From: Rodrigo Rosenfeld Rosas <rr.rosas@...>
Date: 2014-01-15T17:58:44-02:00
Subject: [ruby-core:59791] About unmarshallable DRb objects life-time

This is a multi-part message in MIME format.
--------------000301080304040307060804
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

A while ago I created a proof-of-concept that I intended to use in my 
project for using Java libraries from MRI through the use of JRuby and DRb:

http://rosenfeld.herokuapp.com/en/articles/ruby-rails/2013-07-16-running-java-from-mri-ruby-through-drb

It worked in my little tests and today I finally decided to use it in 
the process of migrating a Grails application to a Rails one. I couldn't 
find any Ruby library to edit an Excel (XLS, not XLSX) template and 
create a new spreadsheet so I decided to keep up with the implementation 
using Apache POI, but I needed to access it in the JVM somehow.

At the same time I didn't want to convert the whole project to JRuby, as 
MRI is much faster for me to develop due to the short start-up time of 
the processes.

When I put the concept to work with real data from my application I got 
tons of random errors in different places. It didn't happen if I 
selected only a few records to export to Excel, but if I decided to 
generate the spreadsheet for all (~ 1k) results, it would always fail in 
random places.

The only explanation I could think of, and after analyzing some data I 
logged from the DRb server, is that that some previously objects created 
in the DRb server (JRuby application) were being garbage collected even 
if I still had a reference for them in the MRI side.

I could somehow "confirm" my suspicions by implementing another 
interface in the DRb server where only regular marshallable Ruby data 
would be transmitted so that I wouldn't take the risk of having my data 
being garbage collected in the server-side. After this change it worked 
all the times I tried but I could no longer use the DRb server as a 
generic platform for accessing the JVM libraries.

So, I'd like to confirm my theory by asking you if this is the intended 
behavior of the DRb protocol.

I mean, suppose I get a Java object from a call to the DRb server. It's 
not marshallable to MRI, of course, so I only get a reference to that 
object in the MRI side. Then, later I want to use that object, but it no 
longer exists in the DRb server as it has no references in that side.

Is DRb supposed to be called only with marshallable data? Or is there a 
way for DRb to know that a reference still exists in a DRb client and 
prevent the object from being GC'ed?

I also read about this:

http://www.ruby-doc.org/stdlib-2.1.0/libdoc/drb/rdoc/DRb/DRbUndumped.html

And if you take a look at the code in my article, you'll find this:

result.extend DRb::DRbUndumped if result.respond_to? :java_class

But even extending the object with DRb::DRbUndumped doesn't seem to 
prevent it from being garbage collected by JRuby. And Charles told me he 
only uses the MRI implementation of DRb, so I'm asking here to try to 
understand if this is how it works by design.

Also, in case this is a bug and DRbUndumped should mark the object as 
not available for being GC'ed, how would it be freed after the DRb 
client no longer needs it?

Sorry if it got confused but I did my best to try to make it as clear as 
possible.

Thanks in advance,
Rodrigo.


--------------000301080304040307060804
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit

<html>
  <head>

    <meta http-equiv="content-type" content="text/html; charset=ISO-8859-1">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    A while ago I created a proof-of-concept that I intended to use in
    my project for using Java libraries from MRI through the use of
    JRuby and DRb:<br>
    <br>
    <meta http-equiv="content-type" content="text/html;
      charset=ISO-8859-1">
    <a
href="http://rosenfeld.herokuapp.com/en/articles/ruby-rails/2013-07-16-running-java-from-mri-ruby-through-drb">http://rosenfeld.herokuapp.com/en/articles/ruby-rails/2013-07-16-running-java-from-mri-ruby-through-drb</a><br>
    <br>
    It worked in my little tests and today I finally decided to use it
    in the process of migrating a Grails application to a Rails one. I
    couldn't find any Ruby library to edit an Excel (XLS, not XLSX)
    template and create a new spreadsheet so I decided to keep up with
    the implementation using Apache POI, but I needed to access it in
    the JVM somehow.<br>
    <br>
    At the same time I didn't want to convert the whole project to
    JRuby, as MRI is much faster for me to develop due to the short
    start-up time of the processes.<br>
    <br>
    When I put the concept to work with real data from my application I
    got tons of random errors in different places. It didn't happen if I
    selected only a few records to export to Excel, but if I decided to
    generate the spreadsheet for all (~ 1k) results, it would always
    fail in random places.<br>
    <br>
    The only explanation I could think of, and after analyzing some data
    I logged from the DRb server, is that that some previously objects
    created in the DRb server (JRuby application) were being garbage
    collected even if I still had a reference for them in the MRI side.<br>
    <br>
    I could somehow "confirm" my suspicions by implementing another
    interface in the DRb server where only regular marshallable Ruby
    data would be transmitted so that I wouldn't take the risk of having
    my data being garbage collected in the server-side. After this
    change it worked all the times I tried but I could no longer use the
    DRb server as a generic platform for accessing the JVM libraries.<br>
    <br>
    So, I'd like to confirm my theory by asking you if this is the
    intended behavior of the DRb protocol.<br>
    <br>
    I mean, suppose I get a Java object from a call to the DRb server.
    It's not marshallable to MRI, of course, so I only get a reference
    to that object in the MRI side. Then, later I want to use that
    object, but it no longer exists in the DRb server as it has no
    references in that side.<br>
    <br>
    Is DRb supposed to be called only with marshallable data? Or is
    there a way for DRb to know that a reference still exists in a DRb
    client and prevent the object from being GC'ed?<br>
    <br>
    I also read about this:<br>
    <br>
    <meta http-equiv="content-type" content="text/html;
      charset=ISO-8859-1">
    <a
href="http://www.ruby-doc.org/stdlib-2.1.0/libdoc/drb/rdoc/DRb/DRbUndumped.html">http://www.ruby-doc.org/stdlib-2.1.0/libdoc/drb/rdoc/DRb/DRbUndumped.html</a><br>
    <br>
    And if you take a look at the code in my article, you'll find this:<br>
    <br>
    <meta http-equiv="content-type" content="text/html;
      charset=ISO-8859-1">
    result.extend DRb::DRbUndumped if result.respond_to? :java_class<br>
    <br>
    But even extending the object with DRb::DRbUndumped doesn't seem to
    prevent it from being garbage collected by JRuby. And Charles told
    me he only uses the MRI implementation of DRb, so I'm asking here to
    try to understand if this is how it works by design.<br>
    <br>
    Also, in case this is a bug and DRbUndumped should mark the object
    as not available for being GC'ed, how would it be freed after the
    DRb client no longer needs it?<br>
    <br>
    Sorry if it got confused but I did my best to try to make it as
    clear as possible.<br>
    <br>
    Thanks in advance,<br>
    Rodrigo.<br>
    <br>
  </body>
</html>

--------------000301080304040307060804--