LMDB memory is overcommitted by default on Raspberry Pi

[Swiddis]

Describe the Bug

Running the default config of 0.1.0-rc2 on a raspberry Pi 4-B (aarch64, libc, Debian 13) immediately OOMs:

/ferrumc-0.1.0-rc2 $ ./target/release/ferrumc
2026-04-20T06:26:12.312360Z  INFO Starting server...
2026-04-20T06:26:12.312663Z  INFO Server setup complete.

thread 'main' (120408) panicked at src/lib/world/src/lib.rs:98:14:
Failed to initialize database: DatabaseInitError("Cannot allocate memory (os error 12)")
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

With memory stracing:

/ferrumc-0.1.0-rc2 $ RUST_BACKTRACE=1 strace -f -e trace=mmap,brk  ./target/release/ferrumc
brk(NULL)                               = 0x55b230d000
mmap(NULL, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fba3f5000
mmap(NULL, 27392, PROT_READ, MAP_PRIVATE, 3, 0) = 0x7fba3ee000
mmap(NULL, 263104, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_DENYWRITE, -1, 0) = 0x7fba37c000
mmap(0x7fba380000, 197568, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0) = 0x7fba380000
mmap(0x7fba3af000, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x1f000) = 0x7fba3af000
mmap(NULL, 720920, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_DENYWRITE, -1, 0) = 0x7fba2cf000
mmap(0x7fba2d0000, 655384, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0) = 0x7fba2d0000
mmap(0x7fba36f000, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x8f000) = 0x7fba36f000
mmap(NULL, 1892368, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_DENYWRITE, -1, 0) = 0x7fba101000
mmap(0x7fba110000, 1826832, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0) = 0x7fba110000
mmap(0x7fba2bd000, 20480, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x19d000) = 0x7fba2bd000
mmap(0x7fba2c2000, 49168, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x7fba2c2000
mmap(NULL, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fba3ec000
brk(NULL)                               = 0x55b230d000
brk(0x55b232e000)                       = 0x55b232e000
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_DROPPABLE|MAP_ANONYMOUS, -1, 0) = 0x7fba3f4000
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fba3f3000
mmap(NULL, 20480, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_STACK, -1, 0) = 0x7fba3ee000
2026-04-20T06:26:35.273296Z  INFO Starting server...
2026-04-20T06:26:35.274054Z  INFO Server setup complete.
mmap(NULL, 1052672, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fba00f000
mmap(NULL, 2101248, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fb9e0e000
mmap(NULL, 8256, PROT_READ|PROT_WRITE, MAP_SHARED, 4, 0) = 0x7fba3e9000
mmap(NULL, 1073741824000, PROT_READ, MAP_SHARED, 5, 0) = -1 ENOMEM (Cannot allocate memory)

thread 'main' (120501) panicked at src/lib/world/src/lib.rs:98:14:
Failed to initialize database: DatabaseInitError("Cannot allocate memory (os error 12)")
stack backtrace:
mmap(NULL, 32654832, PROT_READ, MAP_PRIVATE, 4, 0) = 0x7fb81eb000
brk(0x55b2356000)                       = 0x55b2356000
brk(0x55b23d4000)                       = 0x55b23d4000
brk(0x55b2464000)                       = 0x55b2464000
brk(0x55b2584000)                       = 0x55b2584000
   0: __rustc::rust_begin_unwind
   1: core::panicking::panic_fmt
   2: core::result::unwrap_failed
   3: <ferrumc_world::World>::new::<&alloc::string::String>
   4: ferrumc::launch::create_state
   5: ferrumc::main
mmap(NULL, 1716616, PROT_READ, MAP_PRIVATE, 4, 0) = 0x7fb8047000
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
+++ exited with 101 +++

The mmap(NULL, 1073741824000, PROT_READ, MAP_SHARED, 5, 0) = -1 ENOMEM (Cannot allocate memory) line points at the default config of taking 1 TB of shmem. This is workaroundable by just changing the limit to e.g. 100GB in config.toml.

To Reproduce

Just cargo run --release on this machine, specifically with the default map_size = 1000:

OS: Debian GNU/Linux 13 (trixie) aarch64
Host: Raspberry Pi 4 Model B Rev 1.4
Kernel: Linux 6.12.75+rpt-rpi-v8
CPU: BCM2711 (4) @ 1.80 GHz
GPU: Broadcom bcm2711-vc5 [Integrated]

Expected Behaviour

Work OOTB.

The bug is easy to work around: reduce the map_size. I'm creating this issue to help others who run into it, and to see if the default could be fixed on this platform.

Screenshots / Videos

N/A

FerrumC Version

0.1.0-rc2 (built from source)

Rust Version

rustc 1.97.0-nightly (e22c616e4 2026-04-19)

Server Logs

See above

Additional Context

No response

Is there an existing issue for this?

• I have searched current existing issues before creating this.
• I checked all the boxes without reading them
• I have provided all necessary and relevant details, including logs, reproduction steps, etc.