Tarantool因磁盘写入错误无法启动。

Question

我尝试从头开始在Docker中启动Tarantool（没有现有数据）。 我使用他们在教程中建议的Docker命令，并在MacOS 10.15.6（Catalina）上的Docker Desktop 2.4.0.0下运行：

docker run \
  --name mytarantool \
  -d -p 3301:3301 \
  -v /data/dir/on/host:/var/lib/tarantool \
  tarantool/tarantool:2.5.1

(/data/dir/on/host是在我的笔记本电脑上的本地目录，用于替换). 我还尝试使用最新版本2.6.0。

容器启动后很快就会终止。docker logs显示如下：

2020-10-02 20:51:10.331 [1] main/103/tarantool-entrypoint.lua C> Tarantool 2.6.0-0-g47aa4e01e
2020-10-02 20:51:10.331 [1] main/103/tarantool-entrypoint.lua C> log level 5
2020-10-02 20:51:10.332 [1] main/103/tarantool-entrypoint.lua I> mapping 268435456 bytes for memtx tuple arena...
2020-10-02 20:51:10.332 [1] main/103/tarantool-entrypoint.lua I> mapping 134217728 bytes for vinyl tuple arena...
2020-10-02 20:51:10.335 [1] main/103/tarantool-entrypoint.lua I> instance uuid 1811ff01-13d1-45c8-9878-0974bf27ee40
2020-10-02 20:51:10.335 [1] iproto/101/main I> binary: bound to 0.0.0.0:3301
2020-10-02 20:51:10.335 [1] main/103/tarantool-entrypoint.lua I> initializing an empty data directory
2020-10-02 20:51:10.351 [1] main/103/tarantool-entrypoint.lua I> assigned id 1 to replica 1811ff01-13d1-45c8-9878-0974bf27ee40
2020-10-02 20:51:10.351 [1] main/103/tarantool-entrypoint.lua I> cluster uuid 12ca546b-29ea-4af3-a407-f24e91c0e636
2020-10-02 20:51:10.357 [1] snapshot/101/main I> saving snapshot `/var/lib/tarantool/00000000000000000000.snap.inprogress'
2020-10-02 20:51:10.361 [1] snapshot/101/main I> done
2020-10-02 20:51:10.364 [1] main/103/tarantool-entrypoint.lua I> ready to accept requests
2020-10-02 20:51:10.365 [1] main/103/tarantool-entrypoint.lua I> set 'log_level' configuration option to 5
2020-10-02 20:51:10.365 [1] main/105/checkpoint_daemon I> scheduled next checkpoint for Fri Oct  2 22:10:11 2020
2020-10-02 20:51:10.367 [1] main/103/tarantool-entrypoint.lua I> set 'listen' configuration option to "3301"
2020-10-02 20:51:10.367 [1] main/103/tarantool-entrypoint.lua I> set 'log_format' configuration option to "plain"
2020-10-02 20:51:10.384 [1] wal/101/main xlog.c:1026 !> SystemError /var/lib/tarantool/00000000000000000000.xlog: can't allocate disk space: Invalid argument
2020-10-02 20:51:10.384 [1] main/103/tarantool-entrypoint.lua txn.c:876 E> ER_WAL_IO: Failed to write to disk
2020-10-02 20:51:10.391 [1] main txn.c:876 E> ER_WAL_IO: Failed to write to disk
2020-10-02 20:51:10.391 [1] main F> fatal error, exiting the event loop

同时，容器成功创建了5.9K个00000000000000000000.snap文件和97B个00000000000000000000.xlog文件。

$ ls -hal
total 24
drwxr-xr-x@ 4 user  staff   128B  2 Oct 13:51 .
drwxr-xr-x  3 user  staff    96B  2 Oct 12:56 ..
-rw-r--r--  1 user  staff   5.9K  2 Oct 13:51 00000000000000000000.snap
-rw-r--r--  1 user  staff    97B  2 Oct 13:51 00000000000000000000.xlog

如果我启动容器而不挂载本地目录，它会成功。

我猜测我的本地文件系统存在问题（或者以某种方式从容器中可见）或者可能有权限问题，但是我无法确定具体原因。

如果我在成功启动的容器中使用exec作为shell，我将看到xlog文件更大且文件的所有者是tarantool:tarantool。

$ docker exec -it 016 sh
/opt/tarantool # ls -hal /var/lib/tarantool/
total 1044
drwxr-xr-x    2 tarantoo tarantoo    4.0K Oct  2 20:40 .
drwxr-xr-x    1 root     root        4.0K Aug  2 16:31 ..
-rw-r--r--    1 tarantoo tarantoo    5.9K Oct  2 20:40 00000000000000000000.snap
-rw-r--r--    1 tarantoo tarantoo     273 Oct  2 20:40 00000000000000000000.xlog

但是如果绑定了目录，情况就不同了：

$ docker run -it -p 3031:3031 -v /Users/user/project/storage:/var/lib/tarantool tarantool/tarantool:2.6.0 sh
/opt/tarantool # ls -hal /var/lib/tarantool/
total 16
drwxr-xr-x    4 tarantoo root         128 Oct  2 21:18 .
drwxr-xr-x    1 root     root        4.0K Aug  2 16:31 ..
-rw-r--r--    1 root     root        5.9K Oct  2 21:18 00000000000000000000.snap
-rw-r--r--    1 root     root          97 Oct  2 21:18 00000000000000000000.xlog

我试图更改目录和文件的所有者：

$ docker run -it -p 3031:3031 -v /Users/user/project/storage:/var/lib/tarantool tarantool/tarantool:2.6.0 sh
/opt/tarantool # chown tarantool:tarantool -R /var/lib/tarantool/

并检查更改是否在容器重启后持久存在：

$ docker run -it -p 3031:3031 -v /Users/user/project/storage:/var/lib/tarantool tarantool/tarantool:2.6.0 sh
/opt/tarantool # ls -hal /var/lib/tarantool/
total 16
drwxr-xr-x    4 tarantoo tarantoo     128 Oct  2 21:18 .
drwxr-xr-x    1 root     root        4.0K Aug  2 16:31 ..
-rw-r--r--    1 tarantoo tarantoo    5.9K Oct  2 21:18 00000000000000000000.snap
-rw-r--r--    1 tarantoo tarantoo      97 Oct  2 21:18 00000000000000000000.xlog

现在权限看起来跟工作容器里的一样。但是正常启动容器最终也会遇到同样的问题：

$ docker run -it -p 3031:3031 -v /Users/user/project/storage:/var/lib/tarantool tarantool/tarantool:2.6.0
Creating configuration file: /etc/tarantool/config.yml
Config:
---
force_recovery: false
memtx_dir: /var/lib/tarantool
listen: 3301
pid_file: /var/run/tarantool/tarantool.pid
vinyl_dir: /var/lib/tarantool
wal_dir: /var/lib/tarantool
```
2020-10-02 21:22:29.680 [1] main/103/tarantool-entrypoint.lua C> Tarantool 2.6.0-0-g47aa4e01e
2020-10-02 21:22:29.681 [1] main/103/tarantool-entrypoint.lua C> log level 5
2020-10-02 21:22:29.685 [1] main/103/tarantool-entrypoint.lua I> mapping 268435456 bytes for memtx tuple arena...
2020-10-02 21:22:29.685 [1] main/103/tarantool-entrypoint.lua I> mapping 134217728 bytes for vinyl tuple arena...
2020-10-02 21:22:29.687 [1] main/103/tarantool-entrypoint.lua I> instance uuid 74d33452-7f39-4ebf-a2f7-c1da6cb8c54b
2020-10-02 21:22:29.691 [1] main/103/tarantool-entrypoint.lua I> instance vclock {}
2020-10-02 21:22:29.691 [1] iproto/101/main I> binary: bound to 0.0.0.0:3301
2020-10-02 21:22:29.693 [1] main/103/tarantool-entrypoint.lua I> recovery start
2020-10-02 21:22:29.694 [1] main/103/tarantool-entrypoint.lua I> recovering from `/var/lib/tarantool/00000000000000000000.snap'
2020-10-02 21:22:29.695 [1] main/103/tarantool-entrypoint.lua I> cluster uuid 258099e8-803c-4a10-a3e0-57f6cd796f18
2020-10-02 21:22:29.708 [1] main/103/tarantool-entrypoint.lua I> assigned id 1 to replica 74d33452-7f39-4ebf-a2f7-c1da6cb8c54b
2020-10-02 21:22:29.709 [1] main/103/tarantool-entrypoint.lua I> recover from `/var/lib/tarantool/00000000000000000000.xlog'
2020-10-02 21:22:29.710 [1] main/103/tarantool-entrypoint.lua recovery.cc:156 W> file `/var/lib/tarantool/00000000000000000000.xlog` wasn't correctly closed
2020-10-02 21:22:29.713 [1] main/103/tarantool-entrypoint.lua I> ready to accept requests
2020-10-02 21:22:29.713 [1] main/103/tarantool-entrypoint.lua C> leaving orphan mode
2020-10-02 21:22:29.713 [1] main/103/tarantool-entrypoint.lua I> set 'log_level' configuration option to 5
2020-10-02 21:22:29.713 [1] main/105/checkpoint_daemon I> scheduled next checkpoint for Fri Oct  2 23:01:09 2020
2020-10-02 21:22:29.720 [1] main/103/tarantool-entrypoint.lua I> set 'listen' configuration option to "3301"
2020-10-02 21:22:29.720 [1] main/103/tarantool-entrypoint.lua I> set 'log_format' configuration option to "plain"
2020-10-02 21:22:29.723 [1] wal/101/main xlog.c:1026 !> SystemError /var/lib/tarantool/00000000000000000000.xlog: can't allocate disk space: Invalid argument
2020-10-02 21:22:29.723 [1] main/103/tarantool-entrypoint.lua txn.c:876 E> ER_WAL_IO: Failed to write to disk
2020-10-02 21:22:29.726 [1] main txn.c:876 E> ER_WAL_IO: Failed to write to disk
2020-10-02 21:22:29.726 [1] main F> fatal error, exiting the event loop

奇怪的是，在本地电脑上检查权限时，权限根本没有改变。我猜这是Docker的魔法，但考虑到权限更改在容器重启之间仍然存在，我不确定它是如何工作的。

但也许问题与权限无关...那么该怎么解决问题呢？

- greatvovan

---

我在日志中看到了这个错误：无法分配磁盘空间：无效的参数。你的磁盘空间用完了吗？（运行 df -h 命令） - Nick ODell

@NickODell 磁盘空间没问题，它显示在容器中：grpcfuse 465.6G 405.0G 43.4G 90% /var/lib/tarantool。但是在 Docker 中检查磁盘空间的过程可能存在错误... - greatvovan

嗯...但这给了我一个想法：他们在某个时候引入了gRPC FUSE技术进行文件共享，并默认启用。我将其关闭，这有所帮助！现在它可以工作了！ - greatvovan

Answer 1

回答我的问题。

显然问题与权限无关，问题在文件系统虚拟化层面上。

可以通过在 Docker Desktop 首选项中关闭 "gRPC FUSE" 功能来解决此问题。据推测（因为它说 can't allocate disk space: Invalid argument），问题在于此实现不支持某些特定参数的 fallocate()（请参见 https://github.com/docker/for-mac/issues/4964#issuecomment-702748937）。

更新：如果您想使用 gRPC FUSE 功能进行文件共享，请考虑升级到修复了此问题的版本 2.4.2.0。

- greatvovan