[gdal-dev] Sharing GDAL Dataset objects between threads (JNI)

[Even Rouault]

Selon Markus Schneider <schneider at occamlabs.de>:

> Hi Even,
>
> thanks for you help so far.
>
> I now observed that handing around a (ECW) Dataset between threads works
> fine, as long as I open it using:
>
> - gdal.Open(String name)
>
> However, when I open it using:
>
> - gdal.OpenShared(String name)
>
> I see strange effects and VM crashes.
>
> Does this make sense?

Yes, GDALOpenShared() ( see the doc of the C function ) must be used with great
caution in a multi-threaded scenario. As long as the thread that has called
GDALOpenShared() continues to use it until it closes it, it is OK. But if that
thread handles the dataset to another one, then chaos might happen.

For example in the following scenario :

         Thread 1                                  Thread 2
ds1 = OpenShared("foo")
                           -- handle it to -->
ds2 = OpenShared("foo") ( == ds1)
  do_things(ds2)                                do_things(ds1)
ds2.Close()
                                                ds1.Close()

At a point, you can call methods on the same dataset object from different
threads...

By the way, the issue you've observed with ECW would also be true with other
drivers. It is just by chance that you didn't notice it.

> What do you believe is preferrable with regard to
> performance and memory consumption:
>
> 1. Using a ThreadPool and obtain a Dataset using gdal.OpenShared for
> each thread
> 2. Using gdal.Open and simply pool the obtained Dataset objects (and
> hand them around between threads)

I'm not sure to completely understand what you've in mind, but I would rather
recommand option 2.

Best regards,

Even

-- 
Geospatial professional services
http://even.rouault.free.fr/services.html