trafficserver-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sudheer Vinukonda <>
Subject Re: Host status metric change in ATS 8.1
Date Thu, 16 Apr 2020 22:11:05 GMT
 Ah apologies, John - I totally misread it. For a stat, it sounds fine to me!
    On Thursday, April 16, 2020, 03:06:30 PM PDT, John Rushford <>
 typo in my previous, it's a persistent stat to records.snap
On Thu, Apr 16, 2020 at 4:05 PM John Rushford <> wrote:

Hi Sudheer, It's not a records.config entry it's a stat that is persisted to records.config. 
HostStatus loads and deserializes the text into a struct kept in a hash table on startup.
On Thu, Apr 16, 2020 at 3:56 PM Sudheer Vinukonda <> wrote:

 We don't use this, so I'm not opposed to the proposed change particularly if it means adding
more capability :)
However, I do have one comment/concern - can we explore using a different config format to
be less error prone/more readable? It seems a bit too complex and easily error-prone to have
to encode the entire config into a records.config string. Perhaps, a separate config file
or part of one of the existing config files and yaml or json format might be easier on the
Just my 2c!

    On Thursday, April 16, 2020, 02:08:08 PM PDT, John Rushford <>
With ATS 8.1 I would like find out if a change in the host status metric from ATS 8.0 can
be made.  Host status is new with 8.0 and I suspect not many are using it yet so, we would
like to make a change to the metric in 8.1.

The host status metrics are used in ATS by parent selection and traffic_ctl to report and
determine if a parent is available for use.  A user may use the 'traffic_ctl host' command
to mark a parent as up or down for use by parent selection and that hosts status is then kept
in a metric.  For example in ATS 8.0 a metric could be:
host_status. 1.   

the metric prefix "host_status" followed by a host name or IP and it's current state: 0 =
unavailable 1 = available.
In ATS 8.1 more features would be added to support health checking, downtime, and more. 
In order to serialize the state I've used this new metric format:
proxy.process.host_status. HOST_STATUS_UP,ACTIVE:UP:0:0,LOCAL:UP:0:0,MANUAL:UP:0:0,SELF_DETECT:UP:0

This new format serializes the state of a host with reason codes and the times that they were
marked down and the state is loaded across restarts.  This format and the feature change
is used in ATS 9.0 and was cherry-picked into the new 8.1 branch.  Details regarding use
of this feature are here:
I would like to see if the community is agreeable to this metric and feature change from 8.0
to 8.1.

John Rushford

John Rushford  
View raw message