www-infrastructure-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Christofer Dutz <christofer.d...@c-ware.de>
Subject AW: Tool proposal for helping run and monitor the ASF Infra Services
Date Sat, 27 Aug 2016 17:29:36 GMT
Hi Chris,

Ok so I started the Agent ... the Windows build agent is rather unexciting as it only has
a hand full of java processes and a MsSQL Server with all network shut down. When encountering
software, that is explicitly instrumented such as an Apache Webserver, Tomcat, the introspection
will provide more insight.

Eventually I could whip up a puppet script which you could use to setup the Instana Agent
on a linux machine. Then you should be able to setup an agent on one of the more interesting
machines. Are there any test VMs I could test such a script on?

I just sent you (Chris L.) an invitation to the Instana Web Console, so feel free to sign
up. If anyone else is interested, I'll be glad to invite him/her too :-)

The "Datacenter A / Rack 42" is just a dummy as I don't have any clue about what datacenter
the machine/vm is running on.


Von: Chris Lambertus <cml@apache.org>
Gesendet: Samstag, 27. August 2016 08:20:15
An: infrastructure-dev@apache.org
Cc: Christofer Dutz
Betreff: Re: Tool proposal for helping run and monitor the ASF Infra Services


We’re interested, but we don’t really have the bandwidth to set up evaluations for tools
like this right now. If you truly believe it’s worthwhile, then go ahead and fire up the
agent on hudson-win and let’s see what we get out of it. If it doesn’t require any elevated
permissions, I’m happy to see a live demo.

On Aug 23, 2016, at 11:46 PM, Christofer Dutz <christofer.dutz@c-ware.de<mailto:christofer.dutz@c-ware.de>>

Hi all,

How about trying it out for one chain of services? I bet that would explain a lot more than
anything else? As I mentioned, the SAAS account is already setup and I can create some user
accounts for you guys.

On the windows1 machine (hudson-win.apache.org<http://hudson-win.apache.org/> in directory
"f:\instana-agent") I have the agent installed (but inactive) so you could have a look at
the agent itself.

We could fire up the agent and start a Jenkins build and it will definitely show us communication
with the jenkins machine, repository.apache.org<http://repository.apache.org/> and the
proxy (Well probably with the proxy). So if we setup the agent on windows, the proxy and repository.apache.org<http://repository.apache.org/>
I guess this should give a first impression. And if you don't like or want it in the end,
just stop and delete the agent and no harm was done. What do you think?

I would be glad to assist you in this.


Von: Mirko Novakovic <mirko.novakovic@codecentric.de<mailto:mirko.novakovic@codecentric.de>>
Gesendet: Dienstag, 23. August 2016 16:04:52
An: infrastructure-dev@apache.org<mailto:infrastructure-dev@apache.org>
Betreff: Re: AW: Tool proposal for helping run and monitor the ASF Infra Services

Hi Daniel,

I am the Instana CEO and happy to answer your questions. First of all I would like to point
out that we are an APM company/product, so more in the space of NewRelic than DataDog. But
as we are monitoring the application as a whole (and model a dynamic dependency graph for
it) we also monitor the infrastructure. And have out of the box dashboards etc but the core
is our distributed tracing and service quality management - all automatic and without configuration.

Regarding the questions:

- I can guarantee that there is no time limit on this offering for Apache - the only thing
we should discuss is the number of components you would monitor, as it is SAAS there should
be some maximum defined so that the costs for us do not explode. If you want to have Instana
On Premise we wouldn’t need that limit.

- Our whole system is streaming based, so we stream the data form the agent to the server
and from the server the data is pushed to the UI. On the backend we build a graph and apply
a knowledge base on each component to find changes and issues in realtime. Based on service
KPIs (Four Golden Signals by Google) we identify if the changes/issues are having impact on
the service quality - if yes we report. The algorithms are based on machine learning, so we
see things like sudden drops, slow responses, high error rates, etc etc

- Don’t know Circonus in details, so I cannot really compare it. Instana for sure is not
time consuming as it was designed for auto discovery and zero configuration.

- We have different types of integration using API and SDK. For alarms we have out of the
box integrations for PagerDuty, OpsGenie and Slack, but also provide a generic Webhook to
integrated with any other system that has an API.

- The agent is Apache Karaf based, can update itself using a private or public sensor (plugin)
repository that is based on Apache Maven technology. We provide a set of sensors for approx
40 technologies right now but enable the users to extend this with own sensors.

We can give you more insights or a demo at any time.

Best regards

On 2016-08-19 09:16 (+0200), Daniel Gruno <h...@apache.org<http://apache.org>>
On 08/19/2016 08:20 AM, Christofer Dutz wrote:>
Hi Chris,>

I knew that someone asked exactly the "how does it compare to datadog" question somewhere.
Here's the link to that mail thread https://news.ycombinator.com/item?id=12147219>

And I can confirm the shortcomings of the time series approach, cause in jenkins, I'd say
about 70% of recent failures of flex builds were due to timeouts when uploading Maven artifacts
to nexus. The current solution doesn't seem to detect that. Not only that I couldn't see any
hipchat notifications. The infra guys always had to start looking for the real reason of the
timeouts as nexus wasn't having any problems at all.>

I'd say the real reason we weren't being notified is because httpd>
wasn't telling us it was jammed, so it would have made little difference>
whether we used product X or Y to monitor it, when it wasn't sending out>
data that could lead us to what the problem was. It wasn't a question of>
granularity, the data just wasn't enabled for any agent to see.>

I would imagine this would be something _instead_ of datadog, which>
leads me to some questions:>

- For how long into the future could we have a guarantee that this isn't>
gonna cost us $$$/year or whatever the price would end up being with a>
non-free version?>

- What is understood by real time checks here? and what exactly is checked?>

- How does this relate to more advanced monitoring systems like>
Circonus? As you may know, we had to drop that as it proved to be rather>
time consuming.>

- What sort of integrations does this system have? How are alerts>

I'd also be interested in learning about plugins and how to customize>
the agents.>

With regards,>

And I really live the feature of tracking down the response time for one service back to other
servers to find it the real reason for a system being slow (have a look in the presentation
for this. There's a great slide on this)>


Von meinem Samsung Galaxy Smartphone gesendet.>

-------- Ursprngliche Nachricht -------->
Von: Chris Lambertus <cm...@apache.org<http://apache.org>>>
Datum: 19.08.16 07:03 (GMT+01:00)>
An: infrastructure-dev@apache.org<mailto:infrastructure-dev@apache.org>, Christofer
Dutz <ch...@c-ware.de<http://c-ware.de>>>
Cc: mirko.novakovic@codecentric.de<mailto:mirko.novakovic@codecentric.de>>
Betreff: Re: Tool proposal for helping run and monitor the ASF Infra Services>

Hiya Chris,>

Thanks for the info and the legwork on this. We currently use DataDog, which is very similar
to what Instana appears to provide an agent-based monitoring solution that gives us that kind
of look into our infra. We also have a number of internal tools that report on various goings-on
as well. You might see some of this in #asfinfra on hipchat from SNMP2HipChat. DataDog also
reports various problems there, as does our monitoring via PingMyBox.>

Since youre not root@, you may not see some of the stuff that we see, but I think by and large,
the majority of the monitors do direct to #asfinfra. Have you noticed gaps in the monitoring?
Since we moved to DataDog, weve been quite happy with the resolution and metrics weve been
able to get. Its been on my back burner for awhile to expose some of our DD dashboards as
public, but for right now its somewhat limited access. In the interests of transparency (but
not at the expense of security,) Id be happy to work with you to expose more of this, and
Im happy to address any questions or concerns about shortcomings in our monitoring.>

Many thanks to Instana for offering the ASF free services! Id definitely like to hear more
about what they might be able to offer on top of what we already get from DataDog. Ill take
a look at the info you sent out. Please feel free to follow up with me directly, either via
email or hipchat.>


On Aug 18, 2016, at 2:13 AM, Christofer Dutz <ch...@c-ware.de<http://c-ware.de>>


I have been on the Infra Hipchat for a few weeks now while trying to migrate the Flex project
to Maven and back to the ASF Infra build system. Thanks for your support in this and even
more thanks for the trust in granting me access and Admin rights on the windows1 build agent.>

In the chat I observed the daily work of you guys, having to maintain quite a zoo of all sorts
of different systems on different platforms. Some problems you were having seem quite easy
to track down ... if the hard disk is full, you clean up. But not all problems are that easy
to track down. Thinking of the problems with repository.apache.org<http://repository.apache.org>
... here the cause was the proxy being flooded with connections (I think this was the case)
... regular restarts of this helped temporarily, but I don't think that helps on the long
term as no one had an idea why those connections were hanging there in the first place.>

A few years ago the company I work for - codecentric - have founded a company called Instana.
They are developing an agent based system for monitoring IT infrastructure. In contrast to
most established solutions, they use machine learning strategies to analyze the root cause
for problems. While you can probably achieve similar results with normal tools, the problem
is that you need a very detailed domain knowledge to do so and in a regularly changing environment
you need to continuously keep adjusting your metrics. Instana does this automatically. I think
you can imagine how tricky it is to follow the root cause for bad response times through a
network of interconnected services.>

Investing almost all of my free time (and a lot of my paid time) for Apache, noticing a lot
of the problems you have to deal with every day, I asked Instana if they would be willing
to provide their service to the ASF for free and they agreed and immediately setup a dedicated

I wanted to try the thing out as I would prefer to grab a few beers with you at ApcheCon in
Cevillia and not get punched in the face for recommending something bad ;-) ... so I tried
this on my private Server playground. I unpacked and started the agent and the host appeared
on the web console and reported the problems it was having (ones I didn't even know about)
as well as other systems it communicates with ... as soon as I added agents on these machines
the analytics started doing their work across system and I built up a map view of my services
and their correlation. So it's really a system that needs almost no configuration at all :-)>

I uploaded the internal product presentation here: https://public.centerdevice.de/1a9dc4ed-515e-482e-9fd6-6d60a5562598
(please don't share this outside of the ASF)>

Please use the password: 4p4cheR0cks (I'll remove that document in about two weeks)>

By the way ... the screenshots in the presentation are real ... I was amazed of seeing a 3D
web UI in production for the first time ;-)>

So if there is any interest in this offer, I would be more than happy to provide credentials
to you and assist you in getting started, so you could easily try it out. The guys at Instana
would also be delighted to give you guys an online demo and answer any questions you might
be having. Feel free to conatact Mirco directly for this: mirko.novakovic@codecentric.de<mailto:mirko.novakovic@codecentric.de>>


Mirko Novakovic | Vorstand

codecentric AG | Merscheider Straße 1 | 42699 Solingen | Deutschland
tel: +49 (0) 212.23362811 | fax: +49 (0) 212.23362879 | mobil: +49 (0) 163.6681500
| blog.codecentric.de<http://blog.codecentric.de/> | www.meettheexperts.de<http://www.meettheexperts.de/><http://www.meettheexperts.de<http://www.meettheexperts.de/>>
| www.more4fi.de<http://www.more4fi.de/><http://www.more4fi.de<http://www.more4fi.de/>>

Sitz der Gesellschaft: Solingen | HRB 25917 | Amtsgericht Wuppertal
Vorstand: Michael Hochgürtel . Mirko Novakovic . Rainer Vehns
Aufsichtsrat: Patric Fedlmeier (Vorsitzender) . Klaus Jäger . Jürgen Schütz

Diese E-Mail einschließlich evtl. beigefügter Dateien enthält vertrauliche und/oder rechtlich
geschützte Informationen. Wenn Sie nicht der richtige Adressat sind oder diese E-Mail irrtümlich
erhalten haben, informieren Sie bitte sofort den Absender und löschen Sie diese E-Mail und
evtl. beigefügter Dateien umgehend. Das unerlaubte Kopieren, Nutzen oder Öffnen evtl. beigefügter
Dateien sowie die unbefugte Weitergabe dieser E-Mail ist nicht gestattet.

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