Coordinator cannot load mysql and postgres extensions for sql ingest

[kevin-fibich]

Affected Version

0.21.1
but everything up to master seems affected

Description

mysql and postgres sql ingest require the corresponding metadata extensions to be loaded, which results in an error because both modules are extenting SQLMetadataStorageDruidModule and calling its configure function which includes:

JsonConfigProvider.bind(binder, "druid.audit.manager", SQLAuditManagerConfig.class);

Since this bind can only be called once, the second metadata extension loaded encounters an exception at this point and fails to load, which in turn makes the sql ingestion / firehose module of that extension not load.

Steps to reproduce

put

druid.extensions.loadList=["postgres-metadata-storage","mysql-metadata-storage"]

in your config and start coordinator.

Stacktrace

Exception in thread "main" java.lang.RuntimeException: com.google.inject.CreationException: Unable to create injector, see the following errors:
coordinator |
coordinator | 1) A binding to com.google.common.base.Supplier<org.apache.druid.server.audit.SQLAuditManagerConfig> was already configured at org.apache.druid.guice.JsonConfigProvider.bind(JsonConfigProvider.java:151) (via modules: com.google.inject.util.Modules$OverrideModule -> org.apache.druid.metadata.storage.postgresql.PostgreSQLMetadataStorageModule).
coordinator | at org.apache.druid.guice.JsonConfigProvider.bind(JsonConfigProvider.java:151) (via modules: com.google.inject.util.Modules$OverrideModule -> org.apache.druid.metadata.storage.mysql.MySQLMetadataStorageModule)
coordinator |
coordinator | 2) A binding to org.apache.druid.server.audit.SQLAuditManagerConfig was already configured at org.apache.druid.guice.JsonConfigProvider.bind(JsonConfigProvider.java:152) (via modules: com.google.inject.util.Modules$OverrideModule -> org.apache.druid.metadata.storage.postgresql.PostgreSQLMetadataStorageModule).
coordinator | at org.apache.druid.guice.JsonConfigProvider.bind(JsonConfigProvider.java:152) (via modules: com.google.inject.util.Modules$OverrideModule -> org.apache.druid.metadata.storage.mysql.MySQLMetadataStorageModule)

[dulu98Kurz]

I can reproduce this issue in 0.23.0, 0.20.2
It seems the cause of this issue is from druid.extensions.loadList=["postgres-metadata-storage","mysql-metadata-storage"] , for obvious reasons Druid never expect user to use 2 type of metadata-storage, and when you try to do that, it causes conflict in binding phase because we are binding the shared classes(such as AuditManager.class and AuditManagerProvider.class) twice.
By looking at document here , sql ingestion ties metadata-storage extension for same type of RDBMS, which imply an assumption that when users want to ingest from one RDBMS(for example,MySQL), user would reuse same type of RDBMS for metadata-storage, this poor assumption never considered the case when people use Druid to ingest from more than 1 type of RDBMS.

@a2l007 correct me if I`m wrong please, from my understanding using sql ingest to ingest data from more than 1 type of RDBMS is likely unsupported and unintended in Druid right now.

A workaround for this issue can be :

1. If it`s a one time operation, we can dump the table from mysql and/or postgres into CSV file, then import to druid from CSV
2. Or if it`s a periodically operation, stream the content to message queues like Kafka or Kinesis, then use Kafka/Kinesis indexer

[FrankChen021]

The metadata storage is designed to store meta data such as segments/datasources in Druid, and it's supposed to have 1 storage at any time.

So I think the SQL ingestion should be indepdent from the metadata storage module.

I look through the code, current PostgresqlFirehoseDatabaseConnector and MySQLFirehoseDatabaseConnector does not use any APIs of PostgreSQL or MySQL driver, so one way is to move these classes to 'server' module. Once PostgreSQL or MySQL driver is placed under the lib directory, the sql ingestion should work.