'Normal' is fine, I'll look into it.

I'm guessing I can mart this as normal as this is a blender studio issue (?)

Alternatively offer JPEG2000 as only option (supports Alpha).

Implemented as the 2nd option (open blend file, create scene, fill VSE, render) in 3371e5d, b3c620b, and a bunch of commits on the Worker

Feel free to poke me on #flamenco on Blender Chat ;-)

@Sybren A. Stüvel (sybren) : still have to make myself familiar with flamenco [I will have to do this for an upcoming studio job...].
beauty of the rule-based thingie was [iirc], that you could add something easily without having to touch existing code.
So in the case of netrender I added something like "Priority-frames" [frames to be rendered first] without too much hassle. (I think the rules could even be mixed per job, but I would need to have a look again, this was ages ago...)

It's now implemented as a sub-quadratic curve (power of 1.6666) with an artist-given maximum number of samples per task. As an example, 400 total samples with a maximum of 100 per task gives these render tasks (first number is "rendering and merging up to this sample count", and second number is samples in only that task):

@Philipp Oeser (lichtwerk) thanks for thinking along with us :)

Output to JPEG would also be good enough (w.r.t. PNG).

not sure if this is still relevant: netrender had something like this based on rules, see https://developer.blender.org/diffusion/BA/browse/master/netrender/balancing.py

I tried again with double-clicking Flamenco Manager, and here it really doesn't crash. I'm archiving this now, because I can't reproduce the issue and thus can't find the root cause. If you have more information that can help, please comment here.

This is in the master branch now.

Implemented & deployed to production.

Also only use mapped network drives (so S:/ instead of rackstation/software), as means something else to Blender. That, and of course T60131 should be fixed.

I recommend not using any exclamation marks in your path. Also only use mapped network drives (so S:/ instead of //rackstation/software), as // means something else to Blender. That, and of course T60131 should be fixed.

Please make sure that you have the latest Blender Cloud add-on 1.11.0 (just released) as it contains Windows and Blender 2.8 fixes. If you still see issues, please provide us with a proper bug report.

I wasn't thinking about encoding those as actual Tasks, but more as an extra payload in the same way we now also handle task cancellation requests. It would be just a list of job+task IDs to upload to the server. I'll have to think about it more to come up with a nice design that both works rapidly enough (when there are no workers asking for tasks, the Manager only infrequently polls that endpoint) and also doesn't load the server too much.

So this would be a parallel queue of "tasks" that are accessible from the "dpsgraph/dispatch" endpoint.
The main difference is that they would be performed directly by the manager?
Could this system be also used for uploading render previews to the server, or should that be part of the job definition?

We can have the Manager upload it to the Server on demand, for example by piggy-backing the 'depsgraph' endpoint the Manager uses to pull any changed tasks from the Server.

In T54099#593120, @Dennis Teusink (Teus) wrote:

@Sybren A. Stüvel (sybren) I previously haven't had a chance to test this script properly. For now we are working with Backburner as our main render farm tool. We would like to use Flamenco though. But it is not very compatible with our Windows network. Everytime we try to work with Flamenco we run into strange issues.

@Sybren A. Stüvel (sybren) I previously haven't had a chance to test this script properly. For now we are working with Backburner as our main render farm tool. We would like to use Flamenco though. But it is not very compatible with our Windows network. Everytime we try to work with Flamenco we run into strange issues.

Implemented in 7680c65daf35712d8a9060c9fe0ccc8fd1716e06 and polished in following commits.

Resolved in e2d9166473184

I tried last night with a render. I started Flamenco Manager using cmd.exe and with the - verbose command.
The manager didn't crash and worked perfectly all night long.

I've tried running Flamenco Manager on Windows 10 for close to 7 hours, and it keeps running just fine. It has one Worker connected to it, so it does get frequent polls for more tasks.

Can you start Flamenco Manager from a cmd.exe terminal? That way it stays open and you can read any error message. Start it with flamenco-manager.exe -verbose to get more verbose logging. If it turns out that that still doesn't give enough information we can use flamenco-manager.exe -debug to enable debug logging, but this will produce a rather copious amount of log messages.

Nice analysis, thanks.

Adding the script to my blender variable in flamenco-manager.yaml works for me.