|
| 1 | +import time |
1 | 2 | import unittest
|
2 | 3 | from crate.client import connect
|
3 | 4 | from crate.client.exceptions import ProgrammingError
|
4 | 5 |
|
5 |
| -from crate.qa.tests import NodeProvider, insert_data, wait_for_active_shards, UpgradePath |
| 6 | +from crate.qa.tests import NodeProvider, insert_data, wait_for_active_shards, UpgradePath, assert_busy |
6 | 7 |
|
7 | 8 | ROLLING_UPGRADES_V4 = (
|
8 | 9 | # 4.0.0 -> 4.0.1 -> 4.0.2 don't support rolling upgrades due to a bug
|
|
30 | 31 | )
|
31 | 32 |
|
32 | 33 | ROLLING_UPGRADES_V5 = (
|
33 |
| - UpgradePath('5.0.x', '5.1.x'), |
34 |
| - UpgradePath('5.1.x', '5.2.x'), |
35 |
| - UpgradePath('5.2.x', '5.3.x'), |
36 |
| - UpgradePath('5.3.x', '5.4.x'), |
37 |
| - UpgradePath('5.4.x', '5.5.x'), |
38 |
| - UpgradePath('5.5.x', '5.6.x'), |
39 |
| - UpgradePath('5.6.x', '5.7.x'), |
40 |
| - UpgradePath('5.7.x', '5.8.x'), |
41 |
| - UpgradePath('5.8.x', '5.9.x'), |
42 |
| - UpgradePath('5.9.x', '5.10.x'), |
43 |
| - UpgradePath('5.10.x', '5.10'), |
44 |
| - UpgradePath('5.10', 'latest-nightly'), |
| 34 | + UpgradePath('5.10', 'branch:master'), |
45 | 35 | )
|
46 | 36 |
|
47 | 37 |
|
48 | 38 | class RollingUpgradeTest(NodeProvider, unittest.TestCase):
|
49 | 39 |
|
| 40 | + def _num_docs(self, cursor, schema, table): |
| 41 | + cursor.execute("select sum(num_docs) from sys.shards where schema_name = ? and table_name = ?", (schema, table)) |
| 42 | + return cursor.fetchall()[0][0] |
| 43 | + |
| 44 | + def _assert_num_docs(self, cursor, schema, table, expected_count): |
| 45 | + count = self._num_docs(cursor, schema, table) |
| 46 | + self.assertEqual(expected_count, count) |
| 47 | + |
50 | 48 | def test_rolling_upgrade_4_to_5(self):
|
51 | 49 | print("") # force newline for first print
|
52 | 50 | for path in ROLLING_UPGRADES_V4:
|
@@ -88,6 +86,10 @@ def _test_rolling_upgrade(self, path, nodes):
|
88 | 86 | }
|
89 | 87 | cluster = self._new_cluster(path.from_version, nodes, settings=settings)
|
90 | 88 | cluster.start()
|
| 89 | + replica_cluster = None |
| 90 | + if int(path.from_version.split('.')[0]) >= 5 and int(path.from_version.split('.')[1]) >= 10: |
| 91 | + replica_cluster = self._new_cluster(path.from_version, 1, settings=settings, explicit_discovery=False) |
| 92 | + replica_cluster.start() |
91 | 93 | with connect(cluster.node().http_url, error_trace=True) as conn:
|
92 | 94 | c = conn.cursor()
|
93 | 95 | c.execute("create user arthur with (password = 'secret')")
|
@@ -152,6 +154,24 @@ def _test_rolling_upgrade(self, path, nodes):
|
152 | 154 | # Add the shards of the new partition primaries
|
153 | 155 | expected_active_shards += shards
|
154 | 156 |
|
| 157 | + # Set up tables for logical replications |
| 158 | + if int(path.from_version.split('.')[0]) >= 5 and int(path.from_version.split('.')[1]) >= 10: |
| 159 | + c.execute("create table doc.x (a int) clustered into 1 shards with (number_of_replicas=0)") |
| 160 | + expected_active_shards += 1 |
| 161 | + c.execute("create publication p for table doc.x") |
| 162 | + with connect(replica_cluster.node().http_url, error_trace=True) as replica_conn: |
| 163 | + rc = replica_conn.cursor() |
| 164 | + transport_port = cluster.node().addresses.transport.port |
| 165 | + replica_transport_port = replica_cluster.node().addresses.transport.port |
| 166 | + assert 4300 <= transport_port <= 4310 and 4300 <= replica_transport_port <= 4310 |
| 167 | + rc.execute("create table doc.rx (a int) clustered into 1 shards with (number_of_replicas=0)") |
| 168 | + rc.execute("create publication rp for table doc.rx") |
| 169 | + rc.execute(f"create subscription rs connection 'crate://localhost:{transport_port}?user=crate&sslmode=sniff' publication p") |
| 170 | + assert_busy(lambda: self._assert_num_docs(rc, "doc", "x", 0)) |
| 171 | + c.execute(f"create subscription s connection 'crate://localhost:{replica_transport_port}?user=crate&sslmode=sniff' publication rp") |
| 172 | + assert_busy(lambda: self._assert_num_docs(c, "doc", "rx", 0)) |
| 173 | + expected_active_shards += 1 |
| 174 | + |
155 | 175 | for idx, node in enumerate(cluster):
|
156 | 176 | # Enforce an old version node be a handler to make sure that an upgraded node can serve 'select *' from an old version node.
|
157 | 177 | # Otherwise upgraded node simply requests N-1 columns from old version with N columns and it always works.
|
@@ -282,6 +302,25 @@ def _test_rolling_upgrade(self, path, nodes):
|
282 | 302 | c.execute("select version['created'] from information_schema.table_partitions where table_name = 't3' and values['a'] = ?", [idx])
|
283 | 303 | self.assertEqual(c.fetchall(), [[partition_version]])
|
284 | 304 |
|
| 305 | + # Ensure logical replications works |
| 306 | + if int(path.from_version.split('.')[0]) >= 5 and int(path.from_version.split('.')[1]) >= 10: |
| 307 | + with connect(replica_cluster.node().http_url, error_trace=True) as replica_conn: |
| 308 | + rc = replica_conn.cursor() |
| 309 | + |
| 310 | + # Cannot drop replicated tables |
| 311 | + with self.assertRaises(ProgrammingError): |
| 312 | + rc.execute("drop table doc.x") |
| 313 | + c.execute("drop table doc.rx") |
| 314 | + |
| 315 | + count = self._num_docs(rc, "doc", "x") |
| 316 | + count2 = self._num_docs(c, "doc", "rx") |
| 317 | + |
| 318 | + c.execute("insert into doc.x values (1)") |
| 319 | + rc.execute("insert into doc.rx values (1)") |
| 320 | + |
| 321 | + assert_busy(lambda: self._assert_num_docs(rc, "doc", "x", count + 1)) |
| 322 | + assert_busy(lambda: self._assert_num_docs(c, "doc", "rx", count2 + 1)) |
| 323 | + |
285 | 324 | # Finally validate that all shards (primaries and replicas) of all partitions are started
|
286 | 325 | # and writes into the partitioned table while upgrading were successful
|
287 | 326 | with connect(cluster.node().http_url, error_trace=True) as conn:
|
|
0 commit comments