Voyage: GC collecting random objects?

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
10 messages Options
Reply | Threaded
Open this post in threaded view
|

Voyage: GC collecting random objects?

Bernat Romagosa
Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.
Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

EstebanLM
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.

Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

Bernat Romagosa
It says "Reference not found", and the method signaling the error is VOMongoMaterializer >> #missingContentFor:id:

I can try to set up a minimal test case and sent it to you, would that help?

Thanks!


2013/2/8 Esteban Lorenzano <[hidden email]>
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.




--
Bernat Romagosa.
Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

Ciprian Teodorov-3
In reply to this post by Bernat Romagosa

Hi,

I'm not particulary knowledgeable of inner workings of Iliad or voyage. But I think you should have a look at ghostproxy http://rmod.lille.inria.fr/web/pier/software/Marea/GhostProxies.

The idea being that after storing your objects you should keep proxies to them in the image. These proxies would replace your objects and have knowledge of how to retrieve them from voyage. Something like that is done by the marea project. http://rmod.lille.inria.fr/web/pier/software/Marea

Cheers

On Feb 8, 2013 11:48 AM, "Bernat Romagosa" <[hidden email]> wrote:
Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.
Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

EstebanLM
In reply to this post by Bernat Romagosa
yes, this is what happens: 

You have a graph that can look like this: 

A->b->c->D

where A and D represent roots of voyage (#isVoyageRoot = true)

you removed D, but A is not updated (the logic for auto update references is too complex, and this is a known limitation of voyage). 
So, in a next cycle you try to bring A to the image, and A is trying to load eagerly an instance of D... which does not exists anymore, so you have the error. 
To prevent this kind of problems, I use something that I call "eventual integrity" :) you can configure your collection to not fail when you try to get it and it does not exists. For achieve that, you have to add a setting to the class you want to take as "lazy deletable" (is a very dangerous property to just allow it by defect).
You have to do something like this:

D class>>#mongoContainer
<mongoContainer>
^VOMongoContainer new
enableMissingContent;
yourself

note that who is enabling missing content is D, not A :)

hope it helps.

Cheers,
Esteban

ps: y si no se entiende nada, mandame un mail privado y lo explico en español :)

On Feb 8, 2013, at 12:12 PM, Bernat Romagosa <[hidden email]> wrote:

It says "Reference not found", and the method signaling the error is VOMongoMaterializer >> #missingContentFor:id:

I can try to set up a minimal test case and sent it to you, would that help?

Thanks!


2013/2/8 Esteban Lorenzano <[hidden email]>
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.




--
Bernat Romagosa.

Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

Bernat Romagosa
Hmm I'm not sure I understand. Now I've tried not even having the objects stored in a classvar at all, but saving them from a materialized fuel file straight away:

(FLMaterializer materializeFromFileNamed: 'mongodb.obj') do: [ :each | each save ].

It's all ok and working, until I do Smalltalk garbageCollect and there goes the "Reference not found" error again, without me having removed any object.

Would making ALL of my classes respond true to #isVoyageRoot fix the issue?

Thanks again!

Bernat.

p.s. se entiende el idioma, no el contenido, pero eso es culpa mía! jeje


2013/2/8 Esteban Lorenzano <[hidden email]>
yes, this is what happens: 

You have a graph that can look like this: 

A->b->c->D

where A and D represent roots of voyage (#isVoyageRoot = true)

you removed D, but A is not updated (the logic for auto update references is too complex, and this is a known limitation of voyage). 
So, in a next cycle you try to bring A to the image, and A is trying to load eagerly an instance of D... which does not exists anymore, so you have the error. 
To prevent this kind of problems, I use something that I call "eventual integrity" :) you can configure your collection to not fail when you try to get it and it does not exists. For achieve that, you have to add a setting to the class you want to take as "lazy deletable" (is a very dangerous property to just allow it by defect).
You have to do something like this:

D class>>#mongoContainer
<mongoContainer>
^VOMongoContainer new
enableMissingContent;
yourself

note that who is enabling missing content is D, not A :)

hope it helps.

Cheers,
Esteban

ps: y si no se entiende nada, mandame un mail privado y lo explico en español :)

On Feb 8, 2013, at 12:12 PM, Bernat Romagosa <[hidden email]> wrote:

It says "Reference not found", and the method signaling the error is VOMongoMaterializer >> #missingContentFor:id:

I can try to set up a minimal test case and sent it to you, would that help?

Thanks!


2013/2/8 Esteban Lorenzano <[hidden email]>
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.




--
Bernat Romagosa.




--
Bernat Romagosa.
Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

Bernat Romagosa
(I've tried making all objects be voyageRoot and still no luck)


2013/2/8 Bernat Romagosa <[hidden email]>
Hmm I'm not sure I understand. Now I've tried not even having the objects stored in a classvar at all, but saving them from a materialized fuel file straight away:

(FLMaterializer materializeFromFileNamed: 'mongodb.obj') do: [ :each | each save ].

It's all ok and working, until I do Smalltalk garbageCollect and there goes the "Reference not found" error again, without me having removed any object.

Would making ALL of my classes respond true to #isVoyageRoot fix the issue?

Thanks again!

Bernat.

p.s. se entiende el idioma, no el contenido, pero eso es culpa mía! jeje


2013/2/8 Esteban Lorenzano <[hidden email]>
yes, this is what happens: 

You have a graph that can look like this: 

A->b->c->D

where A and D represent roots of voyage (#isVoyageRoot = true)

you removed D, but A is not updated (the logic for auto update references is too complex, and this is a known limitation of voyage). 
So, in a next cycle you try to bring A to the image, and A is trying to load eagerly an instance of D... which does not exists anymore, so you have the error. 
To prevent this kind of problems, I use something that I call "eventual integrity" :) you can configure your collection to not fail when you try to get it and it does not exists. For achieve that, you have to add a setting to the class you want to take as "lazy deletable" (is a very dangerous property to just allow it by defect).
You have to do something like this:

D class>>#mongoContainer
<mongoContainer>
^VOMongoContainer new
enableMissingContent;
yourself

note that who is enabling missing content is D, not A :)

hope it helps.

Cheers,
Esteban

ps: y si no se entiende nada, mandame un mail privado y lo explico en español :)

On Feb 8, 2013, at 12:12 PM, Bernat Romagosa <[hidden email]> wrote:

It says "Reference not found", and the method signaling the error is VOMongoMaterializer >> #missingContentFor:id:

I can try to set up a minimal test case and sent it to you, would that help?

Thanks!


2013/2/8 Esteban Lorenzano <[hidden email]>
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.




--
Bernat Romagosa.




--
Bernat Romagosa.



--
Bernat Romagosa.
Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

EstebanLM
In reply to this post by Bernat Romagosa
no, isVoyageRoot mark the objects of the graph that you want as roots. All the others are just embedded objects. 
Just try adding enableMissingContent to class "D" (I don't know what is the real name in your case :)

Esteban
 

On Feb 8, 2013, at 12:39 PM, Bernat Romagosa <[hidden email]> wrote:

Hmm I'm not sure I understand. Now I've tried not even having the objects stored in a classvar at all, but saving them from a materialized fuel file straight away:

(FLMaterializer materializeFromFileNamed: 'mongodb.obj') do: [ :each | each save ].

It's all ok and working, until I do Smalltalk garbageCollect and there goes the "Reference not found" error again, without me having removed any object.

Would making ALL of my classes respond true to #isVoyageRoot fix the issue?

Thanks again!

Bernat.

p.s. se entiende el idioma, no el contenido, pero eso es culpa mía! jeje


2013/2/8 Esteban Lorenzano <[hidden email]>
yes, this is what happens: 

You have a graph that can look like this: 

A->b->c->D

where A and D represent roots of voyage (#isVoyageRoot = true)

you removed D, but A is not updated (the logic for auto update references is too complex, and this is a known limitation of voyage). 
So, in a next cycle you try to bring A to the image, and A is trying to load eagerly an instance of D... which does not exists anymore, so you have the error. 
To prevent this kind of problems, I use something that I call "eventual integrity" :) you can configure your collection to not fail when you try to get it and it does not exists. For achieve that, you have to add a setting to the class you want to take as "lazy deletable" (is a very dangerous property to just allow it by defect).
You have to do something like this:

D class>>#mongoContainer
<mongoContainer>
^VOMongoContainer new
enableMissingContent;
yourself

note that who is enabling missing content is D, not A :)

hope it helps.

Cheers,
Esteban

ps: y si no se entiende nada, mandame un mail privado y lo explico en español :)

On Feb 8, 2013, at 12:12 PM, Bernat Romagosa <[hidden email]> wrote:

It says "Reference not found", and the method signaling the error is VOMongoMaterializer >> #missingContentFor:id:

I can try to set up a minimal test case and sent it to you, would that help?

Thanks!


2013/2/8 Esteban Lorenzano <[hidden email]>
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.




--
Bernat Romagosa.




--
Bernat Romagosa.

Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

Bernat Romagosa
OK! It looks like it's working now! Thank you so much! :)

Bernat.

p.s. After garbage collection, #selectAll is really slow. I guess this is normal as all objects have to be retrieved all over from the DB, is this right?


2013/2/8 Esteban Lorenzano <[hidden email]>
no, isVoyageRoot mark the objects of the graph that you want as roots. All the others are just embedded objects. 
Just try adding enableMissingContent to class "D" (I don't know what is the real name in your case :)

Esteban

 

On Feb 8, 2013, at 12:39 PM, Bernat Romagosa <[hidden email]> wrote:

Hmm I'm not sure I understand. Now I've tried not even having the objects stored in a classvar at all, but saving them from a materialized fuel file straight away:

(FLMaterializer materializeFromFileNamed: 'mongodb.obj') do: [ :each | each save ].

It's all ok and working, until I do Smalltalk garbageCollect and there goes the "Reference not found" error again, without me having removed any object.

Would making ALL of my classes respond true to #isVoyageRoot fix the issue?

Thanks again!

Bernat.

p.s. se entiende el idioma, no el contenido, pero eso es culpa mía! jeje


2013/2/8 Esteban Lorenzano <[hidden email]>
yes, this is what happens: 

You have a graph that can look like this: 

A->b->c->D

where A and D represent roots of voyage (#isVoyageRoot = true)

you removed D, but A is not updated (the logic for auto update references is too complex, and this is a known limitation of voyage). 
So, in a next cycle you try to bring A to the image, and A is trying to load eagerly an instance of D... which does not exists anymore, so you have the error. 
To prevent this kind of problems, I use something that I call "eventual integrity" :) you can configure your collection to not fail when you try to get it and it does not exists. For achieve that, you have to add a setting to the class you want to take as "lazy deletable" (is a very dangerous property to just allow it by defect).
You have to do something like this:

D class>>#mongoContainer
<mongoContainer>
^VOMongoContainer new
enableMissingContent;
yourself

note that who is enabling missing content is D, not A :)

hope it helps.

Cheers,
Esteban

ps: y si no se entiende nada, mandame un mail privado y lo explico en español :)

On Feb 8, 2013, at 12:12 PM, Bernat Romagosa <[hidden email]> wrote:

It says "Reference not found", and the method signaling the error is VOMongoMaterializer >> #missingContentFor:id:

I can try to set up a minimal test case and sent it to you, would that help?

Thanks!


2013/2/8 Esteban Lorenzano <[hidden email]>
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.




--
Bernat Romagosa.




--
Bernat Romagosa.




--
Bernat Romagosa.
Reply | Threaded
Open this post in threaded view
|

Re: Voyage: GC collecting random objects?

EstebanLM
it depends on the amount of objects, but yes... you can play with lazy/eager ways of retrieve subcollections but if you have 10k objects, it will take some time to bring them all :)

Esteban


On Feb 8, 2013, at 12:59 PM, Bernat Romagosa <[hidden email]> wrote:

OK! It looks like it's working now! Thank you so much! :)

Bernat.

p.s. After garbage collection, #selectAll is really slow. I guess this is normal as all objects have to be retrieved all over from the DB, is this right?


2013/2/8 Esteban Lorenzano <[hidden email]>
no, isVoyageRoot mark the objects of the graph that you want as roots. All the others are just embedded objects. 
Just try adding enableMissingContent to class "D" (I don't know what is the real name in your case :)

Esteban

 

On Feb 8, 2013, at 12:39 PM, Bernat Romagosa <[hidden email]> wrote:

Hmm I'm not sure I understand. Now I've tried not even having the objects stored in a classvar at all, but saving them from a materialized fuel file straight away:

(FLMaterializer materializeFromFileNamed: 'mongodb.obj') do: [ :each | each save ].

It's all ok and working, until I do Smalltalk garbageCollect and there goes the "Reference not found" error again, without me having removed any object.

Would making ALL of my classes respond true to #isVoyageRoot fix the issue?

Thanks again!

Bernat.

p.s. se entiende el idioma, no el contenido, pero eso es culpa mía! jeje


2013/2/8 Esteban Lorenzano <[hidden email]>
yes, this is what happens: 

You have a graph that can look like this: 

A->b->c->D

where A and D represent roots of voyage (#isVoyageRoot = true)

you removed D, but A is not updated (the logic for auto update references is too complex, and this is a known limitation of voyage). 
So, in a next cycle you try to bring A to the image, and A is trying to load eagerly an instance of D... which does not exists anymore, so you have the error. 
To prevent this kind of problems, I use something that I call "eventual integrity" :) you can configure your collection to not fail when you try to get it and it does not exists. For achieve that, you have to add a setting to the class you want to take as "lazy deletable" (is a very dangerous property to just allow it by defect).
You have to do something like this:

D class>>#mongoContainer
<mongoContainer>
^VOMongoContainer new
enableMissingContent;
yourself

note that who is enabling missing content is D, not A :)

hope it helps.

Cheers,
Esteban

ps: y si no se entiende nada, mandame un mail privado y lo explico en español :)

On Feb 8, 2013, at 12:12 PM, Bernat Romagosa <[hidden email]> wrote:

It says "Reference not found", and the method signaling the error is VOMongoMaterializer >> #missingContentFor:id:

I can try to set up a minimal test case and sent it to you, would that help?

Thanks!


2013/2/8 Esteban Lorenzano <[hidden email]>
Hi, 

Voyage should keep a weak cache... objects should be disposed as soon as they are not used anymore. 
Now... it is saying "Reference not found", or "Lazy reference not found"?

Esteban

On Feb 8, 2013, at 11:46 AM, Bernat Romagosa <[hidden email]> wrote:

Hi!

We've been trying to migrate the data of our Iliad webapp to Voyage, and so far everything was going great, until the garbage collector collected objects it shouldn't have... Here's what we did:

We used to have all objects stored in an Instances classVar belonging to their own classes, so:

  1. I had our classes return true to #isVoyageRoot
  2. ourClasses do: [ :c | (c classVarNamed: #Instances) do: [ :i | i save ]]
  3. I checked MongoDB and all was looking great. From Pharo, we could get all our instances and explore all of their relationships by OurClass selectAll. Everything was working, so we thought we didn't need the image stored instances anymore, thus:
  4. ourClasses do: [ :c | c classVarNamed: #Instances put: nil ]
  5. Smalltalk garbageCollect
  6. RMUser selectAll → (long delay) → BOOM → 'Reference  not found RMUser: OID(3725800612)'
So, should we actually keep references to our instances in the image? I doubt this is the case, cos that'd mean we still have the 2Gb storage limitation, and then having an external DB would not be helping much...

Thanks a lot in advance!

Bernat.

p.s. I've also tried saving copies of my objects, but the result's been the same, after garbage-collecting, I lose references.

--
Bernat Romagosa.




--
Bernat Romagosa.




--
Bernat Romagosa.




--
Bernat Romagosa.