Faster document operations with pyvespa

• Define Question Answering app
• Deploy to Vespa Cloud
• Load sample data
• Feed data
  • Asynchronous feeding - new default
  • Syncronous feeding - old default
• Conclusion

We will demonstrate that the new default implementation for document operations in pyvespa are now much faster due to async mode.

Define Question Answering app

from vespa.gallery import QuestionAnswering

app_package = QuestionAnswering()

Deploy to Vespa Cloud

import os
from vespa.package import VespaCloud

# define your own WORK_DIR and VESPA_CLOUD_USER_KEY env variables

disk_folder = os.path.join(os.getenv("WORK_DIR"), "sample_application")
vespa_cloud = VespaCloud(
    tenant="vespa-team",
    application="pyvespa-integration",
    key_content=os.getenv("VESPA_CLOUD_USER_KEY").replace(r"\n", "\n"),
    application_package=app_package,
)
app = vespa_cloud.deploy(
    instance="msmarco", 
    disk_folder=os.path.join(os.getenv("WORK_DIR"), "sample_application")
)

Deployment started in run 243 of dev-aws-us-east-1c for vespa-team.pyvespa-integration.msmarco. This may take about 15 minutes the first time.
INFO    [20:02:15]  Deploying platform version 7.452.11 and application version unknown ...
INFO    [20:02:17]  Deployment successful.
INFO    [20:02:17]  Session 43999 for tenant 'vespa-team' prepared and activated.
INFO    [20:02:17]  ######## Details for all nodes ########
INFO    [20:02:17]  h5591e.dev.aws-us-east-1c.vespa-external.aws.oath.cloud: expected to be UP
INFO    [20:02:17]  --- platform vespa/cloud-tenant-rhel8:7.452.11
INFO    [20:02:17]  --- container-clustercontroller on port 19050 has config generation 43999, wanted is 43999
INFO    [20:02:17]  h5580f.dev.aws-us-east-1c.vespa-external.aws.oath.cloud: expected to be UP
INFO    [20:02:17]  --- platform vespa/cloud-tenant-rhel8:7.452.11
INFO    [20:02:17]  --- distributor on port 19111 has config generation 43997, wanted is 43999
INFO    [20:02:17]  --- searchnode on port 19107 has config generation 43999, wanted is 43999
INFO    [20:02:17]  --- storagenode on port 19102 has config generation 43997, wanted is 43999
INFO    [20:02:17]  h5577a.dev.aws-us-east-1c.vespa-external.aws.oath.cloud: expected to be UP
INFO    [20:02:17]  --- platform vespa/cloud-tenant-rhel8:7.452.11
INFO    [20:02:17]  --- logserver-container on port 4080 has config generation 43999, wanted is 43999
INFO    [20:02:17]  h5592d.dev.aws-us-east-1c.vespa-external.aws.oath.cloud: expected to be UP
INFO    [20:02:17]  --- platform vespa/cloud-tenant-rhel8:7.452.11
INFO    [20:02:17]  --- container on port 4080 has config generation 43999, wanted is 43999
INFO    [20:02:31]  Found endpoints:
INFO    [20:02:31]  - dev.aws-us-east-1c
INFO    [20:02:31]   |-- https://qa-container.msmarco.pyvespa-integration.vespa-team.aws-us-east-1c.dev.z.vespa-app.cloud/ (cluster 'qa_container')
INFO    [20:02:32]  Installation succeeded!
Finished deployment.

Load sample data

import json, requests

sentence_data = json.loads(
    requests.get("https://data.vespa.oath.cloud/blog/qa/sample_sentence_data_100.json").text
)

batch_feed = [
    {
        "id": idx, 
        "fields": sentence
    }
    for idx, sentence in enumerate(sentence_data)
]

Feed data

Asynchronous feeding - new default

import time

start = time.time()
response = app.feed_batch(schema="sentence", batch=batch_feed)
print("{} seconds".format(time.time() - start))

1.146148920059204 seconds

Syncronous feeding - old default

start = time.time()
response = app.feed_batch(schema="sentence", batch=batch_feed, asynchronous=False)
print("{} seconds".format(time.time() - start))

68.66188383102417 seconds

Conclusion

Basic document operations are now much faster in pyvespa due to the new async implementation. This includes feed, get, update and delete operations.