macOS下Docker容器多实例运行Linux可执行文件I/O缓慢问题求解

Hey, let's break down why you're seeing such slow I/O when running multiple Docker instances on macOS, and go through some practical fixes to speed things up:

Why the I/O is slow #

First, let's get to the root of the problem:

• Docker on macOS relies on a virtual machine: All container I/O has to pass through a virtualization layer (HyperKit or the newer Virtualization Framework) to reach your macOS filesystem. This extra layer already adds overhead, and running multiple instances at once amplifies this bottleneck.
• Unnecessary cross-filesystem copies: Your run_my_exec.sh copies my_exec from the container's internal filesystem to the macOS-mounted /rundir_id directory every time you spin up an instance. Copying files between the VM's filesystem and macOS is one of the slowest operations you can do in Docker for Mac—especially with multiple instances doing this simultaneously.

Fix 1: Eliminate cross-filesystem executable copies #

This is the biggest win you can get. Stop copying my_exec to the macOS-mounted directory entirely, and run the executable directly inside the container's filesystem. Only mount the directories you need for inputs and outputs.

Here's how to adjust your setup:

1. Update your run_my_exec.sh to work with separate input/output mounts:

   #!/bin/bash
   export MYEXEC="/usr/local/my_exec_folder/bin/my_exec"
   # Create a temporary working directory inside the container
   WORK_DIR="/tmp/my_exec_workdir"
   mkdir -p $WORK_DIR
   cd $WORK_DIR
   
   # Copy input file from the mounted input directory
   cp /inputs/$2 .
   
   # Run the executable directly from its container location (no cross-filesystem copy!)
   $MYEXEC $2
   
   # Copy output files back to the mounted output directory
   cp ./output* /outputs/
   

2. When running the container, mount separate input and output directories from macOS:

   docker run -t \
     -v /my/rundir_id/on/macos/inputs:/inputs \
     -v /my/rundir_id/on/macos/outputs:/outputs \
     my_exec_c $2
   

Now you only copy small input/output files across the VM-macOS boundary, not the entire executable every time.

Fix 2: Use Docker volumes instead of bind mounts #

Docker named volumes perform better than bind mounts on macOS because they live directly inside the VM's filesystem, avoiding the cross-layer I/O overhead.

• For shared inputs/outputs, create a named volume and populate it once:

  docker volume create my_exec_inputs
  docker cp /my/inputs_on_macos/. my_exec_inputs:/
  

• Run your container using the volume instead of a bind mount:

  docker run -t \
    -v my_exec_inputs:/inputs \
    -v my_exec_outputs:/outputs \
    my_exec_c $2
  

If you need per-instance input directories, create a unique volume for each instance instead of using a bind mount.

Fix 3: Optimize Docker Desktop settings #

Tweak these settings to boost overall I/O performance:

• Limit shared directories: Go to Settings > Resources > File Sharing and only include the specific macOS directories your containers need. Removing unnecessary directories reduces filesystem sync overhead.
• Use the Virtualization Framework: If you're on macOS 12+, enable Settings > Features in Development > Use the new Virtualization framework—it's faster than the older HyperKit backend.
• Allocate more resources: Increase the CPU and memory allocated to Docker in Settings > Resources. Insufficient resources can cause I/O operations to queue up and slow down.

Fix 4: Reuse containers instead of spawning new ones #

If your workflow allows, reuse a single container to run multiple instances of my_exec instead of starting a new container each time. Container startup has overhead, and reusing a container avoids reinitializing filesystem layers repeatedly.

Example workflow:

# Start a long-lived container in the background
docker run -d -t \
  -v /my/rundir_id/on/macos/inputs:/inputs \
  -v /my/rundir_id/on/macos/outputs:/outputs \
  my_exec_c sleep infinity

# Run your executable inside the existing container multiple times
docker exec <container_id> /run_my_exec.sh $2

Fix 5: Verify storage driver #

Ensure Docker is using the overlay2 storage driver (the default on Docker Desktop for Mac), which is optimized for performance. Check with:

docker info | grep "Storage Driver"

If it's not set to overlay2, reset Docker Desktop to defaults to switch to it.

Final Notes #

Start with Fix 1—it addresses the most impactful bottleneck. Combine it with Fix 3 for immediate performance gains. If you still need more speed, explore Fixes 2 and 4 depending on your specific workflow.

内容的提问来源于stack exchange，提问作者newkid