PS/Prefetch: Use a timeout for reading data from TCP (#10834)

This reduces pressure on OS TCP buffers, reducing flush times in other
systems like PageServer.

## Problem

## Summary of changes

pgxn/neon/libpagestore.c

@@ -1099,6 +1099,10 @@ pageserver_try_receive(shardno_t shard_no)
 		{
 			neon_shard_log(shard_no, LOG, "pageserver_receive: disconnect due to failure while parsing response");
 			pageserver_disconnect(shard_no);
+			/*
+			 * Malformed responses from PageServer are a reason to raise
+			 * errors and cancel transactions.
+			 */
 			PG_RE_THROW();
 		}
 		PG_END_TRY();
@@ -1122,7 +1126,8 @@ pageserver_try_receive(shardno_t shard_no)
 		char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
 
 		pageserver_disconnect(shard_no);
-		neon_shard_log(shard_no, ERROR, "pageserver_receive disconnect: could not read COPY data: %s", msg);
+		neon_shard_log(shard_no, LOG, "pageserver_receive disconnect: could not read COPY data: %s", msg);
+		resp = NULL;
 	}
 	else
 	{
@@ -1321,6 +1326,16 @@ pg_init_libpagestore(void)
 							PGC_USERSET,
 							0,	/* no flags required */
 							NULL, (GucIntAssignHook) &readahead_buffer_resize, NULL);
+	DefineCustomIntVariable("neon.readahead_getpage_pull_timeout",
+							"readahead response pull timeout",
+							"Time between active tries to pull data from the "
+							"PageStream connection when we have pages which "
+							"were read ahead but not yet received.",
+							&readahead_getpage_pull_timeout_ms,
+							0, 0, 5 * 60 * 1000,
+							PGC_USERSET,
+							GUC_UNIT_MS,
+							NULL, NULL, NULL);
 	DefineCustomIntVariable("neon.protocol_version",
 							"Version of compute<->page server protocol",
 							NULL,
@@ -1334,7 +1349,7 @@ pg_init_libpagestore(void)
 
 	DefineCustomIntVariable("neon.pageserver_response_log_timeout",
 							"pageserver response log timeout",
-							"If the pageserver doesn't respond to a request within this timeout,"
+							"If the pageserver doesn't respond to a request within this timeout, "
 							"a message is printed to the log.",
 							&pageserver_response_log_timeout,
 							10000, 100, INT_MAX,
@@ -1344,7 +1359,7 @@ pg_init_libpagestore(void)
 
 	DefineCustomIntVariable("neon.pageserver_response_disconnect_timeout",
 							"pageserver response diconnect timeout",
-							"If the pageserver doesn't respond to a request within this timeout,"
+							"If the pageserver doesn't respond to a request within this timeout, "
 							"disconnect and reconnect.",
 							&pageserver_response_disconnect_timeout,
 							120000, 100, INT_MAX,

pgxn/neon/neon.c

@@ -437,6 +437,7 @@ _PG_init(void)
 
 	pg_init_libpagestore();
 	pg_init_walproposer();
+	pagestore_smgr_init();
 	Custom_XLogReaderRoutines = NeonOnDemandXLogReaderRoutines;
 
 	InitUnstableExtensionsSupport();

pgxn/neon/neon.h

@@ -22,6 +22,7 @@ extern char *neon_tenant;
 extern char *wal_acceptors_list;
 extern int	wal_acceptor_reconnect_timeout;
 extern int	wal_acceptor_connection_timeout;
+extern int	readahead_getpage_pull_timeout_ms;
 
 #if PG_MAJORVERSION_NUM >= 17
 extern uint32		WAIT_EVENT_NEON_LFC_MAINTENANCE;
@@ -49,6 +50,7 @@ extern uint32		WAIT_EVENT_NEON_WAL_DL;
 
 extern void pg_init_libpagestore(void);
 extern void pg_init_walproposer(void);
+extern void pagestore_smgr_init(void);
 
 extern uint64 BackpressureThrottlingTime(void);
 extern void SetNeonCurrentClusterSize(uint64 size);

pgxn/neon/pagestore_client.h

@@ -209,7 +209,11 @@ typedef struct
 	NeonResponse *(*receive) (shardno_t shard_no);
 	/*
 	 * Try get the next response from the TCP buffers, if any.
-	 * Returns NULL when the data is not yet available. 
+	 * Returns NULL when the data is not yet available.
+	 *
+	 * This will raise errors only for malformed responses (we can't put them
+	 * back into connection). All other error conditions are soft errors and
+	 * return NULL as "no response available".
 	 */
 	NeonResponse *(*try_receive) (shardno_t shard_no);
 	/*