Timing Vespa queries

When comparing different query models, we might want to compute the time difference between them. Doing that from the outside is tricky because it is hard to separate network and other non-search related costs out of the final numbers.

Luckily we can accomplish this by setting presentation.timing equal to true in the request body. Without relying on pyvespa API, we can add the keyword parameter presentation.timing in the app.query call.

from vespa.application import Vespa
from vespa.query import Query, RankProfile, OR

app = Vespa(url = "https://api.cord19.vespa.ai")

result = app.query(
    query="what is covid?",
    query_model=Query(
        match_phase=OR(),
        rank_profile=RankProfile(name="bm25")
    ),
    **{"presentation.timing": "true"}
)

result.json['timing']

{'querytime': 0.01,
 'summaryfetchtime': 0.002,
 'searchtime': 0.013000000000000001}

Such a useful feature could be part of the pyvespa API:

result = app.query(
    query="what is covid?",
    query_model=Query(
        match_phase=OR(),
        rank_profile=RankProfile(name="bm25")
    ),
    timing=True
)