Testing w/ Xapian and other off-line indexing search engines

rubyonrails-talk
1

I am running into problems testing with Xapian and Sphinx, both off-line index updating search engines (i.e., adding/changing an indexed record puts the re uest into a queue with the actual work done in a cronjob). To test I can load the fixtures, build the index, and run tests successfully that don't add new records or modify the fixtures. But I can figure out how to get the indexes updated between the DB modify and the search. For Xapian I am doing the following from within the test program:

`rake xapian:update_index flush=true RAILS_ENV=test`

The new record doesn't make it into the index, though it does get added to the acts_as_xapian table.

TIA,   Jeffrey

2

Quoting Jeffrey L. Taylor <ror@abluz.dyndns.org>:

I am running into problems testing with Xapian and Sphinx, both off-line index updating search engines (i.e., adding/changing an indexed record puts the re uest into a queue with the actual work done in a cronjob). To test I can load the fixtures, build the index, and run tests successfully that don't add new records or modify the fixtures. But I can figure out how to get the indexes updated between the DB modify and the search. For Xapian I am doing the following from within the test program:

`rake xapian:update_index flush=true RAILS_ENV=test`

The new record doesn't make it into the index, though it does get added to the acts_as_xapian table.

I think I've found the answer. The key thing is that the test script is adding a record at run time. ActiveRecord does not commit these records. This is so it can rollback the test database to just the fixtures after each test. The record ID is being passed to Xapian thru acts_as_xapian database, but there is no such record actually in the database, just in the ActiveRecord cache. This problem exists for all search engines that read the database directly instead of going thru Rails.

For the record,   Jeffrey

3

Quoting Jeffrey L. Taylor <ror@abluz.dyndns.org>:

I am running into problems testing with Xapian and Sphinx, both off- line index updating search engines (i.e., adding/changing an indexed record
puts the re uest into a queue with the actual work done in a cronjob). To test
I can load the fixtures, build the index, and run tests successfully that
don't add new records or modify the fixtures. But I can figure out how to get
the indexes updated between the DB modify and the search. For Xapian I am
doing the following from within the test program:

`rake xapian:update_index flush=true RAILS_ENV=test`

The new record doesn't make it into the index, though it does get
added to the acts_as_xapian table.

I think I've found the answer. The key thing is that the test
script is adding a record at run time. ActiveRecord does not commit these
records. This is so it can rollback the test database to just the fixtures
after each test. The record ID is being passed to Xapian thru acts_as_xapian
database, but there is no such record actually in the database, just in the
ActiveRecord cache. This problem exists for all search engines that read the
database directly instead of going thru Rails.

Although the record isn't committed, other connections to the database
can see it if they ask. With Sphinx I get it to execute "SET SESSION TRANSACTION ISOLATION
LEVEL READ UNCOMMITTED" and it can then see the non committed changes

Fred

4

Quoting Frederick Cheung <frederick.cheung@gmail.com>: [snip]

> I think I've found the answer. The key thing is that the test
> script is > adding a record at run time. ActiveRecord does not commit these
> records. > This is so it can rollback the test database to just the fixtures
> after each > test. The record ID is being passed to Xapian thru acts_as_xapian
> database, > but there is no such record actually in the database, just in the
> ActiveRecord > cache. This problem exists for all search engines that read the
> database > directly instead of going thru Rails.

Although the record isn't committed, other connections to the database
can see it if they ask. With Sphinx I get it to execute "SET SESSION TRANSACTION ISOLATION
LEVEL READ UNCOMMITTED" and it can then see the non committed changes
from my tests.

Thank you. Do you know if this is an SQL standard? From the MySQL manual it looks like it is MySQL specific. For the moment I'll just test the Xapian searching against records in the fixtures. But this is good to know it that turns out to be too great a restriction, i.e., an undetected because of this bug bites me.

Jeffrey

5

I've only used it with mysql. The transaction isolation levels it provides are the 4 ones defined by the ISO sql standard (at least they claim to be). Exactly how you tell the database what isolation level you want might vary, I don't know.

Fred