Invalid object data error for generated images

[kyrieb-ekat]

Somewhat similar to what we've seen before with #1196 where we were getting 500 errors, which turned out to be an image size thing, we're now getting "Invalid object data" errors in separated layers.

Occasionally the layers will just show up blank or empty, where text isn't being separated at all or is instead being put in the background. Most recently we're only getting these object data errors- previously only text layers weren't showing up, but now we're getting layer 2 (stafflines) not separating/throwing this error as well.

These images, when downloaded, are still viewable, and when used as inputs in pixel for correction do behave appropriately you just can't 'view' them in Rodan.

Images below of the error when you click 'view' in Rodan, and of the image when downloaded.

[kyrieb-ekat]

Jumping back on this and bumping priority: it's back and it's possibly worse. In trying to run pixel on production, even waiting days, we never get past the 500 error to open pixel from the "open" button when trying to open pixel from the workflow run tab. I have tried a hard reload, etc, and on multiple browsers (Firefox-Chrome-Safari). I would be very interested if @JoyfulGen you were able to replicate this.

Previously shrinking images has worked- I'm going to be taking that as a next step this week by sending a single stave through and seeing if it opens. Otherwise this is officially higher priorty- "mid/large" images can't be opened in pixel (~4000x6000 is the rough size of the images of the resized Ms73 images I fed through earlier).

[kyrieb-ekat]

Size test- this image worked (a quartered section of a page from ms73), 2700x870. Was able to open and load in pixel

[kyrieb-ekat]

I officially am moderately bewildered. I've run several different full size folio tests, and they've all succeeded. This may be a thing where, when it fails, the solution is to rerun the job and see if it works? Or if there's a lot of activity on Rodan it might have a hard time? We might need to default back to a use-schedule if that seems to be the issue, but for now I'm really not certain what's going on.

Perhaps semi-related, but the full size image did load in with its tile lines intermittently visible, which could be showing the memory issue. (see picture; you might need to open it in another tab to see the lines, they're quite fine/thin).

[kyrieb-ekat]

I have attached a video demonstrating the issue; the old pixel job which failed still fails and can't retrieve/load pixel. The new pixel which produced the above screenshot loads in. The images are the same size.

Screen.Recording.2024-12-11.at.11.42.53.AM.mov

[fujinaga]

Did you ever try the Resize job? Try reducing the image by half, to see what happens.

[kyrieb-ekat]

Did a resize job version and it opened alright; I reckon a workflow with resize upstream of pixel should be kept in reserve if 500 errors start popping up again, just to see if it makes a difference. At the moment I can't get the 500 error, but that could just be because I'm the only one on Rodan, or it's in a good mood.