I try to import a CSV file into a PostgreSQL database that runs on a remote server. I use psql via SSH.

AIS_DB=# CREATE TABLE bet_srk_2008 (
AIS_DB(# Unique_VesselTraject numeric,
AIS_DB(# AISTrajectPositionTimestampLocal timestamp,
AIS_DB(# RandomVesselAISKey varchar,
AIS_DB(# AISTrajectPositionSpeedOverGround varchar,
AIS_DB(# AISTrajectPositionCourseOverGround varchar,
AIS_DB(# VesselAISLength varchar,
AIS_DB(# VesselAISBeam varchar,
AIS_DB(# VesselAISMaximumDraught varchar,
AIS_DB(# VesselTypeAISDescription varchar,
AIS_DB(# BOZ varchar,
AIS_DB(# AISTrajectPositionLatitude numeric,
AIS_DB(# AISTrajectPositionLongitude numeric,
AIS_DB(# TrajectAISDestination varchar
AIS_DB(# );  


AIS_DB=# \COPY bet_srk_2008 FROM '/home/thecaptain/Export2008.csv' DELIMITERS ';'CSV HEADER;  

ERROR: invalid input syntax for type numeric: "51,2628135681152" CONTEXT: COPY bet_srk_2008, line 2, column aistrajectpositionlatitude: "51,2628135681152"

I think the problem is that the data uses a comma instead of point as a decimal. Is there a way to circumvent this in the import process?
The datasets have 100 million+ lines and a find and replace is making nano run out of memory.
If I use VARCHAR instead of NUMERIC it won't let me ST_MAKEPOINT it to the geom column.

This is the whole text:

CREATE TABLE bet_srk_2008 (
Unique_VesselTraject numeric,
AISTrajectPositionTimestampLocal timestamp,
RandomVesselAISKey varchar,
AISTrajectPositionSpeedOverGround varchar,
AISTrajectPositionCourseOverGround varchar,
VesselAISLength varchar,
VesselAISBeam varchar,
VesselAISMaximumDraught varchar,
VesselTypeAISDescription varchar,
BOZ varchar,
AISTrajectPositionLatitude numeric,
AISTrajectPositionLongitude numeric,
TrajectAISDestination varchar

\COPY bet_srk_2008 FROM '/home/thecaptain/Export2008.csv' DELIMITERS 

ALTER TABLE bet_srk_2008 ADD COLUMN geom geometry (POINT, 4326 );

UPDATE bet_srk_2008 SET geom 
AISTrajectPositionLatitude) ,4326);

CREATE INDEX bet_srk_2008_index ON bet_srk_2008 USING GIST ( geom );

1 Answer 1


Ok I found a solution for anybody that has the same problem.
I am still interested if somebody has a solution without having to alter the dataset prior to importing.

I use Linux as a operating system on desktop and servers. I changed the comma to point:

sed -i 's/,/./g' Export2008.csv
  • Note that you should be able to \COPY bet_srk_2008 FROM PROGRAM 'sed -i 's/,/./g' Export2008.csv';. Alternatively, you can a) setup a BEFORE INSERT trigger to REPLACE(NEW."AISTrajectPositionLongitude", ',', '.'), (for both columns, of course) or b) run that same string replace (and a cast to FLOAT) in your UPDATE statement within ST_MakePoint. The trigger would likely be the least performant option, and using sed the fastest, but that's only an educated guess.
    – geozelot
    Apr 12, 2020 at 11:21
  • @geozelot the trigger or the cast would work if the lat/long columns were Text.
    – JGH
    Apr 12, 2020 at 11:44
  • Another option, if you can replace the , before (for example if you have a text and don't want to remove all commas from it), is to load the csv to a temp table, then to load your table by reading/converting from the temp table.
    – JGH
    Apr 12, 2020 at 11:48
  • @JGH since CSV input is text per se, and xx,yyy no valid format, I assumed that as given (also I didn't read the table definitions...).
    – geozelot
    Apr 12, 2020 at 12:19
  • And yet another option would be to have a VIEW with all columns defined as TEXT and an INSTEAD OF trigger that inserts cast values into the actual table. But in any case, I'd go for the piped input from sed...
    – geozelot
    Apr 12, 2020 at 12:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.