[ProgClub programming] Configuring network with systemd-networkd

Jedd Rashbrooke jedd.rashbrooke at gmail.com
Thu Mar 21 08:37:55 AEDT 2019


On Wed, 20 Mar 2019 at 21:56, John Elliot V | ProgClub <jj5 at progclub.org>
wrote:

> On 20/3/19 9:37 pm, Jedd Rashbrooke wrote:
> >  Yes, you probably should be running monitoring and log management
> systems.
>
> I'm nearly ready for that (give me a few weeks). What sort of specs will
> I be looking at to provision those services? Do I run Graylog and Zabbix
> servers on different hosts? How much (virtual) CPU/RAM/Disk should I
> provision?


 Goodness.

 I'm happy to give pointers to things you should go read about and think
about and run up some demo instances to play with and evaluate.  These
suggestions have come about from things like 'I've run out of space on my
cloud server and didn't notice' and 'my hard disk lights are flashing like
wild but I can't login to that box' (etc).

 As noted, look at Prometheus first - it's more modern, probably better
lends itself to custom instrumentation.

 For logging, look at Loki first - it's made by the Grafana guys, and
Grafana is *almost definitely* what you want to use for visualisation (of
logs, events, and metrics). It's currently focused on k8s (and similar) but
I expect its feature set to grow quickly.

 You may even want to look at Elasticsearch.  AWS has just forked it
because Elastic.co's licensing is a bit weird - and while AWS isn't all
rainbows and unicorns, this move may make a lot of people's lives easier /
better.  Anyhoo, Elasticsearch also has monitoring facilities.  Research
'metricbeat' -- I haven't played with that component.

 Specs:  it depends.  What granularity data do you want to obtain, how long
do you want to retain it, etc.  Normally I'd say go with the author's
recommendation, or slightly under, as it's very easy to bump up cores,
memory, file system sizes on VM's ... but I gather this is still somewhat
of a challenge in your environment, so it's going to be trickier for you.
Elasticsearch's three components have different JRE version restrictions /
requirements - this is frustrating.  It's not especially CPU intensive
until you start doing whacky stuff with it.  Graylog uses pretty much every
buzzword component (including Mongo) they could find, but it's not CPU
intensive.  Zabbix is pretty light, especially on a small fleet like
yours.  PostgreSQL (say with TimescaleDB) is also not very demanding.
 Grafana runs very light - but it's not doing much work.   I run elastic,
logstash, kibana, netflow (elastiflow), zabbix, postgresql, prometheus,
netdata, etc on a single box (6GB, 2 core, separate LVM-backed file systems
for postgresql, elasticsearch, and prometheus) and it works well - I shared
some screenshots recently to this list.

 j.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.progclub.org/pipermail/programming/attachments/20190321/a05f89ad/attachment-0001.html>


More information about the programming mailing list