Desync in Subnet Proxy subnet on world load after restart

[sibir-ine]

Describe the bug

When first loading a world, existing Subnet Proxy subnets desync with their front network(s). Doing something like disabling & reenabling Subnet Proxy channels is necessary to sync them until the next world load.

To Reproduce

Steps to reproduce the behavior:

1. Create a new world
2. Create a simple Subnet Proxy subnet exposed to a front network with some contents (e.g., a stack of Oak Wood Planks)
3. Restart the game (may require multiple)
4. Load back into the world
5. Try to extract contents from the front network
6. See error

Expected behavior

The networks should be synced on world load.

Logs and crash reports

• latest.log: https://mclo.gs/fT0POLC

Screenshots

N/A

Checklist

• I tested and can reproduce the issue in the latest version of the mod (from Curseforge or GitHub). I guarantee the version mentioned in Environment is the latest one.
• I searched for existing issues that might be related.
• If additional modifications have been made to the mod list, I have provided the exact sources (e.g. github repo for an unofficial build of a mod).

Environment

• C.E.L.L.S version: 0.6.4-beta3
• AE2UEL version: 0.56.5
• Minecraft: 1.12.2
• Forge: 14.23.5.2860
• Java: 8u51
• OS: Windows 11 Version 25H2
• Modpack (if any): N/A

Additional context

N/A

[Aedial]

So, 1 time out of N, the front network will not get the deltas of the back network? Something like race conditions between network initializations?

[sibir-ine]

Yeah, sounds like a race condition. I can't always reproduce it after 1 restart, but I have reliably been able to after 2 so far.

[Aedial]

Beta4 should maybe have some mitigation, but I kinda expect there to be rarer and rarer cases like that. I have not seen it happen again, but it is hard to know due to not having an end-game AE2 network. At least I have not seen the issue with fluid drops missing any more, which means something was fixed in the internal chain.
I added some logging behind the -Dcells.trace.subnetproxy.updateflow=true JVM flag, which precisely traces the timings, but it produces a huge amount of logs (do not let it enabled outside of short debug logging). If providing such logging, always try to mention the exact position(s) of the faulty prox(ies), to allow me filtering the unrelated ones from the log.

[sibir-ine]

I'm still experiencing the desync in my test world, here's the log with the JVM flag:
https://mclo.gs/1DuUFyR
There's only the one Subnet Proxy.

[Aedial]

Do you have a simple setup or a world zip where you can reliably reproduce the setup? I cannot reproduce it in my local world (to validate if the changes fix the issue), as it heavily depends on timings, and I don't want to bother you incessantly with back-and-forth releases.

[sibir-ine]

Yes, here's my simple test world:
New World.zip
I just created a 1.12.2 profile in Prism with only Forge 14.23.5.2860, AE2UEL 0.56.5, & CELLS 0.6.4-beta4 installed, no configs changed. One thing to note is that, sometimes, the desync manifests as the correct amount of items reported to the front network, but it's not extractable (at least through a terminal).

I don't mind testing, I can just download the artifacts from the build action so you don't have to keep making releases.

[Aedial]

Okay, some more fixes with beta5. I have not seen any issues in dozens of loads for the simple setup, so we'll need to move to a more complex (late/end-game) setup to really trigger something.

[sibir-ine]

Haven't been able to replicate it in my test world either, I'll continue to keep an eye out on my MBC server for any more issues.