[sheepdog] [PATCH 0/4] variable vnodes specified by free space

[levin li]

On 2012年08月03日 13:59, MORITA Kazutaka wrote:
> At Wed,  1 Aug 2012 10:56:12 +0800,
> levin li wrote:
>>
>> From: levin li <xingke.lwp at taobao.com>
>>
>> For sheep's first boot, we get the free disk space either by syscall
>> or by user's determination though command line argement, and then we
>> store the size into config file, and for later boot, we just read the
>> size from config file, the space size can never be changed after first
>> boot.
>>
>> The free space size of every node is notified to every other node though
>> join message, and stored in sd_node, every time a new node joins or an
>> old node leaves, we recalculate the vnodes number for every nodes.
>>
>> levin li (4):
>>   sheep: get the free disk space and store it in the config file
>>   sheep: add user-defined free disk space size
>>   sheep: recalculate the vnodes number when epoch changes
>>   sheep: remove command line argument --vnodes
>>
>>  include/internal_proto.h |    1 +
>>  sheep/group.c            |   24 +++++++++++++
>>  sheep/sheep.c            |   22 +++++++-----
>>  sheep/sheep_priv.h       |    3 ++
>>  sheep/store.c            |   86 ++++++++++++++++++++++++++++++++++++++++++++++
>>  5 files changed, 127 insertions(+), 9 deletions(-)
> 
> This patchset doesn't pass the following test:
> 
> ==
> #!/bin/bash
> 
> set -ex
> 
> dd if=/dev/zero of=/tmp/0.img seek=$((2 * 1024 ** 3 - 1)) bs=1 count=1
> dd if=/dev/zero of=/tmp/1.img seek=$((4 * 1024 ** 3 - 1)) bs=1 count=1
> dd if=/dev/zero of=/tmp/2.img seek=$((8 * 1024 ** 3 - 1)) bs=1 count=1
> 
> for i in 0 1 2; do
>     mkfs.xfs /tmp/$i.img
>     mkdir /store/$i
>     mount -o loop /tmp/$i.img /store/$i
>     sheep -d /store/$i/ -z $i -p 700$i
> done
> 
> sleep 1
> 
> for i in 0 1 2; do
>     collie node list -p 700$i
> done
> ==
> 
> The expected result is:
> 
> + collie/collie node list -p 7000
> M   Id   Host:Port         V-Nodes       Zone
> -    0   127.0.0.1:7000         27          0
> -    1   127.0.0.1:7001         54          1
> -    2   127.0.0.1:7002         109          2
> + for i in 0 1 2
> + collie/collie node list -p 7001
> M   Id   Host:Port         V-Nodes       Zone
> -    0   127.0.0.1:7000         27          0
> -    1   127.0.0.1:7001         54          1
> -    2   127.0.0.1:7002         109          2
> + for i in 0 1 2
> + collie/collie node list -p 7002
> M   Id   Host:Port         V-Nodes       Zone
> -    0   127.0.0.1:7000         27          0
> -    1   127.0.0.1:7001         54          1
> -    2   127.0.0.1:7002         109          2
> 
> But I got the following output:
> 
> + collie/collie node list -p 7000
> M   Id   Host:Port         V-Nodes       Zone
> -    0   127.0.0.1:7000      		 42          0
> -    1   127.0.0.1:7001      		 85          1
> + for i in 0 1 2
> + collie/collie node list -p 7001
> M   Id   Host:Port         V-Nodes       Zone
> -    0   127.0.0.1:7000      		 42          0
> -    1   127.0.0.1:7001      		 85          1
> + for i in 0 1 2
> + collie/collie node list -p 7002
> M   Id   Host:Port         V-Nodes       Zone
> -    0   127.0.0.1:7002      		 64          2
> 
> 
> Thanks,
> 
> Kazutaka
> 
I just tested this patch set using your script, and it gives the output:

+ mkdir /store/2
+ mount -o loop /tmp/2.img /store/2
+ sheep/sheep -d /store/2/ -z 2 -p 7002
+ sleep 1
+ for i in 0 1 2
+ collie/collie node list -p 7000
M   Id   Host:Port         V-Nodes       Zone
-    0   127.0.0.1:7000      	27          0
-    1   127.0.0.1:7001      	54          1
-    2   127.0.0.1:7002      	110          2
+ for i in 0 1 2
+ collie/collie node list -p 7001
M   Id   Host:Port         V-Nodes       Zone
-    0   127.0.0.1:7000      	27          0
-    1   127.0.0.1:7001      	54          1
-    2   127.0.0.1:7002      	110          2
+ for i in 0 1 2
+ collie/collie node list -p 7002
M   Id   Host:Port         V-Nodes       Zone
-    0   127.0.0.1:7000      	27          0
-    1   127.0.0.1:7001      	54          1
-    2   127.0.0.1:7002      	110          2

I think it's what we expected, as for your result, I think it's due to not
rebase to the latest master which has just merged the delay recovery patch set,
later I'll resend this patch set.

thanks,

levin