Open Stack

On 02/27/2015 10:50 AM, James E. Blair wrote:
> Ian Wienand <iwienand at redhat.com> writes:
> 
>> Hi,
>>
>> Firstly, being someone who is trying to keep several images going in
>> nodepool and being some distance away from the nodepool servers, the
>> current situation of it logging to a single rather large file is often
>> annoying.  Yes I do know how to use grep, but IMO the use-case for
>> wanting every single log message all jumbled up is weaker than the
>> case for splitting it up.
> 
> Yes, there is no good use case for that, I would love for them to be
> split.
> 
>> I'm trying to get on-top of fedora kernel issues; compare logs for
>> centos builds to some manual testing being done and generally monitor
>> the health of the builds.
> 
> By the way, does a resumed transfer work for a partial download (ie, the
> HTTP Range header)?
> 
>> There is a proposal to have nodepool automatically split out these
>> logs [1], have puppet deal with it [2] and I'm willing to add nodepool
>> support for re-reading the config if we can agree.
> 
> I do not think that OpenStack's logging configuration should be
> contained within the nodepool source code.  I think any of these
> alternate approaches would be okay:
> 
> 1) Have a logger definition in a config file that is treated as a
> template and then automatically duplicated for each provider-image
> combination.
> 
> 2) Stop using the python logging module and just take a path for the
> directory and write files there, leaving rotation/cleanup to an external
> tool (that we would manage with puppet).
> 
> 3) Manually create log config file entries for DIB images, since there
> won't be as many of them.  Once we finish moving away from snapshot
> images, the set of logs and configuration for them will be manageable.
> 
> There might be other options I haven't thought of.

4) Just use syslog, and have splitting/rotation handled at the system
level by syslog config

(not saying that's preferrable, just where my brain went wrt listing things)