mod_proxy_hcheck backport

classic Classic list List threaded Threaded
7 messages Options
Reply | Threaded
Open this post in threaded view
|

mod_proxy_hcheck backport

Jim Jagielski
Looking, begging, pleading for tests and votes :)
Reply | Threaded
Open this post in threaded view
|

Re: mod_proxy_hcheck backport

Daniel Ruggeri
On 2016-05-12 16:33, Jim Jagielski wrote:
> Looking, begging, pleading for tests and votes :)

Yup! You got mine.

Side question that popped into my mind as I was reviewing the code...
Would it make sense to have a "transitional" status for the time between
0 failures and when hcfails threshold is breached? I wonder because
there may be a reason to have multiple failures configured before
marking the backend out of service (plus a potentially long hcinterval
for whatever reason) and it'd be nice if an admin can be informed of a
"yellow" condition if they are futzing around elsewhere in the balancer
manager or are monitoring its output otherwise.

--
Daniel Ruggeri
Reply | Threaded
Open this post in threaded view
|

Re: mod_proxy_hcheck backport

Jim Jagielski

> On May 14, 2016, at 9:28 PM, [hidden email] wrote:
>
> On 2016-05-12 16:33, Jim Jagielski wrote:
>> Looking, begging, pleading for tests and votes :)
>
> Yup! You got mine.

THANKS! This feature seemed to cause a lot of buzz @ ApacheCon so
would be very very nice to have in 2.4.21.

> Side question that popped into my mind as I was reviewing the code... Would it make sense to have a "transitional" status for the time between 0 failures and when hcfails threshold is breached? I wonder because there may be a reason to have multiple failures configured before marking the backend out of service (plus a potentially long hcinterval for whatever reason) and it'd be nice if an admin can be informed of a "yellow" condition if they are futzing around elsewhere in the balancer manager or are monitoring its output otherwise.

Hmmm... The balancer-manager page does display the configured
values for 'hcpasses' and 'hcfails' as well as the current count
of passes/fails. Is that sufficient?
Reply | Threaded
Open this post in threaded view
|

Re: mod_proxy_hcheck backport

Daniel Ruggeri
On 5/16/2016 8:19 AM, Jim Jagielski wrote:
> THANKS! This feature seemed to cause a lot of buzz @ ApacheCon so
> would be

I believe I heard and/or used the term "sexy" at least once to describe
it ;-)


On 5/16/2016 8:19 AM, Jim Jagielski wrote:
> Hmmm... The balancer-manager page does display the configured
> values for 'hcpasses' and 'hcfails' as well as the current count
> of passes/fails. Is that sufficient?

Yeah - didn't think about that. It'd be fine for my purposes. It could
be a PITA if someone is monitoring for a specific string like
"transitive" or "fail", but it's probably not worth monkeying with.

--
Daniel Ruggeri

Reply | Threaded
Open this post in threaded view
|

Re: mod_proxy_hcheck backport

Jim Jagielski

> On May 16, 2016, at 3:37 PM, Daniel Ruggeri <[hidden email]> wrote:
>
> On 5/16/2016 8:19 AM, Jim Jagielski wrote:
>> THANKS! This feature seemed to cause a lot of buzz @ ApacheCon so
>> would be
>
> I believe I heard and/or used the term "sexy" at least once to describe
> it ;-)
>
>
> On 5/16/2016 8:19 AM, Jim Jagielski wrote:
>> Hmmm... The balancer-manager page does display the configured
>> values for 'hcpasses' and 'hcfails' as well as the current count
>> of passes/fails. Is that sufficient?
>
> Yeah - didn't think about that. It'd be fine for my purposes. It could
> be a PITA if someone is monitoring for a specific string like
> "transitive" or "fail", but it's probably not worth monkeying with.
>

Yeah... let's mull this over and see what makes the most sense.

Reply | Threaded
Open this post in threaded view
|

Re: mod_proxy_hcheck backport

Stefan Eissing
Jim,

how do you see the possibility of other proxy modules providing their own hc? HTTP/2 has this nice PING frame that is intended for exactly this.

Cheers,
 Stefan

PS. Btw. it's accepted for backport, left the actual work for you...

> Am 17.05.2016 um 12:41 schrieb Jim Jagielski <[hidden email]>:
>
>
>> On May 16, 2016, at 3:37 PM, Daniel Ruggeri <[hidden email]> wrote:
>>
>> On 5/16/2016 8:19 AM, Jim Jagielski wrote:
>>> THANKS! This feature seemed to cause a lot of buzz @ ApacheCon so
>>> would be
>>
>> I believe I heard and/or used the term "sexy" at least once to describe
>> it ;-)
>>
>>
>> On 5/16/2016 8:19 AM, Jim Jagielski wrote:
>>> Hmmm... The balancer-manager page does display the configured
>>> values for 'hcpasses' and 'hcfails' as well as the current count
>>> of passes/fails. Is that sufficient?
>>
>> Yeah - didn't think about that. It'd be fine for my purposes. It could
>> be a PITA if someone is monitoring for a specific string like
>> "transitive" or "fail", but it's probably not worth monkeying with.
>>
>
> Yeah... let's mull this over and see what makes the most sense.
>

Reply | Threaded
Open this post in threaded view
|

Re: mod_proxy_hcheck backport

Jim Jagielski-2
Seems like adding that directly to the module itself makes the most sense. Later versions
will allow for using Provider API to make it easier to add various checks,

On 2016-05-17 05:17, Stefan Eissing <[hidden email]> wrote:

> Jim,
>
> how do you see the possibility of other proxy modules providing their own hc? HTTP/2 has this nice PING frame that is intended for exactly this.
>
> Cheers,
>  Stefan
>
> PS. Btw. it's accepted for backport, left the actual work for you...
>
> > Am 17.05.2016 um 12:41 schrieb Jim Jagielski <[hidden email]>:
> >
> >
> >> On May 16, 2016, at 3:37 PM, Daniel Ruggeri <[hidden email]> wrote:
> >>
> >> On 5/16/2016 8:19 AM, Jim Jagielski wrote:
> >>> THANKS! This feature seemed to cause a lot of buzz @ ApacheCon so
> >>> would be
> >>
> >> I believe I heard and/or used the term "sexy" at least once to describe
> >> it ;-)
> >>
> >>
> >> On 5/16/2016 8:19 AM, Jim Jagielski wrote:
> >>> Hmmm... The balancer-manager page does display the configured
> >>> values for 'hcpasses' and 'hcfails' as well as the current count
> >>> of passes/fails. Is that sufficient?
> >>
> >> Yeah - didn't think about that. It'd be fine for my purposes. It could
> >> be a PITA if someone is monitoring for a specific string like
> >> "transitive" or "fail", but it's probably not worth monkeying with.
> >>
> >
> > Yeah... let's mull this over and see what makes the most sense.
> >
>
>