void
wts_load(void)
{
	WT_CONNECTION *conn;
	WT_CURSOR *cursor;
	WT_ITEM key, value;
	WT_SESSION *session;
	uint8_t *keybuf, *valbuf;
	int is_bulk, ret;

	conn = g.wts_conn;

	if ((ret = conn->open_session(conn, NULL, NULL, &session)) != 0)
		die(ret, "connection.open_session");

	if (g.logging != 0)
		(void)g.wt_api->msg_printf(g.wt_api, session,
		    "=============== bulk load start ===============");

	/*
	 * Avoid bulk load with KVS (there's no bulk load support for a
	 * data-source); avoid bulk load with a custom collator, because
	 * the order of insertion will not match the collation order.
	 */
	is_bulk = !g.c_reverse &&
	    !DATASOURCE("kvsbdb") && !DATASOURCE("helium");
	if ((ret = session->open_cursor(session, g.uri, NULL,
	    is_bulk ? "bulk" : NULL, &cursor)) != 0)
		die(ret, "session.open_cursor");

	/* Set up the default key buffer. */
	key_gen_setup(&keybuf);
	val_gen_setup(&valbuf);

	for (;;) {
		if (++g.key_cnt > g.c_rows) {
			g.key_cnt = g.rows = g.c_rows;
			break;
		}

		/* Report on progress every 100 inserts. */
		if (g.key_cnt % 100 == 0)
			track("bulk load", g.key_cnt, NULL);

		key_gen(keybuf, &key.size, (uint64_t)g.key_cnt, 0);
		key.data = keybuf;
		value_gen(valbuf, &value.size, (uint64_t)g.key_cnt);
		value.data = valbuf;

		switch (g.type) {
		case FIX:
			if (!is_bulk)
				cursor->set_key(cursor, g.key_cnt);
			cursor->set_value(cursor, *(uint8_t *)value.data);
			if (g.logging == LOG_OPS)
				(void)g.wt_api->msg_printf(g.wt_api, session,
				    "%-10s %" PRIu32 " {0x%02" PRIx8 "}",
				    "bulk V",
				    g.key_cnt, ((uint8_t *)value.data)[0]);
			break;
		case VAR:
			if (!is_bulk)
				cursor->set_key(cursor, g.key_cnt);
			cursor->set_value(cursor, &value);
			if (g.logging == LOG_OPS)
				(void)g.wt_api->msg_printf(g.wt_api, session,
				    "%-10s %" PRIu32 " {%.*s}", "bulk V",
				    g.key_cnt,
				    (int)value.size, (char *)value.data);
			break;
		case ROW:
			cursor->set_key(cursor, &key);
			if (g.logging == LOG_OPS)
				(void)g.wt_api->msg_printf(g.wt_api, session,
				    "%-10s %" PRIu32 " {%.*s}", "bulk K",
				    g.key_cnt, (int)key.size, (char *)key.data);
			cursor->set_value(cursor, &value);
			if (g.logging == LOG_OPS)
				(void)g.wt_api->msg_printf(g.wt_api, session,
				    "%-10s %" PRIu32 " {%.*s}", "bulk V",
				    g.key_cnt,
				    (int)value.size, (char *)value.data);
			break;
		}

		if ((ret = cursor->insert(cursor)) != 0)
			die(ret, "cursor.insert");

		if (!SINGLETHREADED)
			continue;

		/* Insert the item into BDB. */
		bdb_insert(key.data, key.size, value.data, value.size);
	}

	if ((ret = cursor->close(cursor)) != 0)
		die(ret, "cursor.close");

	if (g.logging != 0)
		(void)g.wt_api->msg_printf(g.wt_api, session,
//.........这里部分代码省略.........

/*
 * load_dump --
 *	Load from the WiredTiger dump format.
 */
static int
load_dump(WT_SESSION *session)
{
	WT_CURSOR *cursor;
	WT_DECL_RET;
	int hex, tret;
	char **list, **tlist, *uri, config[64];

	cursor = NULL;
	list = NULL;		/* -Wuninitialized */
	hex = 0;		/* -Wuninitialized */
	uri = NULL;

	/* Read the metadata file. */
	if ((ret = config_read(&list, &hex)) != 0)
		return (ret);

	/* Reorder and check the list. */
	if ((ret = config_reorder(list)) != 0)
		goto err;

	/* Update the config based on any command-line configuration. */
	if ((ret = config_update(session, list)) != 0)
		goto err;

	uri = list[0];
	/* Create the items in the list. */
	if ((ret = config_exec(session, list)) != 0)
		goto err;

	/* Open the insert cursor. */
	(void)snprintf(config, sizeof(config),
	    "dump=%s%s%s",
	    hex ? "hex" : "print",
	    append ? ",append" : "", no_overwrite ? ",overwrite=false" : "");
	if ((ret = session->open_cursor(
	    session, uri, NULL, config, &cursor)) != 0) {
		ret = util_err(ret, "%s: session.open", uri);
		goto err;
	}

	/*
	 * Check the append flag (it only applies to objects where the primary
	 * key is a record number).
	 */
	if (append && strcmp(cursor->key_format, "r") != 0) {
		fprintf(stderr,
		    "%s: %s: -a option illegal unless the primary key is a "
		    "record number\n",
		    progname, uri);
		ret = 1;
	} else
		ret = insert(cursor, uri);

err:	/*
	 * Technically, we don't have to close the cursor because the session
	 * handle will do it for us, but I'd like to see the flush to disk and
	 * the close succeed, it's better to fail early when loading files.
	 */
	if (cursor != NULL && (tret = cursor->close(cursor)) != 0) {
		tret = util_err(tret, "%s: cursor.close", uri);
		if (ret == 0)
			ret = tret;
	}
	if (ret == 0)
		ret = util_flush(session, uri);

	for (tlist = list; *tlist != NULL; ++tlist)
		free(*tlist);
	free(list);

	return (ret == 0 ? 0 : 1);
}

/*
 * dump_table_config --
 *	Dump the config for a table.
 */
static int
dump_table_config(
    WT_SESSION *session, WT_CURSOR *cursor, const char *uri, bool json)
{
	WT_CONFIG_ITEM cval;
	WT_CURSOR *srch;
	WT_DECL_RET;
	size_t len;
	int tret;
	bool complex_table;
	const char *name, *v;
	char *p, **cfg, *_cfg[4] = {NULL, NULL, NULL, NULL};

	p = NULL;
	cfg = &_cfg[3];

	/* Get the table name. */
	if ((name = strchr(uri, ':')) == NULL) {
		fprintf(stderr, "%s: %s: corrupted uri\n", progname, uri);
		return (1);
	}
	++name;

	/*
	 * Dump out the config information: first, dump the uri entry itself,
	 * it overrides all subsequent configurations.
	 */
	cursor->set_key(cursor, uri);
	if ((ret = cursor->search(cursor)) != 0)
		WT_ERR(util_cerr(cursor, "search", ret));
	if ((ret = cursor->get_value(cursor, &v)) != 0)
		WT_ERR(util_cerr(cursor, "get_value", ret));
	if ((*--cfg = strdup(v)) == NULL)
		WT_ERR(util_err(session, errno, NULL));

	/*
	 * Workaround for WiredTiger "simple" table handling. Simple tables
	 * have column-group entries, but they aren't listed in the metadata's
	 * table entry, and the name is different from other column-groups.
	 * Figure out if it's a simple table and in that case, retrieve the
	 * column-group's configuration value and the column-group's "source"
	 * entry, where the column-group entry overrides the source's.
	 */
	complex_table = false;
	if (WT_PREFIX_MATCH(uri, "table:")) {
		len = strlen("colgroup:") + strlen(name) + 1;
		if ((p = malloc(len)) == NULL)
			WT_ERR(util_err(session, errno, NULL));
		(void)snprintf(p, len, "colgroup:%s", name);
		cursor->set_key(cursor, p);
		if ((ret = cursor->search(cursor)) == 0) {
			if ((ret = cursor->get_value(cursor, &v)) != 0)
				WT_ERR(util_cerr(cursor, "get_value", ret));
			if ((*--cfg = strdup(v)) == NULL)
				WT_ERR(util_err(session, errno, NULL));
			if ((ret =__wt_config_getones(
			    (WT_SESSION_IMPL *)session,
			    *cfg, "source", &cval)) != 0)
				WT_ERR(util_err(
				    session, ret, "%s: source entry", p));
			free(p);
			len = cval.len + 10;
			if ((p = malloc(len)) == NULL)
				WT_ERR(util_err(session, errno, NULL));
			(void)snprintf(p, len, "%.*s", (int)cval.len, cval.str);
			cursor->set_key(cursor, p);
			if ((ret = cursor->search(cursor)) != 0)
				WT_ERR(util_cerr(cursor, "search", ret));
			if ((ret = cursor->get_value(cursor, &v)) != 0)
				WT_ERR(util_cerr(cursor, "get_value", ret));
			if ((*--cfg = strdup(v)) == NULL)
				WT_ERR(util_err(session, errno, NULL));
		} else
			complex_table = true;
	}

	WT_ERR(print_config(session, uri, cfg, json, true));

	if (complex_table) {
		/*
		 * The underlying table configuration function needs a second
		 * cursor: open one before calling it, it makes error handling
		 * hugely simpler.
		 */
		if ((ret = session->open_cursor(
		    session, "metadata:", NULL, NULL, &srch)) != 0)
			WT_ERR(util_cerr(cursor, "open_cursor", ret));

		if ((ret = dump_table_config_complex(
		    session, cursor, srch, name, "colgroup:", json)) == 0)
			ret = dump_table_config_complex(
			    session, cursor, srch, name, "index:", json);

		if ((tret = srch->close(srch)) != 0) {
			tret = util_cerr(cursor, "close", tret);
			if (ret == 0)
//.........这里部分代码省略.........

//.........这里部分代码省略.........
	ret = session->drop(session, "table:mytable", NULL);
#endif
	/*! [Configure block_allocation] */
	ret = session->create(session, "table:mytable",
	    "key_format=S,value_format=S,block_allocation=first");
	/*! [Configure block_allocation] */
	ret = session->drop(session, "table:mytable", NULL);

	/*! [Create a cache-resident object] */
	ret = session->create(session,
	    "table:mytable", "key_format=r,value_format=S,cache_resident=true");
	/*! [Create a cache-resident object] */
	ret = session->drop(session, "table:mytable", NULL);

	{
	/* Create a table for the session operations. */
	ret = session->create(
	    session, "table:mytable", "key_format=S,value_format=S");

	/*! [Compact a table] */
	ret = session->compact(session, "table:mytable", NULL);
	/*! [Compact a table] */

	/*! [Rebalance a table] */
	ret = session->rebalance(session, "table:mytable", NULL);
	/*! [Rebalance a table] */

	/*! [Rename a table] */
	ret = session->rename(session, "table:old", "table:new", NULL);
	/*! [Rename a table] */

	/*! [Salvage a table] */
	ret = session->salvage(session, "table:mytable", NULL);
	/*! [Salvage a table] */

	/*! [Truncate a table] */
	ret = session->truncate(session, "table:mytable", NULL, NULL, NULL);
	/*! [Truncate a table] */

	/*! [Transaction sync] */
	ret = session->transaction_sync(session, NULL);
	/*! [Transaction sync] */

	/*! [Reset the session] */
	ret = session->reset(session);
	/*! [Reset the session] */

	{
	/*
	 * Insert a pair of keys so we can truncate a range.
	 */
	WT_CURSOR *cursor;
	ret = session->open_cursor(
	    session, "table:mytable", NULL, NULL, &cursor);
	cursor->set_key(cursor, "June01");
	cursor->set_value(cursor, "value");
	ret = cursor->update(cursor);
	cursor->set_key(cursor, "June30");
	cursor->set_value(cursor, "value");
	ret = cursor->update(cursor);
	ret = cursor->close(cursor);

	{
	/*! [Truncate a range] */
	WT_CURSOR *start, *stop;

	ret = session->open_cursor(
	    session, "table:mytable", NULL, NULL, &start);
	start->set_key(start, "June01");
	ret = start->search(start);

	ret = session->open_cursor(
	    session, "table:mytable", NULL, NULL, &stop);
	stop->set_key(stop, "June30");
	ret = stop->search(stop);

	ret = session->truncate(session, NULL, start, stop, NULL);
	/*! [Truncate a range] */
	}
	}

	/*! [Upgrade a table] */
	ret = session->upgrade(session, "table:mytable", NULL);
	/*! [Upgrade a table] */

	/*! [Verify a table] */
	ret = session->verify(session, "table:mytable", NULL);
	/*! [Verify a table] */

	/*! [Drop a table] */
	ret = session->drop(session, "table:mytable", NULL);
	/*! [Drop a table] */
	}

	/*! [Close a session] */
	ret = session->close(session, NULL);
	/*! [Close a session] */

	return (ret);
}

int
transaction_ops(WT_CONNECTION *conn, WT_SESSION *session)
{
	WT_CURSOR *cursor;
	int ret;

	/*! [transaction commit/rollback] */
	/*
	 * Cursors may be opened before or after the transaction begins, and in
	 * either case, subsequent operations are included in the transaction.
	 * Opening cursors before the transaction begins allows applications to
	 * cache cursors and use them for multiple operations.
	 */
	ret =
	    session->open_cursor(session, "table:mytable", NULL, NULL, &cursor);
	ret = session->begin_transaction(session, NULL);

	cursor->set_key(cursor, "key");
	cursor->set_value(cursor, "value");
	switch (ret = cursor->update(cursor)) {
	case 0:					/* Update success */
		ret = session->commit_transaction(session, NULL);
		/*
		 * If commit_transaction succeeds, cursors remain positioned; if
		 * commit_transaction fails, the transaction was rolled-back and
		 * and all cursors are reset.
		 */
		break;
	case WT_ROLLBACK:			/* Update conflict */
	default:				/* Other error */
		ret = session->rollback_transaction(session, NULL);
		/* The rollback_transaction call resets all cursors. */
		break;
	}

	/*
	 * Cursors remain open and may be used for multiple transactions.
	 */
	/*! [transaction commit/rollback] */
	ret = cursor->close(cursor);

	/*! [transaction isolation] */
	/* A single transaction configured for snapshot isolation. */
	ret =
	    session->open_cursor(session, "table:mytable", NULL, NULL, &cursor);
	ret = session->begin_transaction(session, "isolation=snapshot");
	cursor->set_key(cursor, "some-key");
	cursor->set_value(cursor, "some-value");
	ret = cursor->update(cursor);
	ret = session->commit_transaction(session, NULL);
	/*! [transaction isolation] */

	/*! [session isolation configuration] */
	/* Open a session configured for read-uncommitted isolation. */
	ret = conn->open_session(
	    conn, NULL, "isolation=read_uncommitted", &session);
	/*! [session isolation configuration] */

	/*! [session isolation re-configuration] */
	/* Re-configure a session for snapshot isolation. */
	ret = session->reconfigure(session, "isolation=snapshot");
	/*! [session isolation re-configuration] */

	{
	/*! [transaction pinned range] */
	/* Check the transaction ID range pinned by the session handle. */
	uint64_t range;

	ret = session->transaction_pinned_range(session, &range);
	/*! [transaction pinned range] */
	}

	return (ret);
}

/*
 * __wt_las_sweep --
 *	Sweep the lookaside table.
 */
int
__wt_las_sweep(WT_SESSION_IMPL *session)
{
	WT_CONNECTION_IMPL *conn;
	WT_CURSOR *cursor;
	WT_DECL_ITEM(las_addr);
	WT_DECL_ITEM(las_key);
	WT_DECL_RET;
	WT_ITEM *key;
	uint64_t cnt, las_counter, las_txnid;
	int64_t remove_cnt;
	uint32_t las_id, session_flags;
	int notused;

	conn = S2C(session);
	cursor = NULL;
	key = &conn->las_sweep_key;
	remove_cnt = 0;
	session_flags = 0;		/* [-Werror=maybe-uninitialized] */

	WT_ERR(__wt_scr_alloc(session, 0, &las_addr));
	WT_ERR(__wt_scr_alloc(session, 0, &las_key));

	WT_ERR(__wt_las_cursor(session, &cursor, &session_flags));

	/*
	 * If we're not starting a new sweep, position the cursor using the key
	 * from the last call (we don't care if we're before or after the key,
	 * just roughly in the same spot is fine).
	 */
	if (key->size != 0) {
		__wt_cursor_set_raw_key(cursor, key);
		ret = cursor->search_near(cursor, &notused);

		/*
		 * Don't search for the same key twice; if we don't set a new
		 * key below, it's because we've reached the end of the table
		 * and we want the next pass to start at the beginning of the
		 * table. Searching for the same key could leave us stuck at
		 * the end of the table, repeatedly checking the same rows.
		 */
		key->size = 0;
		if (ret != 0)
			goto srch_notfound;
	}

	/*
	 * The sweep server wakes up every 10 seconds (by default), it's a slow
	 * moving thread. Try to review the entire lookaside table once every 5
	 * minutes, or every 30 calls.
	 *
	 * The reason is because the lookaside table exists because we're seeing
	 * cache/eviction pressure (it allows us to trade performance and disk
	 * space for cache space), and it's likely lookaside blocks are being
	 * evicted, and reading them back in doesn't help things. A trickier,
	 * but possibly better, alternative might be to review all lookaside
	 * blocks in the cache in order to get rid of them, and slowly review
	 * lookaside blocks that have already been evicted.
	 */
	cnt = (uint64_t)WT_MAX(100, conn->las_record_cnt / 30);

	/* Discard pages we read as soon as we're done with them. */
	F_SET(session, WT_SESSION_NO_CACHE);

	/* Walk the file. */
	for (; cnt > 0 && (ret = cursor->next(cursor)) == 0; --cnt) {
		/*
		 * If the loop terminates after completing a work unit, we will
		 * continue the table sweep next time. Get a local copy of the
		 * sweep key, we're going to reset the cursor; do so before
		 * calling cursor.remove, cursor.remove can discard our hazard
		 * pointer and the page could be evicted from underneath us.
		 */
		if (cnt == 1) {
			WT_ERR(__wt_cursor_get_raw_key(cursor, key));
			if (!WT_DATA_IN_ITEM(key))
				WT_ERR(__wt_buf_set(
				    session, key, key->data, key->size));
		}

		WT_ERR(cursor->get_key(cursor,
		    &las_id, las_addr, &las_counter, &las_txnid, las_key));

		/*
		 * If the on-page record transaction ID associated with the
		 * record is globally visible, the record can be discarded.
		 *
		 * Cursor opened overwrite=true: won't return WT_NOTFOUND should
		 * another thread remove the record before we do, and the cursor
		 * remains positioned in that case.
		 */
		if (__wt_txn_visible_all(session, las_txnid)) {
			WT_ERR(cursor->remove(cursor));
			++remove_cnt;
		}
	}
//.........这里部分代码省略.........

int
util_stat(WT_SESSION *session, int argc, char *argv[])
{
	WT_CURSOR *cursor;
	WT_DECL_RET;
	size_t urilen;
	int ch;
	bool objname_free;
	const char *config, *pval, *desc;
	char *objname, *uri;

	objname_free = false;
	objname = uri = NULL;
	config = NULL;
	while ((ch = __wt_getopt(progname, argc, argv, "af")) != EOF)
		switch (ch) {
		case 'a':
			/*
			 * Historically, the -a option meant include all of the
			 * statistics; because we are opening the database with
			 * statistics=(all), that is now the default, allow the
			 * option for compatibility.
			 */
			config = NULL;
			break;
		case 'f':
			config = "statistics=(fast)";
			break;
		case '?':
		default:
			return (usage());
		}
	argc -= __wt_optind;
	argv += __wt_optind;

	/*
	 * If there are no arguments, the statistics cursor operates on the
	 * connection, otherwise, the optional remaining argument is a file
	 * or LSM name.
	 */
	switch (argc) {
	case 0:
		objname = (char *)"";
		break;
	case 1:
		if ((objname = util_name(session, *argv, "table")) == NULL)
			return (1);
		objname_free = true;
		break;
	default:
		return (usage());
	}

	urilen = strlen("statistics:") + strlen(objname) + 1;
	if ((uri = calloc(urilen, 1)) == NULL) {
		fprintf(stderr, "%s: %s\n", progname, strerror(errno));
		goto err;
	}
	snprintf(uri, urilen, "statistics:%s", objname);

	if ((ret =
	    session->open_cursor(session, uri, NULL, config, &cursor)) != 0) {
		fprintf(stderr, "%s: cursor open(%s) failed: %s\n",
		    progname, uri, session->strerror(session, ret));
		goto err;
	}

	/* List the statistics. */
	while (
	    (ret = cursor->next(cursor)) == 0 &&
	    (ret = cursor->get_value(cursor, &desc, &pval, NULL)) == 0)
		if (printf("%s=%s\n", desc, pval) < 0) {
			ret = errno;
			break;
		}
	if (ret == WT_NOTFOUND)
		ret = 0;

	if (ret != 0) {
		fprintf(stderr, "%s: cursor get(%s) failed: %s\n",
		    progname, objname, session->strerror(session, ret));
		goto err;
	}

	if (0) {
err:		ret = 1;
	}
	if (objname_free)
		free(objname);
	free(uri);

	return (ret);
}

int
main(int argc, char *argv[])
{
	WT_CONNECTION *conn, *conn2, *conn3, *conn4;
	WT_CURSOR *cursor;
	WT_ITEM data;
	WT_SESSION *session;
	uint64_t i;
	int ch, status, op, ret;
	bool child;
	const char *working_dir;
	char cmd[512];
	uint8_t buf[MAX_VAL];

	if ((progname = strrchr(argv[0], DIR_DELIM)) == NULL)
		progname = argv[0];
	else
		++progname;
	/*
	 * Needed unaltered for system command later.
	 */
	saved_argv0 = argv[0];

	working_dir = "WT_RD";
	child = false;
	op = OP_READ;
	while ((ch = __wt_getopt(progname, argc, argv, "Rh:W")) != EOF)
		switch (ch) {
		case 'R':
			child = true;
			op = OP_READ;
			break;
		case 'W':
			child = true;
			op = OP_WRITE;
			break;
		case 'h':
			working_dir = __wt_optarg;
			break;
		default:
			usage();
		}
	argc -= __wt_optind;
	argv += __wt_optind;
	if (argc != 0)
		usage();

	/*
	 * Set up all the directory names.
	 */
	testutil_work_dir_from_path(home, sizeof(home), working_dir);
	(void)snprintf(home_wr, sizeof(home_wr), "%s%s", home, HOME_WR_SUFFIX);
	(void)snprintf(home_rd, sizeof(home_rd), "%s%s", home, HOME_RD_SUFFIX);
	(void)snprintf(
	    home_rd2, sizeof(home_rd2), "%s%s", home, HOME_RD2_SUFFIX);
	if (!child) {
		testutil_make_work_dir(home);
		testutil_make_work_dir(home_wr);
		testutil_make_work_dir(home_rd);
		testutil_make_work_dir(home_rd2);
	} else
		/*
		 * We are a child process, we just want to call
		 * the open_dbs with the directories we have.
		 * The child function will exit.
		 */
		open_dbs(op, home, home_wr, home_rd, home_rd2);

	/*
	 * Parent creates a database and table.  Then cleanly shuts down.
	 * Then copy database to read-only directory and chmod.
	 * Also copy database to read-only directory and remove the lock
	 * file.  One read-only database will have a lock file in the
	 * file system and the other will not.
	 * Parent opens all databases with read-only configuration flag.
	 * Parent forks off child who tries to also open all databases
	 * with the read-only flag.  It should error on the writeable
	 * directory, but allow it on the read-only directories.
	 * The child then confirms it can read all the data.
	 */
	/*
	 * Run in the home directory and create the table.
	 */
	if ((ret = wiredtiger_open(home, NULL, ENV_CONFIG, &conn)) != 0)
		testutil_die(ret, "wiredtiger_open");
	if ((ret = conn->open_session(conn, NULL, NULL, &session)) != 0)
		testutil_die(ret, "WT_CONNECTION:open_session");
	if ((ret = session->create(session,
	    uri, "key_format=Q,value_format=u")) != 0)
		testutil_die(ret, "WT_SESSION.create: %s", uri);
	if ((ret =
	    session->open_cursor(session, uri, NULL, NULL, &cursor)) != 0)
		testutil_die(ret, "WT_SESSION.open_cursor: %s", uri);

	/*
	 * Write data into the table and then cleanly shut down connection.
	 */
	memset(buf, 0, sizeof(buf));
	data.data = buf;
	data.size = MAX_VAL;
//.........这里部分代码省略.........

/*
 * build --
 *	Build a row- or column-store page in a file.
 */
void
build(int ikey, int ivalue, int cnt)
{
	WT_CONNECTION *conn;
	WT_CURSOR *cursor;
	WT_ITEM key, value;
	WT_SESSION *session;
	char config[256], kbuf[64], vbuf[64];
	int new_slvg;

	assert(wiredtiger_open(NULL, NULL, "create", &conn) == 0);
	assert(conn->open_session(conn, NULL, NULL, &session) == 0);
	assert(session->drop(session, "file:" LOAD, "force") == 0);

	switch (page_type) {
	case WT_PAGE_COL_FIX:
		(void)snprintf(config, sizeof(config),
		    "key_format=r,value_format=7t,"
		    "allocation_size=%d,"
		    "internal_page_max=%d,internal_item_max=%d,"
		    "leaf_page_max=%d,leaf_item_max=%d",
		    PSIZE, PSIZE, OSIZE, PSIZE, OSIZE);
		break;
	case WT_PAGE_COL_VAR:
		(void)snprintf(config, sizeof(config),
		    "key_format=r,"
		    "allocation_size=%d,"
		    "internal_page_max=%d,internal_item_max=%d,"
		    "leaf_page_max=%d,leaf_item_max=%d",
		    PSIZE, PSIZE, OSIZE, PSIZE, OSIZE);
		break;
	case WT_PAGE_ROW_LEAF:
		(void)snprintf(config, sizeof(config),
		    "key_format=u,"
		    "allocation_size=%d,"
		    "internal_page_max=%d,internal_item_max=%d,"
		    "leaf_page_max=%d,leaf_item_max=%d",
		    PSIZE, PSIZE, OSIZE, PSIZE, OSIZE);
		break;
	default:
		assert(0);
	}
	assert(session->create(session, "file:" LOAD, config) == 0);
	assert(session->open_cursor(
	    session, "file:" LOAD, NULL, "bulk", &cursor) == 0);
	for (; cnt > 0; --cnt, ++ikey, ++ivalue) {
		switch (page_type) {			/* Build the key. */
		case WT_PAGE_COL_FIX:
		case WT_PAGE_COL_VAR:
			break;
		case WT_PAGE_ROW_LEAF:
			snprintf(kbuf, sizeof(kbuf), "%010d KEY------", ikey);
			key.data = kbuf;
			key.size = 20;
			cursor->set_key(cursor, &key);
			break;
		}

		switch (page_type) {			/* Build the value. */
		case WT_PAGE_COL_FIX:
			cursor->set_value(cursor, ivalue & 0x7f);
			break;
		case WT_PAGE_COL_VAR:
		case WT_PAGE_ROW_LEAF:
			snprintf(vbuf, sizeof(vbuf),
			    "%010d VALUE----", value_unique ? ivalue : 37);
			value.data = vbuf;
			value.size = 20;
			cursor->set_value(cursor, &value);
		}
		assert(cursor->insert(cursor) == 0);
	}

	/*
	 * The first time through this routine we put a matching configuration
	 * in for the salvage file.
	 */
	new_slvg = (access(SLVG, F_OK) != 0);
	if (new_slvg) {
		assert(session->drop(session, "file:" SLVG, "force") == 0);
		assert(session->create(session, "file:" SLVG, config) == 0);
	}

	assert(conn->close(conn, 0) == 0);

	/*
	 * We created the salvage file above, but all we want is the schema,
	 * we're creating the salvage file by hand.
	 */
	if (new_slvg)
		(void)remove(SLVG);
}

/*
 * __wt_schema_open_table --
 *	Open a named table.
 */
int
__wt_schema_open_table(WT_SESSION_IMPL *session,
    const char *name, size_t namelen, WT_TABLE **tablep)
{
	WT_CONFIG cparser;
	WT_CONFIG_ITEM ckey, cval;
	WT_CURSOR *cursor;
	WT_DECL_RET;
	WT_ITEM buf;
	WT_TABLE *table;
	const char *tconfig;
	char *tablename;

	cursor = NULL;
	table = NULL;

	WT_CLEAR(buf);
	WT_RET(__wt_buf_fmt(session, &buf, "table:%.*s", (int)namelen, name));
	tablename = __wt_buf_steal(session, &buf, NULL);

	WT_ERR(__wt_metadata_cursor(session, NULL, &cursor));
	cursor->set_key(cursor, tablename);
	WT_ERR(cursor->search(cursor));
	WT_ERR(cursor->get_value(cursor, &tconfig));

	WT_ERR(__wt_calloc_def(session, 1, &table));
	table->name = tablename;
	tablename = NULL;

	WT_ERR(__wt_config_getones(session, tconfig, "columns", &cval));

	WT_ERR(__wt_config_getones(session, tconfig, "key_format", &cval));
	WT_ERR(__wt_strndup(session, cval.str, cval.len, &table->key_format));
	WT_ERR(__wt_config_getones(session, tconfig, "value_format", &cval));
	WT_ERR(__wt_strndup(session, cval.str, cval.len, &table->value_format));
	WT_ERR(__wt_strdup(session, tconfig, &table->config));

	/* Point to some items in the copy to save re-parsing. */
	WT_ERR(__wt_config_getones(session, table->config,
	    "columns", &table->colconf));

	/*
	 * Count the number of columns: tables are "simple" if the columns
	 * are not named.
	 */
	WT_ERR(__wt_config_subinit(session, &cparser, &table->colconf));
	table->is_simple = 1;
	while ((ret = __wt_config_next(&cparser, &ckey, &cval)) == 0)
		table->is_simple = 0;
	if (ret != WT_NOTFOUND)
		goto err;

	/* Check that the columns match the key and value formats. */
	if (!table->is_simple)
		WT_ERR(__wt_schema_colcheck(session,
		    table->key_format, table->value_format, &table->colconf,
		    &table->nkey_columns, NULL));

	WT_ERR(__wt_config_getones(session, table->config,
	    "colgroups", &table->cgconf));

	/* Count the number of column groups. */
	WT_ERR(__wt_config_subinit(session, &cparser, &table->cgconf));
	table->ncolgroups = 0;
	while ((ret = __wt_config_next(&cparser, &ckey, &cval)) == 0)
		++table->ncolgroups;
	if (ret != WT_NOTFOUND)
		goto err;

	WT_ERR(__wt_calloc_def(session, WT_COLGROUPS(table), &table->cgroups));
	WT_ERR(__wt_schema_open_colgroups(session, table));
	*tablep = table;

	if (0) {
err:		if (table != NULL)
			__wt_schema_destroy_table(session, table);
	}
	if (cursor != NULL)
		WT_TRET(cursor->close(cursor));
	__wt_free(session, tablename);
	return (ret);
}

/*
 * __wt_las_sweep --
 *	Sweep the lookaside table.
 */
int
__wt_las_sweep(WT_SESSION_IMPL *session)
{
	WT_CONNECTION_IMPL *conn;
	WT_CURSOR *cursor;
	WT_DECL_ITEM(las_addr);
	WT_DECL_ITEM(las_key);
	WT_DECL_RET;
	WT_ITEM *key;
	uint64_t cnt, las_counter, las_txnid;
	uint32_t las_id, session_flags;
	int notused;

	conn = S2C(session);
	cursor = NULL;
	key = &conn->las_sweep_key;
	session_flags = 0;		/* [-Werror=maybe-uninitialized] */

	WT_ERR(__wt_scr_alloc(session, 0, &las_addr));
	WT_ERR(__wt_scr_alloc(session, 0, &las_key));

	WT_ERR(__wt_las_cursor(session, &cursor, &session_flags));

	/*
	 * If we're not starting a new sweep, position the cursor using the key
	 * from the last call (we don't care if we're before or after the key,
	 * just roughly in the same spot is fine).
	 */
	if (conn->las_sweep_call != 0 && key->data != NULL) {
		__wt_cursor_set_raw_key(cursor, key);
		if ((ret = cursor->search_near(cursor, &notused)) != 0)
			goto srch_notfound;
	}

	/*
	 * The sweep server wakes up every 10 seconds (by default), it's a slow
	 * moving thread. Try to review the entire lookaside table once every 5
	 * minutes, or every 30 calls.
	 *
	 * The reason is because the lookaside table exists because we're seeing
	 * cache/eviction pressure (it allows us to trade performance and disk
	 * space for cache space), and it's likely lookaside blocks are being
	 * evicted, and reading them back in doesn't help things. A trickier,
	 * but possibly better, alternative might be to review all lookaside
	 * blocks in the cache in order to get rid of them, and slowly review
	 * lookaside blocks that have already been evicted.
	 *
	 * We can't know for sure how many records are in the lookaside table,
	 * the cursor insert and remove statistics aren't updated atomically.
	 * Start with reviewing 100 rows, and if it takes more than the target
	 * number of calls to finish, increase the number of rows checked on
	 * each call; if it takes less than the target calls to finish, then
	 * decrease the number of rows reviewed on each call (but never less
	 * than 100).
	 */
#define	WT_SWEEP_LOOKASIDE_MIN_CNT	100
#define	WT_SWEEP_LOOKASIDE_PASS_TARGET	 30
	++conn->las_sweep_call;
	if ((cnt = conn->las_sweep_cnt) < WT_SWEEP_LOOKASIDE_MIN_CNT)
		cnt = conn->las_sweep_cnt = WT_SWEEP_LOOKASIDE_MIN_CNT;

	/* Walk the file. */
	for (; cnt > 0 && (ret = cursor->next(cursor)) == 0; --cnt) {
		/*
		 * If the loop terminates after completing a work unit, we will
		 * continue the table sweep next time. Get a local copy of the
		 * sweep key, we're going to reset the cursor; do so before
		 * calling cursor.remove, cursor.remove can discard our hazard
		 * pointer and the page could be evicted from underneath us.
		 */
		if (cnt == 1) {
			WT_ERR(__wt_cursor_get_raw_key(cursor, key));
			if (!WT_DATA_IN_ITEM(key))
				WT_ERR(__wt_buf_set(
				    session, key, key->data, key->size));
		}

		WT_ERR(cursor->get_key(cursor,
		    &las_id, las_addr, &las_counter, &las_txnid, las_key));

		/*
		 * If the on-page record transaction ID associated with the
		 * record is globally visible, the record can be discarded.
		 *
		 * Cursor opened overwrite=true: won't return WT_NOTFOUND should
		 * another thread remove the record before we do, and the cursor
		 * remains positioned in that case.
		 */
		if (__wt_txn_visible_all(session, las_txnid))
			WT_ERR(cursor->remove(cursor));
	}

	/*
	 * When reaching the lookaside table end or the target number of calls,
	 * adjust the row count. Decrease/increase the row count depending on
	 * if the number of calls is less/more than the target.
//.........这里部分代码省略.........

/*
 * __wt_schema_open_indices --
 *	Open the indices for a table.
 */
int
__wt_schema_open_index(WT_SESSION_IMPL *session,
    WT_TABLE *table, const char *idxname, size_t len, WT_INDEX **indexp)
{
	WT_CURSOR *cursor;
	WT_DECL_ITEM(tmp);
	WT_DECL_RET;
	WT_INDEX *idx;
	u_int i;
	int cmp, match;
	const char *idxconf, *name, *tablename, *uri;

	/* Check if we've already done the work. */
	if (idxname == NULL && table->idx_complete)
		return (0);

	cursor = NULL;
	idx = NULL;

	/* Build a search key. */
	tablename = table->name;
	(void)WT_PREFIX_SKIP(tablename, "table:");
	WT_ERR(__wt_scr_alloc(session, 512, &tmp));
	WT_ERR(__wt_buf_fmt(session, tmp, "index:%s:", tablename));

	/* Find matching indices. */
	WT_ERR(__wt_metadata_cursor(session, NULL, &cursor));
	cursor->set_key(cursor, tmp->data);
	if ((ret = cursor->search_near(cursor, &cmp)) == 0 && cmp < 0)
		ret = cursor->next(cursor);
	for (i = 0; ret == 0; i++, ret = cursor->next(cursor)) {
		WT_ERR(cursor->get_key(cursor, &uri));
		name = uri;
		if (!WT_PREFIX_SKIP(name, tmp->data))
			break;

		/* Is this the index we are looking for? */
		match = idxname == NULL || WT_STRING_MATCH(name, idxname, len);

		/*
		 * Ensure there is space, including if we have to make room for
		 * a new entry in the middle of the list.
		 */
		WT_ERR(__wt_realloc_def(session, &table->idx_alloc,
		    WT_MAX(i, table->nindices) + 1, &table->indices));

		/* Keep the in-memory list in sync with the metadata. */
		cmp = 0;
		while (table->indices[i] != NULL &&
		    (cmp = strcmp(uri, table->indices[i]->name)) > 0) {
			/* Index no longer exists, remove it. */
			__wt_free(session, table->indices[i]);
			memmove(&table->indices[i], &table->indices[i + 1],
			    (table->nindices - i) * sizeof(WT_INDEX *));
			table->indices[--table->nindices] = NULL;
		}
		if (cmp < 0) {
			/* Make room for a new index. */
			memmove(&table->indices[i + 1], &table->indices[i],
			    (table->nindices - i) * sizeof(WT_INDEX *));
			table->indices[i] = NULL;
			++table->nindices;
		}

		if (!match)
			continue;

		if (table->indices[i] == NULL) {
			WT_ERR(cursor->get_value(cursor, &idxconf));
			WT_ERR(__wt_calloc_def(session, 1, &idx));
			WT_ERR(__wt_strdup(session, uri, &idx->name));
			WT_ERR(__wt_strdup(session, idxconf, &idx->config));
			WT_ERR(__open_index(session, table, idx));

			table->indices[i] = idx;
			idx = NULL;
		}

		/* If we were looking for a single index, we're done. */
		if (indexp != NULL)
			*indexp = table->indices[i];
		if (idxname != NULL)
			break;
	}
	WT_ERR_NOTFOUND_OK(ret);

	/* If we did a full pass, we won't need to do it again. */
	if (idxname == NULL) {
		table->nindices = i;
		table->idx_complete = 1;
	}

err:	__wt_scr_free(&tmp);
	if (idx != NULL)
		__wt_schema_destroy_index(session, idx);
	if (cursor != NULL)
//.........这里部分代码省略.........

int main(void)
{
	int count, exact, ret;
	WT_CONNECTION *conn;
	WT_SESSION *session;
	WT_CURSOR *cursor;
	CUSTOMER cust, *custp, cust_sample[] = {
		{ 0, "Professor Oak", "LeafGreen Avenue", "123-456-7890" },
		{ 0, "Lorelei", "Sevii Islands", "098-765-4321" },
		{ 0, NULL, NULL, NULL }
	};
	CALL call, *callp, call_sample[] = {
		{ 0, 32, 1, 2, "billing", "unavailable" },
		{ 0, 33, 1, 2, "billing", "available" },
		{ 0, 34, 1, 2, "reminder", "unavailable" },
		{ 0, 35, 1, 2, "reminder", "available" },
		{ 0, 0, 0, 0, NULL, NULL }
	};

	ret = wiredtiger_open(home, NULL, "create", &conn);
	if (ret != 0) {
		fprintf(stderr, "Error connecting to %s: %s\n",
		    home, wiredtiger_strerror(ret));
		return (1);
	}
	/* Note: further error checking omitted for clarity. */

	/*! [call-center work] */
	ret = conn->open_session(conn, NULL, NULL, &session);

	/*
	 * Create the customers table, give names and types to the columns.
	 * The columns will be stored in two groups: "main" and "address",
	 * created below.
	 */
	ret = session->create(session, "table:customers",
	    "key_format=r,"
	    "value_format=SSS,"
	    "columns=(id,name,address,phone),"
	    "colgroups=(main,address)");

	/* Create the main column group with value columns except address. */
	ret = session->create(session,
	    "colgroup:customers:main", "columns=(name,phone)");

	/* Create the address column group with just the address. */
	ret = session->create(session,
	    "colgroup:customers:address", "columns=(address)");

	/* Create an index on the customer table by phone number. */
	ret = session->create(session,
	    "index:customers:phone", "columns=(phone)");

	/* Populate the customers table with some data. */
	ret = session->open_cursor(
	    session, "table:customers", NULL, "append", &cursor);
	for (custp = cust_sample; custp->name != NULL; custp++) {
		cursor->set_value(cursor,
		    custp->name, custp->address, custp->phone);
		ret = cursor->insert(cursor);
	}
	ret = cursor->close(cursor);

	/*
	 * Create the calls table, give names and types to the columns.  All the
	 * columns will be stored together, so no column groups are declared.
	 */
	ret = session->create(session, "table:calls",
	    "key_format=r,"
	    "value_format=qrrSS,"
	    "columns=(id,call_date,cust_id,emp_id,call_type,notes)");

	/*
	 * Create an index on the calls table with a composite key of cust_id
	 * and call_date.
	 */
	ret = session->create(session, "index:calls:cust_date",
	    "columns=(cust_id,call_date)");

	/* Populate the calls table with some data. */
	ret = session->open_cursor(
	    session, "table:calls", NULL, "append", &cursor);
	for (callp = call_sample; callp->call_type != NULL; callp++) {
		cursor->set_value(cursor, callp->call_date, callp->cust_id,
		    callp->emp_id, callp->call_type, callp->notes);
		ret = cursor->insert(cursor);
	}
	ret = cursor->close(cursor);

	/*
	 * First query: a call arrives.  In SQL:
	 *
	 * SELECT id, name FROM Customers WHERE phone=?
	 *
	 * Use the cust_phone index, lookup by phone number to fill the
	 * customer record.  The cursor will have a key format of "S" for a
	 * string because the cust_phone index has a single column ("phone"),
	 * which is of type "S".
	 *
	 * Specify the columns we want: the customer ID and the name.  This
//.........这里部分代码省略.........

int
util_dump(WT_SESSION *session, int argc, char *argv[])
{
	WT_CURSOR *cursor;
	WT_DECL_RET;
	size_t len;
	int ch, i;
	char *checkpoint, *config, *p, *simpleuri, *uri;
	bool hex, json, reverse;

	hex = json = reverse = false;
	checkpoint = config = simpleuri = uri = NULL;
	cursor = NULL;
	while ((ch = __wt_getopt(progname, argc, argv, "c:f:jrx")) != EOF)
		switch (ch) {
		case 'c':
			checkpoint = __wt_optarg;
			break;
		case 'f':			/* output file */
			if (freopen(__wt_optarg, "w", stdout) == NULL)
				return (util_err(
				    session, errno, "%s: reopen", __wt_optarg));
			break;
		case 'j':
			json = true;
			break;
		case 'r':
			reverse = true;
			break;
		case 'x':
			hex = true;
			break;
		case '?':
		default:
			return (usage());
		}
	argc -= __wt_optind;
	argv += __wt_optind;

	/* -j and -x are incompatible. */
	if (hex && json) {
		fprintf(stderr,
		    "%s: the -j and -x dump options are incompatible\n",
		    progname);
		goto err;
	}

	/* The remaining argument is the uri. */
	if (argc < 1 || (argc != 1 && !json))
		return (usage());

	if (json &&
	    (dump_json_begin(session) != 0 ||
	    dump_prefix(session, hex, json) != 0))
		goto err;

	for (i = 0; i < argc; i++) {
		if (json && i > 0)
			if (dump_json_separator(session) != 0)
				goto err;
		free(uri);
		free(simpleuri);
		uri = simpleuri = NULL;

		if ((uri = util_uri(session, argv[i], "table")) == NULL)
			goto err;

		len =
		    checkpoint == NULL ? 0 : strlen("checkpoint=") +
		    strlen(checkpoint) + 1;
		len += strlen(json ? "dump=json" :
		    (hex ? "dump=hex" : "dump=print"));
		if ((config = malloc(len + 10)) == NULL)
			goto err;
		if (checkpoint == NULL)
			config[0] = '\0';
		else {
			(void)strcpy(config, "checkpoint=");
			(void)strcat(config, checkpoint);
			(void)strcat(config, ",");
		}
		(void)strcat(config, json ? "dump=json" :
		    (hex ? "dump=hex" : "dump=print"));
		if ((ret = session->open_cursor(
		    session, uri, NULL, config, &cursor)) != 0) {
			fprintf(stderr, "%s: cursor open(%s) failed: %s\n",
			    progname, uri, session->strerror(session, ret));
			goto err;
		}

		if ((simpleuri = strdup(uri)) == NULL) {
			(void)util_err(session, errno, NULL);
			goto err;
		}
		if ((p = strchr(simpleuri, '(')) != NULL)
			*p = '\0';
		if (dump_config(session, simpleuri, cursor, hex, json) != 0)
			goto err;

		if (dump_record(cursor, reverse, json) != 0)
//.........这里部分代码省略.........

int
main(int argc, char *argv[])
{
	FILE *fp;
	WT_CONNECTION *conn;
	WT_CURSOR *cursor;
	WT_SESSION *session;
	WT_RAND_STATE rnd;
	uint64_t key;
	uint32_t absent, count, timeout;
	int ch, status, ret;
	pid_t pid;
	char *working_dir;

	if ((progname = strrchr(argv[0], DIR_DELIM)) == NULL)
		progname = argv[0];
	else
		++progname;

	working_dir = NULL;
	timeout = 10;
	while ((ch = __wt_getopt(progname, argc, argv, "h:t:")) != EOF)
		switch (ch) {
		case 'h':
			working_dir = __wt_optarg;
			break;
		case 't':
			timeout = (uint32_t)atoi(__wt_optarg);
			break;
		default:
			usage();
		}
	argc -= __wt_optind;
	argv += __wt_optind;
	if (argc != 0)
		usage();

	testutil_work_dir_from_path(home, 512, working_dir);
	testutil_make_work_dir(home);

	/*
	 * Fork a child to insert as many items.  We will then randomly
	 * kill the child, run recovery and make sure all items we wrote
	 * exist after recovery runs.
	 */
	if ((pid = fork()) < 0)
		testutil_die(errno, "fork");

	if (pid == 0) { /* child */
		fill_db();
		return (EXIT_SUCCESS);
	}

	/* parent */
	__wt_random_init(&rnd);
	/* Sleep for the configured amount of time before killing the child. */
	printf("Parent: sleep %" PRIu32 " seconds, then kill child\n", timeout);
	sleep(timeout);

	/*
	 * !!! It should be plenty long enough to make sure more than one
	 * log file exists.  If wanted, that check would be added here.
	 */
	printf("Kill child\n");
	if (kill(pid, SIGKILL) != 0)
		testutil_die(errno, "kill");
	waitpid(pid, &status, 0);

	/*
	 * !!! If we wanted t