MongoDB connector can not select data from read-only node

[alexanderhw]

When I recently deployed trino in stand-alone mode and connected it to mongodb, I was unable to read data from the mongodb read-only instance node. The result indicate that the read-only instance node is not master node of mongodb. The specific error message is：

2024-08-21T21:58:25.873+0800    INFO    Query-20240821_135825_00018_swwvh-199   org.mongodb.driver.cluster      Waiting for server to become available for operation with ID 125. Remaining time: 30000 ms. Selector: ReadPreferenceServerSelector{readPreference=primary}, topology description: {type=UNKNOWN, servers=[{address=xxxxxxxxx:8635, type=UNKNOWN, state=CONNECTING, exception={com.mongodb.MongoNotPrimaryException: Command failed with error 10107 (NotMaster): 'not master' on server xxxxxxx:8635. The full response is {"operationTime": {"$timestamp": {"t": 1724248705, "i": 10}}, "ok": 0.0, "errmsg": "**not master**", "code": 10107, "codeName": "**NotMaster**", "$clusterTime": {"clusterTime": {"$timestamp": {"t": 1724248705, "i": 10}}, "signature": {"hash": {"$binary": {"base64": "J7HW9Nj7fF7lRpJJiIS/4z2zJP0=", "subType": "00"}}, "keyId": 7378427933884416004}}}}}].
2024-08-21T21:58:25.920+0800    INFO    dispatcher-query-10     io.trino.event.QueryMonitor     TIMELINE: Query 20240821_135825_00018_swwvh :: FAILED (USER_CANCELED) :: elapsed 46ms :: planning 46ms :: waiting 0ms :: scheduling 0ms :: running 0ms :: finishing 0ms :: begin 2024-08-21T21:58:25.872+08:00 :: end 2024-08-21T21:58:25.918+08:00
2024-08-21T21:58:25.930+0800    ERROR   page-buffer-client-callback-3   io.trino.operator.HttpPageBufferClient  Request to delete http://172.31.33.113:8080/v1/task/20240821_135825_00017_swwvh.1.0.0/results/0 failed java.util.concurrent.CancellationException: Task was cancelled.

I noticed that the official deployment guide mentioned that trino needs to have write permissions for information schema. Does this mean that mongodb read-only instance nodes cannot connect to trino?

If trino can connect to mongodb read-only instance nodes, do I need to add any configuration? My current mongodb connection config is:

connector.name=mongodb
mongodb.connection-url=mongodb://<user>:<pwd>@<ip>:<port>/?authSource=admin

basic environment:

• trino version: 453
• jdk version: zulu22.32.15-ca-jdk22.0.2-linux_x64
• python version: 3.7
• trino deployment mode: single machine

mongodb cluster environment:

• node1: primary
• node2: secondary
• relicaSet: replica
• node3: read-only node represent for the real mongo replica

Update:

I have read the source code of trino-mongodb connector and noticed that the MongoSession#getTableMetadata(String schemaName, String tableName) function will read configuration mongodb.schema-collection to infer the collection field structure. If the schema collection does not defined, trino needs to create this collection and dynamically maintain it, so mongodb's write permission may be required.

So will trino support maintaining similar table structures within the cluster instead of in specific data source instances? In many production scenarios, from the perspective of data security, data analysis is often connected to read-only instances.

[ErikTheBerik]

I have a very similar situation, where basically I would like to give trino only read access to the data. Since it needs to maintain the _schema document, is there any way this document could exist in a different mongo database (within the same mongo instance)? Then the document would be named something like "_db-name_schema"

[unkn0wn-kgb]

Any update on this? This seem to be a big issue for production setups.
Is it possible somehow for trino to store schema-collection somewhere else?
So it would be possible to run queries on Hidden/Delayed/Secondary replica?

Thanks