Memory consumption in OrientDBRecordDecoder

[fsicre]

OrientDB allows very big documents (2GB).

Having them twice in memory may cause major problems

Do you think that, once parsed, $this->content could be empty-ed in order to reduce memory consumption ?

[AntonTerekhov]

Well, this is least I'd to worry about.

It's clear, that any document big enough to consume all available memory will raise two problems before crash: slow reading from net and slow parsing.

On the other hand, if $this->content isn't emptied, we can introduce flag and do not encode, if data hans't changed.

[fsicre]

I do agree with you on the first part of your comment.

I'm not sure to understand the second part.

For now what I have understood is that

• the first attempt to a $record->data->XXXX triggers the parsing of the serialized ->content
• the ->data is populated
• the ->content is kept asis. At this stage the record contains twice its informations.

When you say "we can introduce flag and do not encode", are you optimizing a futur update of the record without any prior changes of its data, and in this case, sending back the ->content without encode ?

$record->data->addresses[0]->city = "New city name"; But how can we know at $record level that le city name has changed ?

Kind of Composite pattern ?

$record->data->addresses got to be a special Array with tainting capabilities $record->data->addresses[0] got to be a special Object with tainting capabilities

and on sending to server $record got to ask its "childs" if they are tainted ?

[AntonTerekhov]

Yes, something like that. May be Observer.

Not sure if this kind of optimization would be processor-friendly, or even case "we're downloaded record, it got zero changes and we occasionally put it back unmodified" is occurs frequently enough.

[fsicre]

Ok I see, thanks or your explainations.