MongoDB queries

Query objects describe operations independently of a connection. A collection reference contains only a database and collection name; choose the deployment when creating an executor.

from bondi.mongodb import MongoAggregateQuery, MongoCollectionRef, MongoFindQuery

business = MongoCollectionRef("demo", "business")
query = MongoFindQuery(
    business,
    {"stars": {"$gte": 4}},
    projection={"name": 1, "stars": 1, "_id": 0},
    sort=(("stars", -1), ("name", 1)),
    limit=10,
)

Use native MongoDB filter syntax. sort is an ordered tuple of pairs; order matters for compound sorts. Omit limit for unlimited results, rather than using zero.

Aggregation pipelines

pipeline = MongoAggregateQuery(
    business,
    [
        {"$match": {"stars": {"$gte": 4}}},
        {"$group": {"_id": "$city", "businesses": {"$sum": 1}}},
        {"$sort": {"businesses": -1}},
    ],
)

The pipeline stays native and ordered. The toolkit validates basic structure; MongoDB validates operators and expressions. A pipeline can include write stages, so use database permissions appropriate to your experiment.

Edit and save a query

query.limit = 20
query.validate()
text = query.to_json()
restored = MongoFindQuery.from_json(text)
assert restored.fingerprint() == query.fingerprint()

Queries are mutable. Construction copies filters, projections and pipelines, but later edits are ordinary mutations. Do not change them during execution. Fingerprints identify serialized representations, not semantic equivalence.

Metadata and find options use ordinary JSON. Filters and entire aggregation pipelines use Canonical Extended JSON to preserve BSON values. Find projections must be JSON-serializable. Dates normalize to BSON millisecond precision.

See workload persistence to save many queries, or the MongoDB query reference for validation and codec details.