[Gluster-users] Request for user comments
ravishankar at redhat.com
Wed Jun 17 18:04:07 UTC 2015
On 06/17/2015 10:58 PM, Pranith Markup Karampuri wrote:
> On 06/17/2015 10:44 PM, Ravishankar N wrote:
>> On 06/17/2015 10:34 PM, Pranith Kumar Karampuri wrote:
>>> Depends on the log size. Largest logfile I debugged from customers
>>> is >20GB. Production logs are generally of the order of GBs.
>>> Readability hardly matters when things get to that size.
>> On the contrary, readability is what matters the most. What you
>> cannot read easily you cannot interpret easily.
> Nah! what you can not process easily you can not interpret at that scale.
>>> You need to be able to process the logs and get useful information
>>> with least amount of work. With the message-id framework we are
>>> going towards this will be even more important to have the whole log
>>> in same line, so that we can grep using msg-id alone.
>> Again, how does the patch affect the 'grep'-ability of logs?
> If there are 6 code paths which are passing NULL to dict_ref. All of
> them use msg-id LG_INVALID_ARG in the log file. To figure out the uniq
> code paths the command is:
> "grep LG_INVALID_ARG <log-file> | grep dict_ref | sort | uniq -c"
> Assume the log is 20GB. And there are 100000 dict_ref logs. What will
> be the command to get the info above after this change is merged?
Ah, I now see where you are coming from. But to be fair, since there
will be a delta in the time-stamps of the messages, your uniq filter
would anyway end up printing all 100000 log counts.
More information about the Gluster-users