Hello,

Just a suggestion,  if you think the issue is related to the size of the data,  then you could use a random data generator to test it.  We were facing a similar issue with shareable data for testing.  So I wrote a program called tqgen in Golang (https://github.com/niocs/tqgen). 

It can be used to generate large amounts of csv data for testing,  and to share the data,  we just need to share the command we used to generate it,  and the other person can run it to generate the exact same data. 

It produces datewise fictional trade/quote data for fictional stocks.  It puts each date's data into a separate file, and each file is around 128MB,  so if we run it for a daterange of 1.5 years,  it should generate around 70GB of data.  It takes a while to finish because the data is not completely random and it is related to other rows.

To generate, you could run:

    $ git clone https://github.com/niocs/tqgen

    $ cd tqgen

    $ go build tqgen.go

    $ ./tqgen  --OutFilePat  /path/to/data/tq.YYYYMMDD.csv  --NumStk 1000 --Seed 234532 --DateBeg 20200101 --DateEnd 20210831

I usually load them one by one so that I could record load time stats, etc,  but you could use this bash script (https://github.com/niocs/hcsv/blob/master/hcat) to concatenate them without repeating the headers,  and then load to MonetDB. (I will later add a parameter to tqgen to save to a single file)

To do that,

    $ git clone https://github.com/niocs/hcsv

    $ hcsv/hcat  /path/to/data/tq.20*.csv > /path/to/data/tq.big.csv

If this causes the same issue,  then others could regenerate the same data to try.  Hope this helps.

Thanks,

Sahas

On Jul 14, 2016 10:20 AM, "Anthony Damico" <ajdamico@gmail.com> wrote:

can you winnow down the problem enough to construct a reproducible example?  thanks

On Thursday, July 14, 2016, Stefano Fioravanzo <fioravanzos@gmail.com> wrote:

Hello,

No the data is not public, and testing in on another machine is not an easy option right now.

Regards,

Stefano 

On 14 Jul 2016, at 15:49, Anthony Damico <ajdamico@gmail.com> wrote:

hi, any chance these data are public?  if not, can you test on a separate box and see if the exact same thing happens or if it's hardware-specific?

On Thursday, July 14, 2016, Stefano Fioravanzo <fioravanzos@gmail.com> wrote:

Hello,

I currently have a table with 307524407 rows and 52  columns, it was loaded from a 78GB cvs file, but I do not know how much space the table actually takes on disk.
If I query the table with a simple query like:

        select count(*) from big_table where field =1;

mserver5 crashes with ‘segmentation fault'
I tried also the following query (adding the LIMIT statement):
        select count(*) from farmap_movimento where farmacia_id =1 limit 10;

and the output from mserver5 was:
        mserver5: gdk_select.c:869: fullscan_int: Assertion `cnt < (bn)->S->capacity' failed.
        Aborted

Doing a more complex query (the one I am aiming for), with also a group by and a few aggregation functions (sum, min) on the fields, the outputs is:
        mserver5: gdk_bat.c:2066: BATsetcount: Assertion `b->S->capacity >= cnt' failed.
        Aborted

This seems to be a table size problem, so…is there anything I can do?

I am using MonetDB 5 server v11.22.0
The machine has 30GB of RAM and 60GB of swap.
Available disk space: ~50GB

Any help would be much appreciated,

Regards,

Stefano

_______________________________________________
users-list mailing list
users-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/users-list

_______________________________________________
users-list mailing list
users-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/users-list

_______________________________________________
users-list mailing list
users-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/users-list