cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Marcus Sorensen <shadow...@gmail.com>
Subject Re: Review Request 13072: Print agent host name in logging of agent commands
Date Thu, 01 Aug 2013 13:55:16 GMT
I agree, that would be useful.

The issue im resolving here is that 1) not everyone who has access to look
at the logs and troubleshoot also has access to (or knows the schema of)
the database. There might be an issue with a KVM host, but the admin will
waste time manually hunting down which host the command went to because he
has no idea what "via:1237" means and 2) even with db access, its a major
pain to go hunt down the unencrypted db password(because its a long string
that's hard to memorize), log into the database, run a query, just to know
where to continue your debugging.

A little background: Most of the people who surf these logs aren't devs.
For us they are usually devs, but not cloudstack devs. Maybe UI guys or
some other consumer. It has happened several times where someone comes into
my office, points out an agent command, and says "any idea where that
went?". I pretty much have the numbers memorized by now, depending on the
zone, so I tell them which agent the command went to. Then they ask me
where I found that, and when I explain that this little "via: 85" means to
go log into the db and make an SQL query for vm_instance 85, I almost
always get some "gee that's useful, how was I ever supposed to know that"
response.

So I do agree that the correlations you mention would also be good, I
disagree with the idea that there is not much value in doing this.
On Aug 1, 2013 6:06 AM, "Koushik Das" <koushik.das@citrix.com> wrote:

>    This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/13072/
>
> The changes look fine but I don't see much value with this as the name can be easily
identified from the db. I feel the logs are used primarily for debugging issues. And one key
aspect that is missing currently is the correlation of logs. Currently when a API call is
made to MS, it traverses through various layers in the MS and finally hitting the resource
layer and then returns back with response. Logs gets generated from each layer but it is not
very intuitive to relate them to API call. If there is a unique id (some kind of uuid) that
gets appended to the logs for a specific API call then correlation becomes very easy.
>
> I feel adding these kind of correlation would be much more useful.
>
>
> - Koushik Das
>
> On July 30th, 2013, 5:37 p.m. UTC, Marcus Sorensen wrote:
>   Review request for cloudstack.
> By Marcus Sorensen.
>
> *Updated July 30, 2013, 5:37 p.m.*
>  *Bugs: * CLOUDSTACK-3872
>  *Repository: * cloudstack-git
> Description
>
> Print agent name when logging Commands sent to VM hosts. See bug description.  I'm not
super familiar with this code, so I'd like someone to look over it and verify it's the right
thing.
>
>   Testing
>
> Tested on KVM zone, need help testing others.
>
>   Diffs
>
>    - core/src/com/cloud/agent/transport/Request.java (b0fa4cc)
>    - server/src/com/cloud/agent/manager/AgentManagerImpl.java (b157838)
>
> View Diff <https://reviews.apache.org/r/13072/diff/>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message