Elixir/Erlang中Random/Rand模块的Seed函数是否需要在每个进程中调用？

Great question! Let's break this down clearly, since the behavior differs between the older random module and the newer, recommended rand module in Erlang/Elixir:

Core Answer #

Both the random and rand modules maintain process-specific PRNG state — meaning seed configurations only apply to the process that calls the seed function. There is no VM-wide seed setting; every process has its own independent random number generator state.

1. Behavior of the older random module #

• By default, every new process starts with the same hardcoded seed. That's why you saw identical random:uniform() outputs across fresh Erlang VM sessions (the shell process uses the default seed each time you start a VM).
• If you call :random.seed/1 or :random.seed/3, this only updates the PRNG state for the current process. Any other processes (even ones spawned later) will still use the default seed unless they explicitly set their own.

Example in Elixir:

# In the main shell process
IO.puts(:random.uniform()) # Output: 0.4435846174457203 (default seed)
:random.seed(:exsplus, {123, 456, 789})
IO.puts(:random.uniform()) # Output: A value based on the new seed

# Spawn a new process
spawn(fn -> IO.puts(:random.uniform()) end) # Output: 0.4435846174457203 (still uses default seed)

2. Behavior of the modern rand module (Erlang/OTP 18+, recommended) #

• This module fixes the biggest pain point of random: it automatically initializes a unique, cryptographically secure seed for each process the first time you call any rand function. The seed is sourced from the OS's random device (like /dev/urandom), so you don't need to manually set seeds for most use cases.
• If you do want to set a custom seed (e.g., for reproducible test sequences), calling :rand.seed/1 or :rand.seed/3 still only affects the current process.

Example in Elixir:

# In process 1
IO.puts(:rand.uniform()) # Output: A unique random value (auto-seeded)

# In process 2 (spawned separately)
spawn(fn -> IO.puts(:rand.uniform()) end) # Output: A different unique random value (its own auto-seed)

Key Takeaways #

• If you're using random: You need to set the seed per process if you want non-identical random sequences across processes.
• If you're using rand: You almost never need to manually set seeds — the module handles per-process initialization automatically. Manual seeds are only for specific use cases like reproducibility.

内容的提问来源于stack exchange，提问作者Jordan Parker