`databasename` part of `jdbc:tc` URIs іs ignored when optional host:port pair is omitted

[Alexander-Kiselyov]

Note that this issue is related to behavior mentioned in #1846.

When one uses "host-less URI" as specified on the JDBC support page, e.g.: jdbc:tc:mysql:5.7.12:///mydatabase, DB name is also ignored and a default test name is used instead.
This code fragment from org.testcontainers.jdbc.ConnectionUrl#parseUrl method clearly shows how it is handled:

//In case it matches to the default pattern
Matcher dbInstanceMatcher = Patterns.DB_INSTANCE_MATCHING_PATTERN.matcher(dbHostString);
if (dbInstanceMatcher.matches()) {
	databaseHost = Optional.of(dbInstanceMatcher.group(1));
        databasePort = Optional.ofNullable(dbInstanceMatcher.group(3)).map(value -> Integer.valueOf(value));
        databaseName = Optional.of(dbInstanceMatcher.group(4));
}

then org.testcontainers.containers.JdbcDatabaseContainerProvider#newInstanceFromConnectionUrl defaults DB name to test:

final String databaseName = connectionUrl.getDatabaseName().orElse("test");

In my particular case an init SQL script, which starts with the following instruction, fails with Caused by: java.sql.SQLSyntaxErrorException: Access denied for user 'test'@'%' to 'mydatabase' error:

CREATE DATABASE IF NOT EXISTS `mydatabase`;

Aforementioned documentation page says:

Note that the hostname, port and database name will be ignored; you can leave these as-is or set them to any value.

which apparently is not the case and /// syntax, which includes DB name only, is recommended and multiple examples are provided.

I suggest to improve "host-less" URI handling by honoring DB name and clarify the documentation on this matter. Will make a PR for this if project owners are OK with such resolution.

[chadlwilson]

I just came across this too, with both 1.15.3 and 1.16.0.

Given it contradicts the documentation and the /// syntax works fine for r2dbc connections, I think this is a bug in the URL parsing?

There are also no tests for such URLs at https://github.com/testcontainers/testcontainers-java/blob/3aa5fbfa17be09f4e42b2300396890f76f508c86/modules/jdbc/src/test/java/org/testcontainers/jdbc/ConnectionUrlTest.java

This seems to be a re-prod:

import org.junit.jupiter.api.Test;
import org.testcontainers.jdbc.ConnectionUrl;

import static org.assertj.core.api.Assertions.assertThat;

public class ConnectionUrlTest {
    @Test
    void shouldParseDatabaseNameWithoutHostPort() {
        var url = ConnectionUrl.newInstance("jdbc:tc:mysql:8:///databasename");
        assertThat(url.getDatabaseType()).isEqualTo("mysql");
        assertThat(url.getImageTag()).hasValue("8");
        assertThat(url.getDatabaseHost()).isEmpty();
        assertThat(url.getDatabasePort()).isEmpty();
        assertThat(url.getDatabaseName()).hasValue("databasename"); // FAILS
    }
}

Other than specifying host/port, I found another workaround was just to add a fourth / which causes it to put / in databaseHost which is subsequently ignored by testcontainers anyway.

[chadlwilson]

Seems this is a dupe of #2544 which keeps getting closed for staleness.

[kiview]

Let's keep this issue open, thanks for providing the reproducer @chadlwilson and thanks @Alexander-Kiselyov for the detailed investigation.

Before we invest any work on this issue, is there a good reason to use the host-less URI format? Bringing the docs in line with the implementation is of course critical.

[chadlwilson]

From my perspective, I think it's less confusing when reading one's config to have host-less URIs because you don't have to see a misleading localhost:3306 which is going to be thrown away/ignored by testcontainers anyway. Probably makes the docs a bit cleaner to read where there are a lot of different examples for different DBs as well.

That's the justification in the current docs as well:

We will use /// (host-less URIs) from now on to emphasis the unimportance of the host:port pair.
From Testcontainers' perspective, jdbc:mysql:5.7.34://localhost:3306/databasename and jdbc:mysql:5.7.34:///databasename is the same URI.

[eschulma]

I like the host-less URIs as well, but until this is fixed please update the documentation. I wasted a day on this.

[odrotbohm]

Is there any reason that this is not approached (either by fixing the implementation and/or the docs)? I just spent half a day trying to find out why test was used instead of the URL-defined database name. 😕

[linghengqian]

In fact, my tests showed that testcontainers-postgresql, testcontainers-mssqlserver, testcontainers-clickhouse, and even the external org.firebirdsql:firebird-testcontainers-java did not have this problem.

This issue only occurs in the Maven module of testcontainers-mysql.