respan-instrumentation-weaviate

Respan instrumentation for the Weaviate Python client v4. It emits canonical task spans for synchronous and asynchronous collection lifecycle, data, reference, vector and keyword query, aggregate, config, batch, and tenant operations.

Install

pip install respan-ai respan-instrumentation-weaviate "weaviate-client>=4.22.0,<5"

Quickstart

import weaviate
from weaviate.classes.config import Configure
from respan import Respan, workflow
from respan_instrumentation_weaviate import WeaviateInstrumentor

respan = Respan(instrumentations=[WeaviateInstrumentor()])


@workflow(name="weaviate_vector_query_workflow")
def run_query():
    client = weaviate.connect_to_local()
    collection = client.collections.create(
        "RespanDocs",
        vector_config=Configure.Vectors.self_provided(),
    )
    collection.data.insert(
        {"text": "Respan traces vector-store operations."},
        vector=[0.1, 0.2, 0.3],
    )
    result = collection.query.near_vector(
        near_vector=[0.1, 0.2, 0.3],
        limit=1,
    )
    client.close()
    return result


print(run_query())
respan.shutdown()

Pass capture_content=False to omit operation arguments and results while retaining names, status, and database attributes. Activation/deactivation and the instrument() / uninstrument() aliases are idempotent.

The adapter targets weaviate-client>=4.22.0,<5 and its v4 collection API. Generative search calls are left to an LLM-specific integration because they require chat semantics rather than vector-store task semantics.

See respan-example-projects/python/tracing/weaviate for runnable sync, async, batch, aggregate, query, mutation, and deterministic error examples.