In this quickstart, we will install Verdict using its public docker image. Using included scripts, we will connect to 100GB test data stored in our Amazon S3 bucket. Our demo will use Facebook Presto as an underlying analytics engine, which we will also install using a docker image.



  1. A machine with 64GB or more memory
  2. docker (get here) and docker-compose (get here).
  3. AWS access key and its secret credential, which can be obtained freely from the official page.


Most computational resource is for running Presto. Verdict itself is quite light-weight.


First, set your AWS access key and is credential to environment variables:

export AWS_ACCESS_KEY_ID=<your access key>
export AWS_SECRET_ACCESS_KEY=<your secret credential>

Second, pull and run docker containers:

curl -s https://raw.githubusercontent.com/verdict-project/verdict/master/docker-compose-64gb.yaml \
    | docker-compose -f - up

Running docker ps will show two containers named docker-verdict and docker-presto.

Third, create external tables that connect to 100GB dataset:

docker exec docker-verdict define_presto_tables_and_samples.sh


First, open the python shell in docker:

docker exec -it docker-verdict python

Then, make a connection to Presto with verdict.

import verdict
v = verdict.presto(presto_host='presto')

Info method

To see the tables indexed by verdict, use the info() method.

# {   'Registered Tables': [   'hive.tpch_sf100.orders',
#                              'hive.tpch_sf1.orders',
#                              'hive.tpch_sf100.lineitem',
#                              'hive.tpch_sf1.lineitem']}

You can pass an argument to info() to see more information about it.

# {   'Column Names and Types': {   'l_comment': 'varchar',
#                                   'l_commitdate': 'date',
#                                   'l_discount': 'double',
#                                   ...
#                                   's_nationkey': 'bigint',
#                                   's_phone': 'varchar'},
#     'Samples': [   's9487fcfadd71477ead92b02cf587e525_rowid',
#                    's63d739590a784d959b3d1e8694ef5e3al_orderkey'],
#     'Row Count': 600037902}

From the above output, you can see two samples have been created for hive.tpch_sf100.lineitem. Verdict uses these samples (automatically) to speed up its query processing. You may be curious how they are created, but before describing how to create them, let’s first run some queries.

Run Queries

Traditional Mode

To issue queries in the traditional mode, we use the sql() method. For example,

v.sql("select count(*) from hive.tpch_sf100.lineitem")
#           c1
# 0  597536768

v.sql("select count(*) from hive.tpch_sf100.lineitem where l_linestatus = 'F'")
#           c1
# 0  299372544

The above queries will return answers almost instantly.

Stream Mode

To run queries in the stream mode, use sql_stream() method. This method returns an iterator from which you can retrieve a series of answers. For example,

itr = v.sql_stream("select count(*) from hive.tpch_sf100.lineitem where l_linestatus = 'F'")

for result in itr:

To see more example queries for both traditional and stream modes, see Example Queries.

Bypass Queries

Finally, you can issue any queries directly to the backend engine (Presto here) by sending a query to sql() method with the prefix bypass. For example,

# this will return an exact answer, but will take longer
v.sql("bypass select count(*) from hive.tpch_sf100.lineitem where l_linestatus = 'F'")
#        _col0
# 0  299979732

The answers are almost identical, but it takes longer this time since the query is directly processed by the backend engine.

You can also issue metadata queries or DDL queries by prefixing bypass.

# regular presto query
v.sql("bypass show catalogs")

# The above method will return this:
#   Catalog
# 0    hive
# 1     jmx
# 2  memory
# 3  system
# 4    tpch

v.sql("bypass show schemas in hive")
#                Schema
# 0             default
# 1  information_schema
# 2             verdict
# 3            tpch_sf1
# 4          tpch_sf100

v.sql("bypass show tables in hive.tpch_sf100")
#       Table
# 0  lineitem
# 1    orders
# 2  partsupp