Why is this seed not part if the "header" of the disk partition? Generated upon 
cache initiization.

> On Aug 12, 2014, at 9:44 AM, "ASF subversion and git services (JIRA)" 
> <[email protected]> wrote:
> 
> 
>    [ 
> https://issues.apache.org/jira/browse/TS-3000?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14093633#comment-14093633
>  ] 
> 
> ASF subversion and git services commented on TS-3000:
> -----------------------------------------------------
> 
> Commit 83248169bbdaab383583fbb0b25f0916b18594c4 in trafficserver's branch 
> refs/heads/master from [~amc]
> [ https://git-wip-us.apache.org/repos/asf?p=trafficserver.git;h=8324816 ]
> 
> TS-3000: Add seed string for cache storage.
> 
> 
>> Enable administrator control of seed strings for stripe assigment
>> -----------------------------------------------------------------
>> 
>>                Key: TS-3000
>>                URL: https://issues.apache.org/jira/browse/TS-3000
>>            Project: Traffic Server
>>         Issue Type: Improvement
>>         Components: Cache
>>           Reporter: Alan M. Carroll
>>           Assignee: Alan M. Carroll
>>            Fix For: 5.1.0
>> 
>> 
>> Objects are assigned to stripes based on a hash. The seed string for this 
>> hash is computed from properties of the storage. This means that if the 
>> storage is changed in any way, the hashing changes which disrupts assignment 
>> on all other stripes.
>> An actual example is for a deployment that uses /dev/sdb .. /dev/sdx. If one 
>> of these disks fails (say /dev/sdm) then all of the other drives will "move 
>> down' at the next system restart which will in turn change the hash seed 
>> string for them which will make almost all of the cached objects on those 
>> drives inaccessible. Losing a single drive could wipe out most of the cache.
>> This fix allows an administrator to specify the hash seed string in the 
>> storage.config file. The seed strings (and hence the resulting hash) can be 
>> kept consistent for a physical device even if the operating system 
>> assignment is changed.
>> P.S. As a side note for Linux users, there are symlinks for raw disk devices 
>> in /dev/disk. These can be used in storage.config to avoid some of these 
>> problems.
> 
> 
> 
> --
> This message was sent by Atlassian JIRA
> (v6.2#6252)

Reply via email to