Making a fossilised repository

From EPrints Documentation
Jump to: navigation, search


(This work requires access to the database & original file-system)

The rough plan is to pull all of the visible pages in the repository and make it available as a static copy - no dynamic stuff, no

This process assumes you have access to the underlying database, and the file-store for the repo.

Assume:

  1. you're working as a non-root user on a workstation
  2. your repo is accessable [to you] at http://my.repo/
  3. your web server will have a document-root of ~/www/my.repo
  4. you know how to set a document-root on your web server
  5. you'll have to do some tidying up (finding images, etc) yourself

Grab a copy of the html pages

  mkdir ~/www
  cd ~/www
  wget --local-encoding=UTF-8 --remote-encoding=UTF-8 --no-cache --mirror -nc -k http://my.repo/

Grab a copy of all the abstract pages

You need check the database - the last eprintid in the eprints table is the number you need to count up to

  cd my.repo
  for id in {1..12345} ; do wget --local-encoding=UTF-8 --remote-encoding=UTF-8 --no-cache -k http://my.repo/$id ; done

[(a) note the lack of --mirror option, abd (b) this *will* take quite a while..... but are you in a hurry?]

This downloads each page into a file called $id, and we need to make that a directory:

---- perl code ----
# run in ~/www/my.repo
use File::Slurp;
my $root = '.';

opendir( my $dh, $root) || die "can't open doc root\n";
my @files = grep { /^\d/ && -f "$root/$_" } readdir($dh);
close $dh;

foreach my $file (@files) {
    my $source = "$root/$file";
    my $content = read_file($source, binmode => ':utf8');
    unlink $source;
    mkdir $source;
    write_file( "$source/index.html", {binmode => ':utf8'}, $content ) ;
}
---- perl code ----

Grab a copy of all the documents

The problem here is that EPrints has a funny directory structure, which doesn't map to URLs, so we grab a copy, and then move them into the right place. (Still in ~/www/my.repo/)

   scp -r eprints_user@my.repo:/path/to/eprints/archives/opendepot/documents/disk0 .

... will copy disk0 and all it's subdirectories into the current directory.

Copy the documents into the right location in the web site

We then use a move script to move each of the documents into the appropriate abstract directory:

---- perl code ----
# you will need to tweek the root & number of nested loops if you have more data.
use File::Copy::Recursive;

my $root = './00/00'; 
my $destination = '.';

opendir( my $dh, $root) || die "can't open doc root\n";
my @tlfs = grep { /^\w/ && -r "$root/$_" } readdir($dh);
close $dh;

foreach my $tlf (@tlfs) {
    my $dir = "$root/$tlf";
    opendir( my $dh, $dir) || die "can't open $dir\n";
    my @blfs = grep { /^\w/ && -r "$dir/$_" } readdir($dh);
    close $dh;
    foreach my $blf (@blfs) {
        my $combined = $tlf . $blf;
        my $final = $combined + 0;
        my $docs = "$dir/$blf";
        my $target = "$destination/$final";
        print "move $docs -> $target\n";
        File::Copy::Recursive::rcopy($docs, $target);
    }
}
---- perl code ----

... This should copy disk0/00/00/02/42/01/something.pdf to 242/01/something.pdf

You can now remove the whole disk0 tree

   rm -rf disk0

Tidy up leading 0's

In the abstract pages, the URLs are 242/1/something.pdf, so we need to delete all the leading 0's:

---- perl code ----
use File::Slurp;

my $root = '.';

opendir( my $dh, $root) || die "can't open doc root\n";
my @tlfs = grep { /^\d/ && -d "$root/$_" } readdir($dh);
close $dh;

foreach my $tlf (@tlfs) {
    my $dir = "$root/$tlf";
    opendir( my $dh, $dir) || die "can't open $dir\n";
    my @blfs = grep { /^\d/ && -r "$dir/$_" } readdir($dh);
    close $dh;
    foreach my $blf (@blfs) {
        my $old = "$dir/$blf";
        my $fn = $blf + 0;
        my $new = "$dir/$fn";
        rename $old, $new;
    }
}
---- perl code ----

Deal with all those Absolute URLs

In a significant number of pages, hrefs are absolute, so need to be made relative. This snippet will fix that:

  find . -type f -exec sed -i 's_http://my.repo/_/_g' {} +

Missing stylesheet & images

These will be under archives/<ARCHIVEID>/html/en/style on your EPrints server - copy the whole style across to your local style directory

Unwanted links to dynamic content

Things like seaching and RSS feeds need to be removed from all web pages. Try:

   find . -type f -exec sed -i '/rel="alternate"\|rel="Search"\|search\/simple\|ep_search_feed/d' {} +

The some of the abstract pages will have the Preview link - that needs to go:

   find . -name '*.html' -exec sed -r -i 's/(\s\|\s)?<a[^>]+>Preview<\/a>//' {} +

The abstract pages will also list the local Persistent URI - which goes to a URL that's not supported - that also needs to go:

find . -name '*.html' -exec perl -p -i -e 's/URI:<\/th> .*?<\/td><\/tr>//' {} +

You may have export options for some of your browse pages - they won't work:

   find view/ -name '*.html' -exec perl -p -i -e '$/=undef;s/<form\b[^>]*>.*?<\/form>//s' {} +

No point in having anything to suggest logins may work - so get rid of all the "Actions" sections:

    find . -name '*.html' -exec perl -p -i -e 's/<h3>Actions \(login required\)<\/h3>.*?<\/table>//' {} +

(that should be most of it...)

Get the web server to serve pages

.... and now you can point a web server's "document root" at /path/to/my.repo and it should all "just work" (you may have file-permissions to deal with - but that's not difficult)