[PROPOSAL] Kill traffic_cop for ATS v8.0.0

[Leif Hedstrom]

Hi all,

I’d like to propose that we kill off traffic_cop for ATS v8.0.0. The consensus is that there are better system level tools to manage health checks out of band like it does. Doing this would require

* Removing the synthetic health check thread from traffic_manager
* Removing the backdoor in traffic_server
* Modifying the init / systemd scripts to start traffic_manager directly


Susan has volunteered to do this work, as long as there are no strong arguments for keeping traffic_cop.

Thanks,

— Leif

[Leif Hedstrom]

Post by Leif Hedstrom
Hi all,
I’d like to propose that we kill off traffic_cop for ATS v8.0.0. The consensus is that there are better system level tools to manage health checks out of band like it does.

Do you mean a local service manager, or a CDN load balancer check? Can you give an example of a something that will manage the health checks?

A load balancer, haproxy, nginx, custom tools that your dev team might have developed. I think the killing of traffic_server that traffic_cop does is a flawed design, which is why we added the option to turn that off, which is how we run it in prod. If you truly want to do such a think, writing a shell script that does a “curl” on some URL and then “kill -9” would not be difficult. :-).

— Leif

Post by Leif Hedstrom
Doing this would require
* Removing the synthetic health check thread from traffic_manager
* Removing the backdoor in traffic_server
* Modifying the init / systemd scripts to start traffic_manager directly
Susan has volunteered to do this work, as long as there are no strong arguments for keeping traffic_cop.
Thanks,
— Leif

[Phil Sorber]

Post by Leif Hedstrom

Post by Leif Hedstrom
Hi all,
I’d like to propose that we kill off traffic_cop for ATS v8.0.0. The

consensus is that there are better system level tools to manage health
checks out of band like it does.
Do you mean a local service manager, or a CDN load balancer check? Can you
give an example of a something that will manage the health checks?

I would think both of those, plus something like Nagios or Prometheus.
Perhaps someone could write ats_exporter that would export metrics to
Prometheus, or even modify stats_over_http to do it.

Post by Leif Hedstrom

Post by Leif Hedstrom
Doing this would require
* Removing the synthetic health check thread from traffic_manager
* Removing the backdoor in traffic_server
* Modifying the init / systemd scripts to start traffic_manager

directly

Post by Leif Hedstrom
Susan has volunteered to do this work, as long as there are no strong

arguments for keeping traffic_cop.

Post by Leif Hedstrom
Thanks,
— Leif