Here is another clue from my NFS mount: the directory entry comes up empty
until I write to it:
0-16:01 djh@c24-03-06 ~> *touch /hadoop/wxxxs/data/foo*
0-16:01 djh@c24-03-06 ~> *ls /hadoop/wxxxs/data/*
foo  packed-hbfs/  raw/  tmp/
0-16:01 djh@c24-03-06 ~> *rm /hadoop/wxxxs/data/foo*
0-16:01 djh@c24-03-06 ~> *ls /hadoop/wxxxs/data/*
packed-hbfs/  raw/  tmp/

On Fri, Jun 12, 2020 at 11:51 AM Daniel Howard <[email protected]> wrote:

> Hello,
>
> I wanted to provide NFS access on a cluster. Since the NFS server is
> lightweight, and since I use config management, I figured I have every node
> run its local nfs3 service, and have the nodes mount localhost:/.
>
> This strategy appears to work fine, at first, but before long, I get into
> a state where each system is seeing a different truth on NFS.
>
> Example:
>
> 0-11:45 djh@c24-03-06 ~> *hadoop fs -ls /data/*
> Found 2 items
> drwxrwxr-x+  - djh djh          0 2020-06-10 20:51 /data/packed-hbfs
> drwxrwxr-x+  - djh djh          0 2020-06-11 14:06 /data/tmp
> 0-11:43 djh@c24-03-06 ~> *ls /hadoop/wxxxs/data/*
> 0-11:43 djh@c24-03-06 ~> *ls /hadoop/wxxxs/data/tmp*
> allocation_2020-06-10_1591799995129_0.json
> allocation_2020-06-10_1591809732060_0.json
> allocation_2020-06-10_1591856129529_0.json
> allocation_2020-06-10_1591856640093_0.json
> allocation_2020-06-10_1591858477153_0.json
> allocation_2020-06-11_1591884434468_0.json
> allocation_2020-06-11_1591886358283_0.json
> allocation_2020-06-11_1591896890160_0.json
>
> 0-11:43 djh@c24-03-08 ~> *ls /hadoop/wxxxs/data*
> packed-hbfs/  tmp/
> 0-11:43 djh@c24-03-08 ~> *ls /hadoop/wxxxs/data/tmp*
> allocation_2020-06-10_1591799995129_0.json
> allocation_2020-06-10_1591809732060_0.json
> allocation_2020-06-10_1591856129529_0.json
> allocation_2020-06-10_1591856640093_0.json
> allocation_2020-06-10_1591858477153_0.json
> allocation_2020-06-11_1591884434468_0.json
> allocation_2020-06-11_1591886358283_0.json
> allocation_2020-06-11_1591896890160_0.json
>
> My best guess is that the NameNode assumes there is only one NFS, so the
> NameNode only sends updates to one node?
>
> Is what I am doing supported? Alternatively, is there a decent FUSE
> implementation for Apache Hadoop? I can switch to having a single NFS
> server if that is what is supported.
>
> Thanks,
> -danny
>
>
> --
> http://dannyman.toldme.com
>


-- 
http://dannyman.toldme.com

Reply via email to