net.box: fix fetching of schema of an old version

After 2.2.0-633-gaa0964ae1 ('net.box: fix schema fetching from 1.10/2.1
servers') net.box expects that _vcollation system view exists on a
tarantool server of 2.2.1+ version. This is however not always so: a
server may be run on a new version of tarantool, but work on a schema of
an old version.

The situation with non last schema is usual for replication cluster in
process of upgrading: all instances run on the new version of tarantool
first (no auto-upgrade is performed by tarantools in a cluster). Then
box.schema.upgrade() should be called, but the instances should be
operable even before the call.

Before the commit net.box was unable to connect a server if it is run on
a schema without _vcollation system view (say, 2.1.3), but the server
executable is of 2.2.1 version or newer.

Follows up #4307
Fixes #4691

src/box/lua/net_box.lua

@@ -53,6 +53,7 @@ local E_UNKNOWN              = box.error.UNKNOWN
 local E_NO_CONNECTION        = box.error.NO_CONNECTION
 local E_TIMEOUT              = box.error.TIMEOUT
 local E_PROC_LUA             = box.error.PROC_LUA
+local E_NO_SUCH_SPACE        = box.error.NO_SUCH_SPACE
 
 -- utility tables
 local is_final_state         = {closed = 1, error = 1}
@@ -779,6 +780,13 @@ local function create_transport(host, port, user, password, callback,
                 local status = hdr[IPROTO_STATUS_KEY]
                 local response_schema_version = hdr[IPROTO_SCHEMA_VERSION_KEY]
                 if status ~= 0 then
+                    -- No _vcollation space (server has an old
+                    -- schema version).
+                    local errno = band(status, IPROTO_ERRNO_MASK)
+                    if id == select3_id and errno == E_NO_SUCH_SPACE then
+                        peer_has_vcollation = false
+                        goto continue
+                    end
                     local body
                     body, body_end = decode(body_rpos)
                     return error_sm(E_NO_CONNECTION, body[IPROTO_ERROR_KEY])
@@ -793,6 +801,7 @@ local function create_transport(host, port, user, password, callback,
                 body, body_end = decode(body_rpos)
                 response[id] = body[IPROTO_DATA_KEY]
             end
+            ::continue::
         until response[select1_id] and response[select2_id] and
               (not peer_has_vcollation or response[select3_id])
         -- trick: response[select3_id] is nil when the key is nil

test/box-tap/gh-4691-net-box-connect-schema-2-1-3.test.lua

@@ -0,0 +1,93 @@
+#!/usr/bin/env tarantool
+
+--
+-- gh-4691: net.box fails to connect to tarantool-2.2+ server with
+-- a schema of version 2.1.3 or below (w/o _vcollation system
+-- space).
+--
+-- Tarantool does not update a schema automatically when an
+-- instance is in a replication cluster, because it may break
+-- instances that are run under an old tarantool version. It is
+-- quite usual to have non-upgraded instances for some time,
+-- because upgrade is usually performed step-by-step.
+--
+-- net.box leans on a server version from greeting to determine
+-- whether _vcollation system view should exist and reports an
+-- error if fetching of the space fails. This causes the problem:
+-- a version may be 2.2+, but _vcollation does not exists, because
+-- schema upgrade is not performed yet. The fix of gh-4691 allows
+-- the server to respond ER_NO_SUCH_SPACE for the query.
+--
+
+local test_run = require('test_run').new()
+local net_box = require('net.box')
+local tap = require('tap')
+
+local function before_all()
+    local opts = {
+        'script = "box-tap/no_auto_schema_upgrade.lua"',
+        'workdir = "box-tap/snap/2.1.3"',
+        'return_listen_uri = True',
+    }
+    local opts_str = table.concat(opts, ', ')
+    local cmd = 'create server schema_2_1_3 with %s'
+    local uri = test_run:cmd(cmd:format(opts_str))
+    test_run:cmd('start server schema_2_1_3')
+
+    -- Create 'test' space with 'unicode_ci' index part.
+    --
+    -- We need it to verify that net.box will expose collation_id
+    -- for an index key part when collation names information is
+    -- not available.
+    --
+    -- Note: read_only = false on reconfiguration does not lead to
+    -- a schema upgrading.
+    test_run:eval('schema_2_1_3', ([[
+        box.cfg{read_only = false}
+        box.schema.create_space('test')
+        box.space.test:create_index('pk', {parts =
+            {{field = 1, type = 'string', collation = 'unicode_ci'}}})
+        box.cfg{read_only = true}
+    ]]):gsub('\n', ' '))
+    return uri
+end
+
+local function test_connect_schema_2_1_3(test, uri)
+    test:plan(3)
+
+    local connection = net_box.connect(uri)
+
+    -- Connection is alive.
+    test:ok(connection:is_connected(), 'connection is alive')
+
+    -- Space metainfo is correct: collation_id is exposed when
+    -- collation name information is not available.
+    local key_part = connection.space.test.index[0].parts[1]
+    test:is(key_part.collation, nil,
+        'collation names are not available')
+    test:is(type(key_part.collation_id), 'number',
+        'collation numeric ids are exposed')
+
+    connection:close()
+end
+
+local function after_all()
+    -- Drop 'test' space.
+    test_run:eval('schema_2_1_3', ([[
+        box.cfg{read_only = false}
+        box.space.test:drop()
+        box.cfg{read_only = true}
+    ]]):gsub('\n', ' '))
+
+    test_run:cmd('stop server schema_2_1_3')
+    test_run:cmd('cleanup server schema_2_1_3')
+    test_run:cmd('delete server schema_2_1_3')
+end
+
+local uri = before_all()
+local test = tap.test('gh-4691-net-box-connect-schema-2-1-3')
+test:plan(1)
+test:test('connect_schema_2_1_3', test_connect_schema_2_1_3, uri)
+after_all()
+
+os.exit(test:check() and 0 or 1)

test/box-tap/no_auto_schema_upgrade.lua

@@ -0,0 +1,30 @@
+#!/usr/bin/env tarantool
+
+--
+-- This instance is intended to be used with existing snapshot
+-- from a previous tarantool version. It does not perform
+-- automatic schema upgrade.
+--
+-- Having tarantool in this state allows us to create test cases
+-- for net.box, relay, applier connected to an instance in such
+-- state or test behaviour of the instance itself.
+--
+-- There are two ways to disable automatic schema upgrade: set
+-- 'replication' box.cfg(<...>) option or set 'read_only' option.
+-- Use the latter one, because it is simpler.
+--
+
+box.cfg({
+    listen = os.getenv('LISTEN'),
+    read_only = true,
+})
+
+-- Give 'guest' user read/write accesses to all spaces.
+--
+-- Note: reconfiguration with read_only = false does not lead to
+-- schema upgrading.
+box.cfg{read_only = false}
+box.schema.user.grant('guest', 'read, write', 'universe')
+box.cfg{read_only = true}
+
+require('console').listen(os.getenv('ADMIN'))