You need to realize that the <p:graphicImage> actually renders a <img src> element with just an URL which is then later individually invoked by the webbrowser when it’s about to parse the obtained HTML markup and present the results.
So, whatever you do in the getter method of <p:graphicImage> it must be designed that way that it can be invoked on a per-request basis. So, the most sane approach would be to create a <p:graphicImage> with a <f:param> wherein the <p:graphicImage value> points an entirely standalone request or application scoped bean (and thus absolutely not view or session scoped), and the <f:param value> points the unique image identifier.
E.g.
<p:graphicImage value="#{images.image}">
<f:param name="id" value="#{someBean.imageId}" />
</p:graphicImage>
Where the Images backing bean can look like this:
@Named
@ApplicationScoped
public class Images {
@EJB
private ImageService service;
public StreamedContent getImage() throws IOException {
FacesContext context = FacesContext.getCurrentInstance();
if (context.getCurrentPhaseId() == PhaseId.RENDER_RESPONSE) {
// So, we're rendering the view. Return a stub StreamedContent so that it will generate right URL.
return new DefaultStreamedContent();
}
else {
// So, browser is requesting the image. Return a real StreamedContent with the image bytes.
String id = context.getExternalContext().getRequestParameterMap().get("id");
Image image = service.find(Long.valueOf(id));
return new DefaultStreamedContent(new ByteArrayInputStream(image.getBytes()));
}
}
}
Or, if you’re already using OmniFaces 2.0 or newer, then consider using its <o:graphicImage> instead which can be used more intuitively, almost exectly the way as you expected. See also the blog on the subject.