mojira.dev
MC-202249

Angering passive mobs in new nether chunks very far away from previously generated chunks, then entering a nether portal causes server to completely freeze

Hello,
Descovered a bug that goes far at least as 1.16, this is possibily also on 1.15 with Zombie Pigmen.
I cannot seem to attach previous versions to the affected versions.

Each test was done after deleting the region files in the nether to start cleanly.
I created a datapack to spawn 40 piglins and give a bow/arrow, this is attached below.
Memory allocation does not seem to matter, I tested this locally with Xms/Xmx 2G, while this still occurs on my main server with 5G set.
The setup:
1. Start a brand new server.
2. Turn off Mob Spawning.
3. Go to nether, go to the roof for simplicity.
4. Teleport to 75,000 either X or Z (Minimum to effect is around 60K, 75K 100% garuntees a crash)
5. Create a nether portal and go through it to generate the overworld chunks to verify it does not crash which it does not.
6. Surround the portal with fencing or another way to prevent access, but have enough room to fire an arrow from.
7. Summon Zombie Piglins outside of the fence contraption, 40 will 100% garuntee a crash in my testing
8. Go into survival while standing in the fence contraption with the portal. anger the Zombie Piglins.
9.Go through nether portal.
Notice the Rx in F3 Debug screen slowly decreasing to 0 before freezing.

Video attached, along with the 60s tick timeout watchdog crash – no other errors are created or shown... video is too large to attach So I have used an external source to link them.

Edit: This causes the Watchdog to do it's job and terminate the hung java process, otherwise the server can hang indefinately if the watchdog is set to -1
This is not a bug with the watchdog.

Video from 1.16.2: https://streamable.com/fzb250
Video testing on 1.16.4-pre1: https://streamable.com/e4ue8l

[media][media][media]

Linked issues

Attachments

Comments 10

can confirm

doesn't need new server. Just go far out in the nether

Duplicate of MC-63590, please read the moderator note on that report.

Actually Galaxy, The watchdog is doing it's job in this case.
With the watchdog set to -1 the server hangs indefinately and the only thing that gets logged into latest.log is authentication (connection) attempts.

Also do you mind trying to reproduce this yourself and have a look?
This is not a duped bug report.

I get what you mean, the crash is irrelevant in that case then, the fact that the game freezes is when server-tick-time is set to -1. Reopened now - however, I did notice you are using a third party launcher for starting the game. Please do note that this does usually invalidate any reports you create, though in this case, I can clearly see that it's not coming from the launcher but the server.

What could also be interesting is to see whether or not the log provides anything after the server freezes for several minutes. Also, whether or not this also happens in Singleplayer as well.

Thanks.
I really wanted to make sure that this bug is not burried, as this had recked havoc on my server for a few versions – didn't feel like I had enough research on it to report at the time until now and after reading other bug reports, and thought to give whatever I had.

I just saw your edit Galaxy.
The launcher is just a bash file to run the java command to include Xms/Xmx and purely launches the downloaded jar, nothing else.

The log litterally provides nothing. No Rcon connections, no cant keep up!'s...
It will show "User Authenticator" messages.

I created an external linux script to look for these messages and act as a different kind of watchdog for when it reaches a specific ammount of them signifying a freeze and players trying to re-join – This was not used in this case and I only use it on my main server... I have it record graphs of when these occur, but I do not believe they will be helpful, and purely for logging the freeze events ( and besides, that server is running fabric for other performance tweaks, so it invalidates using those graphs )

Ah well, it was a long shot, thank you for checking.

Sorry I should have mentioned that it doesnt matter about newly generated chunks, it'll do reguardless.
I dont make reports often, I'll try to be more inclusive next time.

Also to note, i did one test in 1.16.2 where the zpiglins were inside a fence contraption and the portal was not, it didn't crash... but the other way around did.

can confirm for singleplayer 1.16.3.
https://youtu.be/qXU7VBZYUm4

the world completely freezes, and will hang indefinitely on "Saving world" if you try to exit.

edit: reuploaded video to youtube.

Some clarification on trigger conditions... Whether or not the area has been generated or not does not matter, as long as it's amply far out. Teleporting away can also trigger it.

Jake Spencer

slicedlime

Community Consensus

Very Important

Crash

1.16.3, 1.16.4 Pre-release 1, 1.16.4 Pre-release 2, 1.16.4 Release Candidate 1, 1.16.4, 1.16.5, 21w17a

1.17 Pre-release 1

Retrieved