How can I do a scrolled search on MetaCPAN?

Question

I'm trying to convert this script to use the new Elasticsearch official client instead of the older (now deprecated) ElasticSearch.pm, but I can't get the scrolled search to work. Here's what I've got:

#! /usr/bin/perl

use strict;
use warnings;
use 5.010;

use Elasticsearch ();
use Elasticsearch::Scroll ();

my $es = Elasticsearch->new(
  nodes => 'http://api.metacpan.org:80',
  cxn   => 'NetCurl',
  cxn_pool => 'Static::NoPing',
  #log_to   => 'Stderr',
  #trace_to => 'Stderr',
);

say 'Getting all results at once works:';
my $results = $es->search(
  index => 'v0',
  type  => 'release',
  body  => {
    filter => { range => { date => { gte => '2013-11-28T00:00:00.000Z' } } },
    fields => [qw(author archive date)],
  },
);

foreach my $hit (@{ $results->{hits}{hits} }) {
  my $field = $hit->{fields};
  say "@$field{qw(date author archive)}";
}

say "\nUsing a scrolled search does not work:";
my $scroller = Elasticsearch::Scroll->new(
  es          => $es,
  index       => 'v0',
  search_type => 'scan',
  size        => 100,
  type        => 'release',
  body => {
    filter => { range => { date => { gte => '2013-11-28T00:00:00.000Z' } } },
    fields => [qw(author archive date)],
  },
);

while (my $hit = $scroller->next) {
  my $field = $hit->{fields};
  say "@$field{qw(date author archive)}";
} # end while $hit

The first search, where I'm just getting all the results in 1 chunk, works fine. But the second search, where I'm trying to scroll through the results, produces:

Using a scrolled search does not work:
[Request] ** [http://api.metacpan.org:80]-[500]
ActionRequestValidationException[Validation Failed: 1: scrollId is missing;],
called from sub Elasticsearch::Transport::try {...}
at .../Try/Tiny.pm line 83. With vars: {'body' =>
'ActionRequestValidationException[Validation Failed: 1: scrollId is missing;]',
'request' => {'path' => '/_search/scroll','serialize' => 'std',
'body' => 'c2Nhbjs1OzE3MjU0NjM2MjowakFELUU3VFFibTJIZW1ibUo0SUdROzE3MjU0NjM2NDowakFELUU3VFFibTJIZW1ibUo0SUdROzE3MjU0NjM2MTowakFELUU3VFFibTJIZW1ibUo0SUdROzE3MjU0NjM2MDowakFELUU3VFFibTJIZW1ibUo0SUdROzE3MjU0NjM2MzowakFELUU3VFFibTJIZW1ibUo0SUdROzE7dG90YWxfaGl0czoxNDQ7',
'method' => 'GET','qs' => {'scroll' => '1m'},'ignore' => [],
'mime_type' => 'application/json'},'status_code' => 500}

What am I doing wrong? I'm using Elasticsearch 0.75 and Elasticsearch-Cxn-NetCurl 0.02, and Perl 5.18.1.

Answer 1

I finally got it working with the newer Search::Elasticsearch official client. Here's the short version:

#! /usr/bin/perl

use strict;
use warnings;
use 5.010;

use Search::Elasticsearch ();

my $es = Search::Elasticsearch->new(
  cxn_pool => 'Static::NoPing',
  nodes    => 'api.metacpan.org:80',
);

my $scroller = $es->scroll_helper(
  index       => 'v0',
  type        => 'release',
  search_type => 'scan',
  scroll      => '2m',
  size        => 100,
  body        => {
    fields => [qw(author archive date)],
    query  => { range => { date => { gte => '2015-02-01T00:00:00.000Z' } } },
  },
);

while (my $hit = $scroller->next) {
  my $field = $hit->{fields};
  say "@$field{qw(date author archive)}";
} # end while $hit

Note that the records are not sorted when you do a scrolled search. I wound up dumping the records into a temporary database and sorting them locally. The updated script is on GitHub.

Answer 2

I don't have a direct answer, but I might have an approach to trouble shooting:

I followed your link to the Elasticsearch::Client and found a scroll() method:

https://metacpan.org/pod/Elasticsearch::Client::Direct#scroll

This method takes scroll and scroll_id as parameters. scroll is the number of minutes that you can keep calling the scroll method before the search expires. scroll_id is a marker to the place where the last call to scroll() ended.

$results = $e->scroll(
    scroll      => '1m',
    scroll_id   => $id
);

Elasticsearch::Scroll is an object oriented wrapper around scroll() which hides scroll and scroll_id.

I would run perl -d on your script, and step in to $scroller->next and follow that as far down the rabbit hole as you can. Something in there is trying a search which should be populating scroll_id or scrollId and is failing.

My description here is admittedly pretty rough... I ran across an accurate description of what the scroll id is and does during my googling, but I can't seem to find it again.