Coffee|Code : Dan Scott

Doing interesting things with Evergreen serials data

I'm working on a project to compare our electronic journal holdings with our print journal holdings. This is probably a task that most academic libraries have been working on over the past few years, as collection space dwindles, the duplication of holdings in electronic and print formats increases, and electronic delivery and 24/7 access becomes the default expectation of our patrons.

In my previous post, I worked through the hoops required to get our SFX holdings into a usable database for query purposes. In this post, I'll walk through the steps required to get the serials holdings from Evergreen into the same database so that we can generate reports based on the authoritative sources for both our electronic and print holdings.

We'll start by dumping the schema for the biblio.record_entry and serial.record_entry tables from our Evergreen database. In the previous post, we could have added the tables from the SFX export to the Evergreen database, but I don't like mixing these more experimental projects with our production system - so we'll work with the a database named periodicals instead.

pg_dump --no-owner --data-only --table biblio.record_entry \
    --table serial.record_entry --database evergreen \
    > bre_sre_schema.ddl

We have to munge the schema to not create the indexes on the tables - should lead to faster loads. Also, remove any triggers that point at other stuff that doesn't exist in this limited subset of data. Then create the schema in our periodicals holdings database:

psql -f bre_sre_schema.ddl -d periodical

Now dump the data for those tables from the Evergreen database. If you have a large set of bibliographic records like we do, make sure you have a few gigabytes of space available in the output location.

pg_dump --no-owner --schema-only --table biblio.record_entry \
    --table serial.record_entry --database evergreen > bre_sre_data.dump

Okay, now you can load the data into your serials holdings database:

psql -f bre_sre_data.dump -d periodical

And now we add the indexes that we previously culled from the schema. You can be more selective in the indexes you create, if you know what you're doing.

For some reason, I opted to play with PostgreSQL's support for XML as a native column type and converted the plain text marc column into an XML column:

ALTER TABLE biblio.record_entry 
    ALTER COLUMN marc 
    SET DATA TYPE XML USING marc::XML;

Now we add the Evergreen holdings to the holdings.conifer table. We use the xpath() function to retrieve the desired values from the MARC XML in biblio.record_entry, and wrap the results in the unnest() function to return the nodeset as a plain text string, rather than an array of values. The WHERE clause restricts the holdings to those owned by the library in which I am interested.

CREATE TABLE holdings.conifer (
    record BIGINT, 
    issn TEXT, 
    coverage TEXT, 
    call_number TEXT
);

INSERT INTO holdings.conifer (record, issn)
    SELECT bre.id, UNNEST(XPATH('//*[local-name()="datafield"][@tag="022"]' ||
            '/*[local-name()="subfield"][@code="a"]/text()', bre.marc))
        FROM biblio.record_entry bre INNER JOIN serial.record_entry sre
        ON sre.record = bre.id
        WHERE sre.owning_lib = 103
;

We'll populate the call number based on the 852 field in the serial record. We could pull this from the asset.call_number table, but this will be good enough for the first pass.

UPDATE holdings.conifer 
    SET call_number = UNNEST(
      XPATH(
        '//*[local-name()="datafield"][@tag="852"]/' ||
            '*[local-name()="subfield"][@code="h"]/text()', 
        (
            SELECT sre.marc::xml
            FROM serial.record_entry sre
              INNER JOIN holdings.conifer hc ON hc.record = sre.record
            WHERE hc.record = holdings.conifer.record
            LIMIT 1
        )
      )
    )
;

Now we need to generate usable holdings statements for the print. Evergreen includes a great MFHD parsing library written in Perl, and PostgreSQL thankfully enables you to create functions written in Perl, but to get the following to work on a non-Evergreen machine, I had to copy Open-ILS/src/perlmods/OpenILS/Utils/MFHD/* to /usr/local/share/perl/5.10.0 and edit the occurrences of OpenILS::Utils::MFHD::* to *.

CREATE OR REPLACE FUNCTION holdings.parse_mfhd ( xml TEXT ) RETURNS TEXT AS $_$

    use MARC::Record;
    use MARC::File::XML;
    use MFHD;

    my $xml = shift;
    my $text;
    my $captions;

    my $marc = MARC::Record->new_from_xml( $xml );
    my $mfhd = MFHD->new($marc);

    foreach my $field ($marc->field('866')) {

        my $holdings = $field->subfield('a');
        if ($holdings) {
            my $public_note = $field->subfield('z');
            if ($public_note) {
                $text .= "$holdings - $public_note";
            } else {
                $text .= "$holdings";
            }
        }
    }

    foreach my $cap_id ($mfhd->captions('853')) {
        my @curr_holdings = $mfhd->holdings('863', $cap_id);
        next unless scalar @curr_holdings;
        foreach (@curr_holdings) {
            if ($captions) {
                $captions .= ', ';
            }
            $captions .= $_->format();
        }
    }

    if ($text and $captions) {
        $text = "$text / $captions";
    } else {
        $text = "$text$captions";
    }

    return $text;

$_$ LANGUAGE PLPERLU;

And update the table:

UPDATE holdings.conifer SET coverage = (
    SELECT holdings.parse_mfhd(marc)
    FROM serial.record_entry
    WHERE serial.record_entry.record = holdings.conifer.record
    LIMIT 1
);

That almost works, but it only retrieves the coverage from a single serial holdings record for a given bibliographic record, even though there might be multiple serial holdings records. To amend that, we'll create a PL/pgSQL function that concatenates all of the coverage statements from all of the pertinent serial holdings records for a given bibliographic record:

CREATE OR REPLACE FUNCTION holdings.print_coverage(marc_record BIGINT)
    RETURNS TEXT AS $$
    DECLARE 
        r RECORD;
        coverage TEXT;
    BEGIN
        -- If coverage is NULL to begin with, then concatenating to it results in NULL
        coverage := '';

        -- RAISE NOTICE 'marc_record = %', marc_record;

        -- Loop over the serial records attached to the targeted bib record
        FOR r IN SELECT marc FROM serial.record_entry 
            WHERE record = marc_record
            ORDER BY id
        LOOP
            coverage := coverage || holdings.parse_mfhd(r.marc);
            -- RAISE NOTICE 'r.marc = %', r.marc;
        END LOOP;

        -- RAISE NOTICE 'coverage = %', coverage;
        RETURN coverage;
    END
$$ LANGUAGE 'plpgsql';

And we'll use this fancy new function to update the print holdings statements again with the more complete coverage:

UPDATE holdings.conifer SET coverage = (
    SELECT holdings.print_coverage(record)
        FROM serial.record_entry
        WHERE serial.record_entry.record = holdings.conifer.record
        LIMIT 1
    )
;

Now the payoff: generating a list of matching ISSNs from the electronic holdings and our print holdings, with the coverage statements for each, for a subset of the SFX collections to which we have access:

-- Set the display to expanded format for easy reading
\x
-- Basic report for perusal
SELECT hsfx.issn AS "ISSN", hsfx.title AS "Title",
        hsfx.collection AS "SFX Collection",
        hsfx.coverage AS "Electronic Coverage",
        hc.coverage AS "Print Coverage", hc.call_number AS "Call Number"
    FROM holdings.sfx_complete hsfx
        INNER JOIN holdings.conifer hc ON hsfx.issn = hc.issn
    WHERE (hsfx.collection ILIKE '%JStor%' OR hsfx.collection LIKE '%Scholars%')
        AND hc.coverage > ''
    LIMIT 5;

That results in:

-[ RECORD 1 ]-------+--------------------------------------------------------------------------
ISSN                | 0142-2774
Title               | Journal of Occupational Behavior
SFX Collection      | JSTOR Arts and Sciences 4
Electronic Coverage | Available from 1980 until 1987. 
Print Coverage      | Vol. 1 No.  - Vol. 8 No. 4 (1980-1987)
Call Number         | DESM-PER
-[ RECORD 2 ]-------+--------------------------------------------------------------------------
ISSN                | 0741-6261
Title               | The Rand Journal of Economics
SFX Collection      | JSTOR Arts and Sciences 2
Electronic Coverage | Available from 1984 until 2006. 
Print Coverage      | V.17 (1986) - v.23 (1992)
Call Number         | DESM-PER
-[ RECORD 3 ]-------+--------------------------------------------------------------------------
ISSN                | 0002-8614
Title               | Journal of the American Geriatrics Society
SFX Collection      | Scholars Portal
Electronic Coverage | Available from 2001 volume: 49 issue: 1 until 2009 volume: 57 issue: 10. 
Print Coverage      | Vol. 1 - 37 (1953-1989)
Call Number         | DESM-PER
-[ RECORD 4 ]-------+--------------------------------------------------------------------------
ISSN                | 0023-7639
Title               | Land Economics
SFX Collection      | JSTOR Arts and Sciences 7
Electronic Coverage | Available from 1948 until 2005. 
Print Coverage      | v.62 (1986) - v.68 (1992)
Call Number         | DESM-PER
-[ RECORD 5 ]-------+--------------------------------------------------------------------------
ISSN                | 0090-2616
Title               | Organizational dynamics
SFX Collection      | Scholars Portal
Electronic Coverage | Available from 1995 volume: 23 issue: 3 until 2009 volume: 38 issue: 3. 
Print Coverage      | Vol. 15 No.  - Vol. 23 No. 5 (Summer 1986-Spring 1995)
Call Number         | DESM-PER

Looks pretty good to these eyes. Okay, now we'll get serious and dump the output to a tab-delimited file so we can easily open it in OpenOffice.org Calc or another spreadsheet:

-- Set delimiter to TAB (CTRL-V )
\f '^V'
-- Set the output to being unaligned
\a
-- Dump the output to a file
\o /tmp/periodicals.tsv
-- Generate URLs for quick catalogue lookups
SELECT 'http://laurentian.concat.ca/opac/en-CA/skin/lul/xml/rdetail.xml?r=' 
            || hc.record || '&l=105&d=1' AS "URL",
        hsfx.issn AS "ISSN", hsfx.title AS "Title",
        hsfx.collection AS "SFX Collection",
        hsfx.coverage AS "Electronic Coverage", hc.coverage AS "Print Coverage",
        hc.call_number AS "Call Number"
    FROM holdings.sfx_complete hsfx
        INNER JOIN holdings.conifer hc ON hsfx.issn = hc.issn
    WHERE (hsfx.collection ILIKE '%JStor%' OR hsfx.collection LIKE '%Scholars%')
        AND hc.coverage > ''
;

And that's it. It might seem complex, but I've found that investing the effort into learning how to lean on PostgreSQL to do the hard work pays plenty of dividends. This exploration should help me contribute more functionality to Evergreen core; for example, I hope to use my experiments with the pl/Perl function to start populating the serial.bib_summary tables using an INSERT/UPDATE/DELETE trigger on serial.record_entry so that we don't have to generate the summaries for every item details request in the catalogue.

There must be other people who have much more intelligent things than me with the TXT dump of SFX holdings that you can generate via the Web administration interface, but as I've gone through this process at least twice and rediscovered it each time, perhaps I'll save myself an hour or two by writing this down. Maybe it will save you a bit of time, too. Or give you an idea of a day in the life of a systems librarian. In part 1, I turn the TXT dump into something that I can load back into a local database. Maybe I want to massage the data, write arbitrary SQL queries, join it against data that I load into other tables in the database... I'll talk about that in a subsequent post. For now, let's get munging!

1. First, I load the raw TXT file into a text editor (gvim, of course) and munge all of the lines so that they are ready to bulk load into my local database. That local database is PostgreSQL, so these commands will be vim-and-PostgreSQL specific, but you should be able to adapt this to whatever your target may be. Note the duplication of the first command is intentional!

   :%s/^I^I/\\N/g
   :%s/^I^I/\\N/g
   :%s/^I$/\\N/
   

2. Then I prepend the DDL required to create the holdings table and copy the data into the newly created holdings table to the top of the TXT file:

   -- I don't want a huge scrolling list of errors if there's something wrong in the data
   SET ON_ERROR_STOP = 1;
   -- I typically get something wrong the first time, so this avoids having to manually
   -- drop and re-rcreate the schema the second and subsequent times
   DROP SCHEMA IF EXISTS holdings CASCADE;
   -- This is PostgreSQL, so we can use schemas. Yay!
   CREATE SCHEMA holdings;
   -- Note that I have no idea what many of these fields are, and frankly don't care
   CREATE TABLE holdings.sfx (title_key TEXT, title TEXT, misc1 TEXT, issn TEXT, 
       ejournal_id TEXT, collection TEXT, coverage TEXT, eissn TEXT, complete_title TEXT, 
       target TEXT, misc2 TEXT, misc3 TEXT, misc4 TEXT, misc5 TEXT, misc6 TEXT, 
       misc7 TEXT, misc8 TEXT, target_parser TEXT, target_id BIGINT, service_id BIGINT, 
       misc_id BIGINT, category TEXT, misc9 TEXT, misc10 TEXT, misc11 TEXT, 
       publisher TEXT, pub_place TEXT, pub_date TEXT, item_type TEXT, active TEXT, 
       misc12 TEXT, misc13 TEXT, misc14 TEXT, language TEXT, title2 TEXT, misc15 TEXT
   );
   -- The COPY command bulk-loads the data from the subsequent lines into the table
   COPY holdings.sfx (title_key, title, misc1, issn, ejournal_id, collection, coverage, eissn, 
       complete_title, target, misc2, misc3, misc4, misc5, misc6, misc7, misc8, target_parser,
        target_id, service_id, misc_id, category, misc9, misc10, misc11, publisher, pub_place, 
       pub_date, item_type, active, misc12, misc13, misc14, language, title2, misc15) 
   FROM STDIN;
   -- All of the TXT data follows
   

3. Assuming that all went well, I can connect to the database and create a few indexes that will probably be useful while doing queries.

   -- Create indexes on the ISSN and E-ISSN
   CREATE INDEX holdings_sfx_issn_idx ON holdings.sfx(issn);
   CREATE INDEX holdings_sfx_eissn_idx ON holdings.sfx(eissn);
   -- Create index on the collection, because we'll limit some queries on this basis
   CREATE INDEX holdings_sfx_collection_idx ON holdings.sfx(collection);
   -- Update the statistics on the table
   ANALYZE;
   

4. Now I can load this into my database, where "datafile.txt" is the munged TXT dump, and "holdings" is the name of the database:

   psql -f datafile.txt -d holdings
   

Now I'm ready to load data from another source - say, from my ILS that includes print holdings - so that I can start matching on ISSN and identify candidates for pruning our print collection based on criteria such as the electronic collection. Perhaps I'll talk about that in part 2...

Update: 2009-11-24 James Forrester of the Ontario Academy of Art and Design has posted a short video (Internet Archive) of the presentation. Thanks, James!

On Friday, October 30th, I presented a status update on Project Conifer at the Free Software Open Source Symposium (FSOSS). This was a follow-up to the talk I gave with John Fink at last year's FSOSS, with the hopefully interesting twist that instead of talking about what we were going to do, I talked about what we had done, and the lessons learned along the way.

This was a slightly modified version of the talk I gave at the Lyrasis/NELINET open source conference earlier in October, aimed at a more general audience. The talk was recorded and will be posted online at the FSOSS site at some point.

Here are the slides in (ODP) and (PDF) format. The speaker notes on the slides will give you the meat of the content.

Dear Dan:

You felt pretty pleased with yourself for jumping on the latest Ubuntu release back in the alpha stages on your virtual machine, then having it running on your laptop and Lynn's Asus EEE 701 the day after the final release came out.

But did you ever have to change the default runlevel? NO, you didn't. You didn't anticipate that you would want to convert that alpha-6 server into a workstation. Nice thinking, hero. And now that you've figured it out, you're going to remind yourself that the place to change it is in /etc/init/rc-sysinit.conf. You might want to look at the env DEFAULT_RUNLEVEL setting. Oh sure, you could manually create an old-school /etc/inittab file, and it would get picked up from there. But, you know, throwing away a grand old tradition like that really feels liberating, doesn't it?

So now you know. Don't forget, okay? But if you do, you might just find yourself coming back here.