July 2013 - users-list

aggregate function in MAL
by Scott Mathieson 05 Jul '13

05 Jul '13

Hi, I am trying to create a simple aggregate function entirely in MAL and am now struggling at the final hurdle. I have created the aggregate function such that it aggregates across an entire table but in the sub aggregate function I can't work out how to link the group to data BAT (g) to the data. The aggregate is designed to simply collect a sample of the values from the named column and return them as a pipe delimited string value. Can anyone tell me how to use the 'g' bat to join to the 'b' bat in the sub... aggregate function to get only the data rows for a specific group - currently I only get 2 rows of data per group even though I know that there should be 100k+ rows in each group? Here's the code: function bin_sample(b:bat[:oid, :any]):str; value := ""; bsamp := sample.uniform(b, 1000:wrd); barrier (h, t) := iterator.new(bsamp); tmp := "" + t; type := bat.getTailType(bsamp); test := str.startsWith(type, "str"); # Force the string type to be cast to a string or you just end up with numeric values barrier xxxx := test; tmp := str.str(t); exit xxxx; value := value + tmp; value := value + "||"; redo (h, t) := iterator.next(bsamp); exit (h, t); return value; end bin_sample; function subbin_sample(b:bat[:oid,:any_1],g:bat[:oid,:oid],e:bat[:oid,:any_2],skip_nils:bit) :bat[:oid,:str]; bn:=bat.new(:oid, :str); # check 'e' has some values - this is the list of group IDs in the head # 'g' is the group to the data BAT head ID mapping # 'b' is the data column BAT that we're aggregating over barrier (grpid, t) := iterator.new(e); # create a single row BAT of the group id we're processing bn_gid := bat.new(:oid, :oid); bat.insert(bn_gid, grpid, grpid); # join this to the GID->DATA map BAT to get the Group DATA oids b_g2dr := bat.reverse(g); b_gid2data := algebra.leftjoin(bn_gid, b_g2dr); grpval := ""; # Get the BAT of the actual data to be processed # # !!! HOW TO GET THIS TO WORK???????? # b_data := algebra.join(b_gid2data, b); grpval := bin_sample(b_data); # Store the result for this group bat.insert(bn, grpid, grpval); redo (grpid, t) := iterator.next(e); exit (grpid, t); return bn; end subbin_sample; Scott Mathieson, ________________________________

2 3

Failing to load a table
by Franck Routier 05 Jul '13

05 Jul '13

Hi all, I am trying to copy a table from a postgresql database into monetdb, but it fails, with monetdb filling the disks. I am using latest version of Monetdb (Feb2013-SP3) on Ubuntu Linux, with 32 GB Ram and SSD disks. To copy into Monetdb, I have a java program that runs mclient as: mclient --language=sql --databas=database --host=host --user=user /tmp/command.ctl with command.ctl : COPY nnn RECORDS INTO table FROM STDIN USING DELIMITERS'|', '\n', '\' NULL AS '' Then the Postgresql resultset is read from jdbc, formatted and written into STDIN. So far so good, it work with my dimensions tables (small), and with a 295 millions rows fact table that has 109 columns. This fact table takes 86 GB on the disk for Postgresql storage (plus indices). I have another form of the fact table, with less rows (184 millions) but more columns (505). While the first form of the table has a measureId column and only one value column, the other form of the table has one value column per measure. (It is a "sum(case when measureId=X then value else null ... group by" version of the first table). Loading this "columnar" table fails, with monetdb filling the disks (something like 500 GB was left before the loading process) and quitting (not sure what happens exactly here on the server, but I get a broken pipe error in java, and the monetdb database is lost). The table size on disk for Postgresql is 37 GB. Using VectorWise, the table loads, and take only 7 GB on disks, but the loading process (very similar to what we do with MonetDB) takes a very long time( 25 hours), much longer than what happens with the 109 columns tables. I am trying to make sense of that and understand why this table is giving a (very) hard time to monetdb (and somehow VectorWise as well). Does anyone have an explanation ? thanks, Franck

4 12

how are user defined function arguments passed ?
by integrity 05 Jul '13

05 Jul '13

Hi everyone, In http://www.monetdb.org/Documentation/Cookbooks/SQLrecipes/UserDefinedFuncti… In thesection about how to extend SQL with a simple scalar function to reverse a string step 3. Extension starts with a definitin of the MAL signatures. See the example given, or browse through the files in monetdb5/modules/mal/*.mal to get a glimpse on how to write them. The MonetDB kernel documentation provides more details. The file contains the MAL snippet: command reverse(ra1:str):str address UDFreverse comment "Reverse a string"; step 4. The signature says that it expects a command body implementation under the name UDFreverse, shown below. The C-signature is a direct mapping, where arguments are passed by reference and the return value(s) references are the first in the arguments list. The body should return a (malloced) string to denote an exception being raised or MAL_SUCCEED upon access. #include "udf.h" static str reverse(const char *src) { size_t len; str ret, new; /* The scalar function returns the new space */ len = strlen(src); ret = new = GDKmalloc(len + 1); if (new == NULL) return NULL; new[len] = 0; while (len > 0) *new++ = src[--len]; return ret; } str UDFreverse(str *ret, str *src) { if (*src == 0 || strcmp(*src, str_nil) == 0) *ret = GDKstrdup(str_nil); else *ret = reverse(*src); return MAL_SUCCEED; } I don't know why the argument name of "command reverse(ra1:str):str" different from function definition UDFreverse(str *ret, str *src)? how does ret come from? is it according to the naming rules that the first argument is the result, must it be a string type? Thanks! Meng

1 0

Announcement: New Feb2013-SP3 Bugfix release of MonetDB suite
by Hannes Muehleisen 02 Jul '13

02 Jul '13

The MonetDB team at CWI/MonetDB BV is pleased to announce the Feb2013-SP3 bugfix release of the MonetDB suite of programs. More information about MonetDB can be found on our website at <http://www.monetdb.org/>. For details on this release, please see the release notes at <http://www.monetdb.org/Downloads/ReleaseNotes>. As usual, the download location is <http://dev.monetdb.org/downloads/>.

2 1