MongoDB

MongoDB suits the parts of an industrial application that do not fit a fixed schema: batch records, recipe documents, per-device configuration blobs, inspection results with varying fields, event payloads.

Configuration

Field	Description
Name	Connection identifier.
Host / Port	Server address; 27017 by default.
Database	Default database for operations on this connection.
Username / Password	Credentials. Encrypted at rest, never returned to the browser.
Auth source	The database holding the user, usually admin.
URI	Full connection string, as an alternative to the fields above. Use it for replica sets, SRV records and Atlas.
AI username / password	Optional read-only credential for the AI assistant.

Using a URI

For anything beyond a single host, supply the URI:

mongodb://user:pass@host1:27017,host2:27017/mydb?replicaSet=rs0&authSource=admin
mongodb+srv://user:pass@cluster.example.mongodb.net/mydb?retryWrites=true&w=majority

The driver's own topology discovery (SDAM) then handles failover, and QUBIQ's connection monitor reads its events for health — so a replica-set election shows up as a health blip, not an outage.

Operations

The MongoDB pipeline node and the scripting API expose:

Operation	Notes
find	Filter, projection, sort, limit, skip.
insert	One document or many.
update	Filter plus update document; optional upsert.
delete	Filter-based.

find — filter

{ "line": "L1", "ts": { "$gte": "2026-01-01T00:00:00Z" }, "status": "complete" }

update — filter and update

{ "filter": { "batchId": "B-1042" },
  "update": { "$set": { "status": "released", "releasedBy": "operator1" } } }

Browsing

Browse (needs View connections) lists databases and collections, and samples documents so you can see the actual field names — which, in a schemaless store, is the only reliable way to know them.

Writing from pipelines

MongoDB writes issued by a pipeline are covered by store-and-forward: buffered while the server is unreachable, replayed in order, quarantined if they fail permanently.

Because replay is possible, make writes idempotent where you can. An update with a filter is naturally idempotent; a bare insert is not. Use a deterministic _id or an upsert when the same logical event might be replayed.

Indexing

Two indexes cover most industrial usage:

• The time field, for range queries driving trends and reports.
• The equipment or line identifier, usually compound with time: { line: 1, ts: -1 }.

Without them, a "last 24 hours for line 1" query becomes a collection scan, and it will be slow exactly when the collection has grown enough to matter.

Choosing MongoDB over SQL

Fits MongoDB	Fits SQL
Documents whose fields vary by product or recipe	Fixed, relational records
Nested structures you read and write whole	Data joined across many tables
Rapidly evolving schema during commissioning	Reporting tools that expect SQL
Event payloads with arbitrary detail	Anything requiring multi-row transactions

For tag time-series, use neither — use the historian, which is built for it.

Security

• Create a dedicated user with rights on one database, not a cluster-wide root account.
• Set the AI credential to a read-only user if the assistant may touch this connection.
• Enable TLS in the URI (tls=true) on any network you do not fully control.
• Filters coming from user input are treated as data, not merged into a query string — but validate ranges and expected shapes anyway.

Troubleshooting

Symptom	Check
Authentication fails	authSource — the user usually lives in admin, not in the target database.
Connects to a replica set then times out	Members advertise internal hostnames the QUBIQ host cannot resolve. Fix DNS or use the SRV URI.
Works from a shell, not from QUBIQ	The account may be restricted by source host, or lacks rights on the specific collection.
Writes appear then vanish	Read preference pointing at a secondary with replication lag; read from the primary to confirm.
Slow finds	Missing index on the filter fields.

Next

→ QuestDB