如何用mpi4py（Python2）实现其他进程向进程0发消息及recv报错修复

Hey there, let's break down why your comm.recv(source=tempRank) line is getting stuck, and walk through how to fix it so rank 0 can successfully receive messages from all other processes.

Why Your Code Is Hanging #

The most common culprits here are:

• Missing or mismatched send calls: If any worker process (rank != 0) isn't actually sending a message to rank 0, rank 0 will wait forever for that missing message.
• Incorrect source rank in your loop: If your tempRank variable includes rank 0 itself, rank 0 will wait for a message from itself that never arrives.
• Race conditions or early process exit: Sometimes worker processes might exit before executing their send call, leaving rank 0 hanging.

Working Implementation Example #

Here's a tested, minimal code snippet that does exactly what you want—all non-0 processes send a message to rank 0, which receives each one:

from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.Get_rank()
process_count = comm.Get_size()

if rank == 0:
    # Receive messages from every other process
    for worker_rank in range(1, process_count):
        print(f"Rank 0 waiting for message from rank {worker_rank}...")
        received_msg = comm.recv(source=worker_rank)
        print(f"Rank 0 got: {received_msg} from rank {worker_rank}")
else:
    # Worker processes send their message to rank 0
    outgoing_msg = f"Task completed from rank {rank}"
    print(f"Rank {rank} sending message to rank 0...")
    comm.send(outgoing_msg, dest=0)

How to Fix Your Specific Code #

Let's target the likely issues in your code:

1. Check your source rank loop: Make sure you're only iterating over ranks 1 to process_count - 1 (not including 0). If you accidentally include rank 0, your code will hang waiting for a message that rank 0 never sends to itself.
2. Verify worker processes are sending: Double-check that every non-0 rank has a comm.send(..., dest=0) call. If any worker skips this (e.g., due to an unhandled condition), rank 0 will stall.
3. Test with simple messages first: If you're sending complex objects, try swapping them for a simple string or integer to rule out serialization issues (mpi4py handles most types, but edge cases can cause problems).
4. Confirm you're launching processes correctly: Use mpiexec or mpirun to launch your script, e.g., mpiexec -n 5 python your_script.py. If you launch fewer processes than your code expects, rank 0 will wait for non-existent workers.

Debugging Tips #

• Add print statements right before comm.send() in worker processes to confirm they're executing the send.
• Print the value of tempRank in rank 0's loop to ensure you're targeting valid worker ranks.
• Use comm.Probe() to check if a message is available before receiving:

  if rank == 0:
      for worker_rank in range(1, process_count):
          status = MPI.Status()
          # Wait until a message is available from this worker
          comm.Probe(source=worker_rank, status=status)
          print(f"Message ready from rank {worker_rank} (tag: {status.Get_tag()})")
          received_msg = comm.recv(source=worker_rank)
  

  This helps confirm whether the issue is that no message is being sent, or something else is wrong with the receive.

内容的提问来源于stack exchange，提问作者greg