Bug 1378072

regards
Aravinda

On Wednesday 21 September 2016 03:53 PM, Ravishankar N wrote:
> On 09/21/2016 03:34 PM, Aravinda wrote:
>> Hi,
>>
>> We have following SPLIT_BRAIN events
>>
>> EVENT_AFR_SPLIT_BRAIN, "subvol=%s;msg=file type mismatch;gfid=%s;ia_type-%d=%s;ia_type-%d=%s"
>> EVENT_AFR_SPLIT_BRAIN, "subvol=%s;msg=gfid mismatch. Skipping conservative merge.;file=<gfid:%s>/%s>;count=2;child-%d=%s;gfid-%d=%s;child-%d=%s;gfid-%d=%s",
>> EVENT_AFR_SPLIT_BRAIN, "subvol=%s;msg=file type mismatch. Skipping conservative merge;file=<gfid:%s>/%s>;count=2;child-%d=%s;type-%d=%s;child-%d=%s;type-%d=%s",
>> EVENT_AFR_SPLIT_BRAIN, "subvol=%s;msg=file type mismatch;file=<gfid:%s>/%s;count=2;child-%d=%s;type-%d=%s;child-%d=%s;type-%d=%s"
>> EVENT_AFR_SPLIT_BRAIN, "subvol=%s;msg=gfid mismatch;file=<gfid:%s>/%s;count=2;child-%d=%s;gfid-%d=%s;child-%d=%s;gfid-%d=%s"
>>
>> Message keys are not same even though Split brain type is same. For example,
>>
>> EVENT_AFR_SPLIT_BRAIN, "subvol=%s;msg=file type mismatch;gfid=%s;ia_type-%d=%s;ia_type-%d=%s"
>> EVENT_AFR_SPLIT_BRAIN, "subvol=%s;msg=file type mismatch;file=<gfid:%s>/%s;count=2;child-%d=%s;type-%d=%s;child-%d=%s;type-%d=%s"
>>
>> We can split these events into two types.(Message included as EVENT name itself, so that no separate msg is required.)
>>
>> EVENT_AFR_SPLIT_BRAIN_FILE_TYPE_MISMATCH, "subvol=%s;file=<gfid:%s>/%s;count=2;child-%d=%s;type-%d=%s;child-%d=%s;type-%d=%s"
>> EVENT_AFR_SPLIT_BRAIN_GFID_MISMATCH, "subvol=%s;file=<gfid:%s>/%s>;count=2;child-%d=%s;gfid-%d=%s;child-%d=%s;gfid-%d=%s"
>>
>> Let me know your thoughts.
>>
>
> I think it is better to retain one EVENT type for split-brains. Otherwise we need to keep on adding different EVENT types for example when we need to propagate data split-brain or metadata split-brain etc.
> The msg=<message> also gives us a way to specific a more verbose message that is immediately available to the consumer, should they decide to parse it. Also, for all types of split-brains, there has to be a remedial action (i.e. resolving the split-brains) required by the admin. Having to monitor only one EVENT type for all split-brains would make  it easier is what I feel.
If it is one event type, then I would prefer "type" instead of msg. (type can be gfid|file|data|meta etc). Descriptive message is not required since these APIs are consumed programatically, free text and non uniform key names breaks the parsing.
>
> -Ravi