pcp
[Top] [All Lists]

Re: pcp updates: dynamic metrics, pmdapapi, containers, qa

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: pcp updates: dynamic metrics, pmdapapi, containers, qa
From: fche@xxxxxxxxxx (Frank Ch. Eigler)
Date: Thu, 11 Dec 2014 09:24:35 -0500
Cc: pcp <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <1300433870.14907994.1418289576168.JavaMail.zimbra@xxxxxxxxxx> (Nathan Scott's message of "Thu, 11 Dec 2014 04:19:36 -0500 (EST)")
References: <494364595.14790394.1418280789809.JavaMail.zimbra@xxxxxxxxxx> <1300433870.14907994.1418289576168.JavaMail.zimbra@xxxxxxxxxx>
User-agent: Gnus/5.1008 (Gnus v5.10.8) Emacs/21.4 (gnu/linux)
> [...]
> Nathan Scott (1):
>       libpcp/pmcd: protocol-level support for containers
> [...]

Can you explain further your intended design for this, so we can get a
chance to review before it gets cast in stone?

What is the exact interpretation of "container name" that is to be
passed?  A path component under the global /sys/fs/cgroup?  Same for
all PMDAs?  Which cgroup manager-type (devices? cpu,*)?  Does that
work with nested containers a la lxc?  What about systemd?  nested
systemd a la atomic/rocket?  libvirt?  Is a process running within a
container namespace required (& able) to know its global container
name in order to get stats for its own cgroup?

(The user interface to the various container tools usuall give a
friendly name, but for pcp purposes, we may well need to canonicalize
them to lower level identifiers for a clear operational purposes.)

- FChE

<Prev in Thread] Current Thread [Next in Thread>