Empty values in comma-delimited data causes parsing issues

[marklit]

I have data files where each line contains a record of data delimited by only a comma. There are no commas within any of the contents of the data so there shouldn't be any escaping issues. The first 22 columns of the table look as follows:

CREATE TABLE trips (
    trip_id                 UInt32,
    vendor_id               String,
    pickup_datetime         DateTime,
    dropoff_datetime        Nullable(DateTime),
    store_and_fwd_flag      Nullable(FixedString(1)),
    rate_code_id            Nullable(UInt8),
    pickup_longitude        Nullable(Float64),
    pickup_latitude         Nullable(Float64),
    dropoff_longitude       Nullable(Float64),
    dropoff_latitude        Nullable(Float64),
    passenger_count         Nullable(UInt8),
    trip_distance           Nullable(Float64),
    fare_amount             Nullable(Float32),
    extra                   Nullable(Float32),
    mta_tax                 Nullable(Float32),
    tip_amount              Nullable(Float32),
    tolls_amount            Nullable(Float32),
    ehail_fee               Nullable(Float32),
    improvement_surcharge   Nullable(Float32),
    total_amount            Nullable(Float32),
    payment_type            Nullable(String),
    trip_type               Nullable(UInt8),
    ...
) ENGINE = Log;

When I attempt to import the data with the following command:

time (gunzip -c /home/mark/trips/trips_x*.csv.gz | \
        clickhouse-client \
        --query="INSERT INTO trips FORMAT CSV")

I'm getting an error message on one of the records:

Code: 27. DB::Exception: Cannot parse input: expected , before: -6,52,,,,,,,,,,,,,,,,,,,,\n504769614,CMT,2011-11-05 06:07:07,2011-11-05 06:12:53,N,1,-73.990061999999995,40.756630000000001,-73.996763999999999,40.73723799999999: (at row 73452711)

Row 73452710:
Column 0,   name: trip_id,               type: UInt32,                   parsed text: "504769613"
Column 1,   name: vendor_id,             type: String,                   parsed text: "VTS"
Column 2,   name: pickup_datetime,       type: DateTime,                 parsed text: "2011-11-04 13:21:00"
Column 3,   name: dropoff_datetime,      type: Nullable(DateTime),       parsed text: "2011-11-04 13:31:00"
Column 4,   name: store_and_fwd_flag,    type: Nullable(FixedString(1)), parsed text: <EMPTY>
Column 5,   name: rate_code_id,          type: Nullable(UInt8),          parsed text: "1"
Column 6,   name: pickup_longitude,      type: Nullable(Float64),        parsed text: "-74.005681999999993"
Column 7,   name: pickup_latitude,       type: Nullable(Float64),        parsed text: "40.745690000000003"
Column 8,   name: dropoff_longitude,     type: Nullable(Float64),        parsed text: "-73.996515000000002"
Column 9,   name: dropoff_latitude,      type: Nullable(Float64),        parsed text: "40.732435000000002"
Column 10,  name: passenger_count,       type: Nullable(UInt8),          parsed text: "1"
Column 11,  name: trip_distance,         type: Nullable(Float64),        parsed text: "1.3799999999999999"
Column 12,  name: fare_amount,           type: Nullable(Float32),        parsed text: "7.2999999999999998"
Column 13,  name: extra,                 type: Nullable(Float32),        parsed text: "0"
Column 14,  name: mta_tax,               type: Nullable(Float32),        parsed text: "0.5"
Column 15,  name: tip_amount,            type: Nullable(Float32),        parsed text: "2"
Column 16,  name: tolls_amount,          type: Nullable(Float32),        parsed text: "0"
Column 17,  name: ehail_fee,             type: Nullable(Float32),        ERROR: text ",,9.800000" is not like Nullable(Float32)

Is there a better way to define the table schema and/or the format parameters in the INSERT INTO statement so that the missing values don't cause such an issue when importing?

[alexey-milovidov]

CSV format doesn't treat empty as NULL to avoid ambiguity with empty strings.
We should add an option to treat empty as NULL.

Please note, that NULLs support is still shallow. We use NULLs only to import data and to convert to non-Nullable types before any further data analysis.

I suggest to use TSV (TabSeparated) format. Postgres, MySQL, Hive, etc. print NULLs as \N in this format and ClickHouse understands it without ambiguity.

[marklit]

Thanks for the explanation. I'll use sed to convert the data to TSV format.

Just in case it's helpful for anyone looking at this issue in the future this is the command I'm running:

time (gunzip -c /home/mark/trips/trips_x*.csv.gz | \
        sed 's/,/\t/g' | \
        clickhouse-client \
        --query="INSERT INTO trips FORMAT TSV")

[marklit]

Looks like I'll need to transform the blanks to \N. Switching to tabs and using TSV format for importing is raising the same issues.

[alexey-milovidov]

Yes.

Although we can introduce option to treat empty as NULL rather quickly (few days).

[marklit]

If you could that would be great. In the mean time I've put together a Python script to transform the data:

import sys


for line in sys.stdin:
    print ','.join([item if len(item.strip()) else '\N'
                    for item in line.strip().split(',')])

I'm running the import now and it seems to be running well.

for filename in /home/mark/trips/trips_x*.csv.gz; do
    gunzip -c $filename | \
        python trans.py | \
        clickhouse-client \
            --query="INSERT INTO trips FORMAT CSV"
done

[Imaskar]

This problem occurs specifically for floating point types (Nullable(Float64), Nullable(Float32)). I don't think there is any ambiguity what empty floating point value means. Also this problem occurs for version 1.1.54343, but not for 1.1.54310.

[anjia0532]

my python script

https://gist.github.com/anjia0532/6db48b0886d91d9a663e5a9fd19f2aaa

src csv

val1,val2,val3
aa,bb,cc
a
a,bb,cc
aa
,bb,cc
aa,
bb,cc
a\a,bb,cc

python cmd python clean_csv.py --src=src.csv --dest=dest.csv --chunksize=50000 --cols --encoding=utf-8 --delimiter=,

dest csv

val1,val2,val3
aa,bb,cc
aa,bb,cc
aa,bb,cc
aa,bb,cc
aa,bb,cc

[filimonov]

#6055

[tavplubix]

Resolved in #5625