Details
-
Bug
-
Status: Resolved
-
Normal
-
Resolution: Fixed
-
None
-
Cassandra 3.10 dev environment running on a MacOS Sierra
-
Normal
Description
Overview
Querying duration related data from the cqlsh prompt results in an error.
Consider the following create table and insert statement.
CREATE TABLE duration_test (
primary_key text,
col20 duration,
PRIMARY KEY (primary_key)
);
INSERT INTO duration_test (primary_key, col20) VALUES ('primary_key_example', 1y5mo89h4m48s);
On executing a select query on col20 in cqlsh I get an error "Failed to format value '"\x00\xfe\x02GS\xfc\xa5\xc0\x00' : 'ascii' codec can't decode byte 0xfe in position 2: ordinal not in range(128)"
Select col20 from duration_test;
Investigation
On investigating this further I found that the current python Cassandra driver used found in lib/cassandra-driver-internal-only-3.7.0.post0-2481531.zip does not seem to support duration data type. This was added in Jan this year https://github.com/datastax/python-driver/pull/689.
So I downloaded the latest driver release https://github.com/datastax/python-driver/releases/tag/3.9.0. I embedded the latest driver into cassandra-driver-internal-only-3.7.0.post0-2481531.zip. This fixed the driver related issue but there was still a formatting issue.
I then went on to modify the format_value_duration methos in the pylib/cqlshlib/formatting.py. Diff posted below
@formatter_for('Duration') def format_value_duration(val, colormap, **_): - buf = six.iterbytes(val) - months = decode_vint(buf) - days = decode_vint(buf) - nanoseconds = decode_vint(buf) - return format_python_formatted_type(duration_as_str(months, days, nanoseconds), colormap, 'duration') + return format_python_formatted_type(duration_as_str(val.months, val.days, val.nanoseconds), colormap, 'duration')
This resulted in fixing the issue and duration types are now correctly displayed.
Happy to fix the issue if I can get some guidance on:
- If this is a valid issue. Tried searching JIRA but did not find anything reported.
- If my assumptions are correct i.e. this is actually a bug
- how to package the new driver into the source code.