Empty result if all columns of primary key are part of the WHERE conditions

CrateDB version: 1.1.2 and 1.1.5

JVM version: Java 8

java version "1.8.0_131"
Java(TM) SE Runtime Environment (build 1.8.0_131-b11)
Java HotSpot(TM) 64-Bit Server VM (build 25.131-b11, mixed mode)

OS version / environment description: Ubuntu 14.04.5 LTS - amd64

Problem description: If I try to run a SELECT or DELETE statement, where all columns of the primary key are part of the WHERE condition, Crate always return an empty result or don’t delete the record.

Steps to reproduce:

Create schema and import testing data

create table statusengine_host_scheduleddowntimes_test (
    hostname string,
    entry_time timestamp,
    author_name string,
    comment_data string,
    internal_downtime_id int,
    triggered_by_id int,
    is_fixed boolean,
    duration int,
    scheduled_start_time timestamp,
    scheduled_end_time timestamp,
    was_started boolean,
    actual_start_time timestamp,
    node_name string,
    PRIMARY KEY ("hostname", "node_name", "scheduled_start_time", "internal_downtime_id")
) CLUSTERED INTO 4 shards with (number_of_replicas = '1-all');

INSERT INTO statusengine_host_scheduleddowntimes_test (hostname, entry_time, author_name, comment_data, internal_downtime_id, triggered_by_id, is_fixed, duration, scheduled_start_time, scheduled_end_time, was_started, actual_start_time, node_name)VALUES
('hplj2605dn', 1497981544, 'Anonymous', '60sekunden111', 12, 0, true, 60, 1497981638, 1497981698, true, 1497981638, 'Crowbar'),
('hplj2605dn', 1497903139, 'Anonymous', '300sekunden111', 7, 0, true, 300, 1497903501, 1497903801, true, 1497903501, 'Crowbar'),
('hplj2605dn', 1497982034, 'Anonymous', '60sekunden222', 14, 0, true, 300, 1497982068, 1497982128, true, 1497982068, 'Crowbar');

Query i like the execute, which always return nothing

select * from statusengine_host_scheduleddowntimes_test where hostname='hplj2605dn' and node_name='Crowbar' AND scheduled_start_time=1497981638 and internal_downtime_id=12;

As you can see, i use all primary key columns in the WHERE condition…

I also noticed, that a SELECT count(*) will return 1.

select Count(*) from statusengine_host_scheduleddowntimes_test where hostname='hplj2605dn' and node_name='Crowbar' AND scheduled_start_time=1497981638 and internal_downtime_id=12 limit 100;

As workaround i need to add a random column from the table as well

select * from statusengine_host_scheduleddowntimes_test where hostname='hplj2605dn' and node_name='Crowbar' AND scheduled_start_time=1497981638 and internal_downtime_id=12 and duration=60 limit 100;

Did i just miss something?

//Update Same results on a 3 node CrateDB cluster on Version 1.1.2 as on my single node dev system at 1.1.5

[seut]

@nook24 @grestocla this isssue has been fixed by https://github.com/crate/crate/commit/919fec3c4e47f61336d8db9970b42a9bb7b408ed and will be released with the next hotfix releases. thanks for reporting!

[nook24]

@seut, I’m not sure if i got you right. Did you mean that I need to define the columns in the same order in my crate table statement:

create table statusengine_host_scheduleddowntimes_test (
    hostname string,
    node_name string,
    scheduled_start_time timestamp,
    internal_downtime_id, int
    <rest of columns/>
PRIMARY KEY ("hostname", "node_name", "scheduled_start_time", "internal_downtime_id")

Or did you mean, that I need to pass the columns in the WHERE condition in the same order, as they are defined in the PK? Because in my example query, they are already in the same order as in the PK definition…

PRIMARY KEY ("hostname", "node_name", "scheduled_start_time", "internal_downtime_id")

select * from
statusengine_host_scheduleddowntimes_test where hostname='hplj2605dn'
and node_name='Crowbar' AND scheduled_start_time=1497981638
and internal_downtime_id=12;

and this is returning an empty result set.

@grestocla Nice idea. Based on this i found an other workaround. I just need to cast the internal_downtime_id from integer to integer, this works as well

select * from
statusengine_host_scheduleddowntimes_test
where hostname='hplj2605dn' and node_name='Crowbar'
AND scheduled_start_time=1497981638 and cast(internal_downtime_id as integer)=12

//Update Ordering the columns doesn’t seems to be the fix

create table statusengine_host_scheduleddowntimes_test (
    hostname string,
    node_name string,
    scheduled_start_time timestamp,
    internal_downtime_id int,
    
    entry_time timestamp,
    author_name string,
    comment_data string,
    triggered_by_id int,
    is_fixed boolean,
    duration int,
    scheduled_end_time timestamp,
    was_started boolean,
    actual_start_time timestamp,
    PRIMARY KEY ("hostname", "node_name", "scheduled_start_time", "internal_downtime_id")
) CLUSTERED INTO 4 shards with (number_of_replicas = '1-all');

Same behavior… Looks like CAST() or add a column which is not defined the the PK is the only workaround at the moment

//Update 2 @seut Sorry, missed the INSERT INTO part… Yes, looks like this is working as well 😃

INSERT INTO statusengine_host_scheduleddowntimes_test (hostname, node_name, scheduled_start_time, internal_downtime_id, entry_time, author_name, comment_data, triggered_by_id, is_fixed, duration, scheduled_end_time, was_started, actual_start_time)VALUES
('hplj2605dn', 'Crowbar', 1497981638, 12, 1497981544, 'Anonymous', '60sekunden111', 0, true, 60,  1497981698, true, 1497981638),
('hplj2605dn', 'Crowbar', 1497903501, 7, 1497903139, 'Anonymous', '300sekunden111', 0, true, 300,  1497903801, true, 1497903501),
('hplj2605dn', 'Crowbar', 1497982068, 14, 1497982034, 'Anonymous', '60sekunden222', 0, true, 300, 1497982128, true, 1497982068);