Add command to remove lock files

[RichardLitt]

We should have a command - ipfs repo repair or ipfs repo fsck - which removes lockfiles. See bug issue here.

[whyrusleeping]

ipfs repo fsck has been planned. We should do that soon.

[TheNain38]

I think that when starting IPFS, it should try to remove stale lock file, by searching if the a proccessus with the owner's PID exists, if it doesn't, then it should remove the repo.lock file and start normally

[scottscheper]

For the record, it's February 11, 2019 and I just experienced this issue.

Removing BOTH of these files solved it:

1. ~/.ipfs/repo.lock
2. ~/.ipfs/datastore/LOCK

[Stebalien]

You should never need to do this at this point. If you do, something is wrong.

[Kubuxu]

@sscheper which version of go-ipfs are you using?

[mrusme]

Ran into Error: resource temporarily unavailable inside a self-built Docker container that currently runs 0.5.1. Added rm -rf /data/ipfs/repo.lock /data/ipfs/datastore/LOCK to the very first line of the startup script, so that it always removes the locks that might still exist on the persistent storage from a previously running container. The daemon now starts again. However, if I dial into that container and try to perform an ipfs add -Q -r ./site (while the daemon is running within that container) it still shows Error: resource temporarily unavailable. Not exactly sure what's going on there, maybe someone here can provide further insights?

[Stebalien]

I really wouldn't do that. It no longer matters if the lockfile exists unless a go-ipfs daemon has the lockfile open and locked. That would mean that the daemon is running.

Your second issue may be that ipfs is running but your IPFS node can't find and/or connect to it. I don't know what your exact setup is so I'm not sure why. If you'd like more help on this, please ask on https://discuss.ipfs.io.

[koxon]

While running kubo IPFS on containers and network storage. I see that when a container is killed or fails, the lock files remains and prevent a new daemon from starting.

[xueshuai0922]

While running kubo IPFS on containers and network storage. I see that when a container is killed or fails, the lock files remains and prevent a new daemon from starting.在容器和网络存储上运行 kubo IPFS 时。我看到当容器被杀死或失败时，锁定文件会保留并阻止新的守护程序启动。

yes ,i meet a same problem,should kill lock first?

[koxon]

How to remove the lock file when using Docker? I try to override the entrypoint in the dockerfile to perform the rm as root but I have no luck so far. Anyone managed to make this work when using Docker?

My repo.lock file is still owned by root which shouldn't be the case.

Also, can't we trap the SIGTERM signal and cleanup the lock file before shutting down?

[koxon]

I've built my own entrypoint.sh and Dockerfile to deal with the repo.lock files. I also make use of ipfs startup script to deep clean.

I'd love some feedback.

Dockerfile:

FROM ipfs/kubo:latest

COPY 001-config.sh /container-init.d/001-config.sh

COPY entrypoint.sh /entrypoint.sh
RUN chmod +x /entrypoint.sh

ENTRYPOINT ["/entrypoint.sh"]

entrypoint.sh:

#!/bin/sh

trap stop SIGTERM SIGINT SIGQUIT SIGHUP ERR

start(){
    echo "Deleting lock files"
    rm -rf /data/ipfs/repo.lock /data/ipfs/datastore/LOCK /data/ipfs/blocks/.temp

    echo "Starting IPFS daemon as ipfs user"
    tini -- gosu ipfs /usr/local/bin/start_ipfs daemon --migrate=true --agent-version-suffix=docker
}

stop(){
   echo "Shutting down IPFS as ipfs user"
   tini -- gosu ipfs /usr/local/bin/ipfs shutdown

   exit
}

start

001-config.sh:

#!/bin/sh
set -ex

rm -rf /data/ipfs/repo.lock /data/ipfs/datastore/LOCK /data/ipfs/blocks/.temp

The locks files are removed in the entrypoint.sh which runs as root and again in the 001-config_efs.sh script that is executed by the ipfs daemon at startup as the user ipfs and not root. I take no chance!

It appears that my cluster has recovered and the repo.lock file has been removed since I updated my Docker image.

The entrypoint also attempt to shut down ipfs gracefully if it receives a SIGTERM. This is useful if you're running your task on AWS Fargate SPOT where your tasks can be shutdown by AWS with a 2 min warning.