		 * but mocked heavily by our fault injection in regression tests.
+		 */
+		if (errno == EAGAIN || errno == EFAULT)
+			return n;
 
 		/*
 		 * If Linux iptables (nf_conntrack?) drops an outgoing packet, it may
@@ -5394,20 +5388,52 @@ sendOnce(ChunkTransportState *transportStates, ChunkTransportStateEntry *pChunkE
 			ereport(LOG,
 					(errcode(ERRCODE_GP_INTERCONNECTION_ERROR),
 					 errmsg("Interconnect error writing an outgoing packet: %m"),
-					 errdetail("error during sendto() for Remote Connection: contentId=%d at %s",
-							   conn->mConn.remoteContentId, conn->mConn.remoteHostAndPort)));
-			return;
+					 errdetail("error during sendto() %s", errDetail)));
+			return n;
 		}
 
 		ereport(ERROR, (errcode(ERRCODE_GP_INTERCONNECTION_ERROR),
 						errmsg("Interconnect error writing an outgoing packet: %m"),
 						errdetail("error during sendto() call (error:%d).\n"
-								  "For Remote Connection: contentId=%d at %s",
-								  save_errno, conn->mConn.remoteContentId,
-								  conn->mConn.remoteHostAndPort)));
+								  "%s", save_errno, errDetail)));
 		/* not reached */
 	}
 
+	return n;
+}
+
+/*
+ * sendOnce
+ * 		Send a packet.
+ */
+static void
+sendOnce(ChunkTransportState *transportStates, ChunkTransportStateEntry *pChunkEntry, ICBuffer *buf, MotionConn *mConn)
+{
+	int32 n;
+	ChunkTransportStateEntryUDP *pEntry = NULL;
+	MotionConnUDP *conn = NULL;
+
+	pEntry = CONTAINER_OF(pChunkEntry, ChunkTransportStateEntryUDP, entry);
+	Assert(pEntry);
+
+	conn = CONTAINER_OF(mConn, MotionConnUDP, mConn);
+
+#ifdef USE_ASSERT_CHECKING
+	if (testmode_inject_fault(gp_udpic_dropxmit_percent))
+	{
+#ifdef AMS_VERBOSE_LOGGING
+		write_log("THROW PKT with seq %d srcpid %d despid %d", buf->pkt->seq, buf->pkt->srcPid, buf->pkt->dstPid);
+#endif
+		return;
+	}
+#endif
+
+	char errDetail[100];
+	snprintf(errDetail, sizeof(errDetail), "For Remote Connection: contentId=%d at %s",
+					  conn->mConn.remoteContentId,
+					  conn->mConn.remoteHostAndPort);
+	n = sendtoWithRetry(pEntry->txfd, buf->pkt, buf->pkt->len, 0,
+                          (struct sockaddr *) &conn->peer, conn->peer_len, -1, errDetail);
 	if (n != buf->pkt->len)
 	{
 		if (DEBUG1 >= log_min_messages)
@@ -5419,7 +5445,6 @@ sendOnce(ChunkTransportState *transportStates, ChunkTransportStateEntry *pChunkE
 		logPkt("PKT DETAILS ", buf->pkt);
 #endif
 	}
-
 	return;
 }
 
@@ -5903,7 +5928,7 @@ checkNetworkTimeout(ICBuffer *buf, uint64 now, bool *networkTimeoutIsLogged)
 		ereport(ERROR,
 				(errcode(ERRCODE_GP_INTERCONNECTION_ERROR),
 				 errmsg("interconnect encountered a network error, please check your network"),
-				 errdetail("Failed to send packet (seq %d) to %s (pid %d cid %d) after %d retries in %d seconds.",
+				 errdetail("Failed to send packet (seq %u) to %s (pid %d cid %d) after %u retries in %d seconds.",
 						   buf->pkt->seq, buf->conn->remoteHostAndPort,
 						   buf->pkt->dstPid, buf->pkt->dstContentId,
 						   buf->nRetry, Gp_interconnect_transmit_timeout)));
@@ -5924,7 +5949,7 @@ checkNetworkTimeout(ICBuffer *buf, uint64 now, bool *networkTimeoutIsLogged)
 	{
 		ereport(WARNING,
 				(errmsg("interconnect may encountered a network error, please check your network"),
-				 errdetail("Failed to send packet (seq %d) to %s (pid %d cid %d) after %d retries.",
+				 errdetail("Failing to send packet (seq %u) to %s (pid %d cid %d) after %u retries.",
 						   buf->pkt->seq, buf->conn->remoteHostAndPort,
 						   buf->pkt->dstPid, buf->pkt->dstContentId,
 						   buf->nRetry)));

src/test/regress/init_file

@@ -88,6 +88,9 @@ m/^WARNING:  table ".*" contains rows in segment .*, which is outside the # of s
 # The following output is generated by \d on foreign tables, so ignore it.
 m/Distributed by: \(.*\)/
 m/Distributed randomly/
+# The following output is an interconnect network warning, but still not error out, so ignore it.
+m/WARNING:  interconnect may encountered a network error, please check your network/
+m/Failing to send packet/
 
 # directory_table test output is sensitive to the user running the tests
 m/^NOTICE:.*storage user mapping for .* does not exist for storage server/