Hi John, Many thanks for the clarification.
Cheers, Nick Sent from my iPhone > On Feb 20, 2023, at 7:20 PM, John Rushford <jjrushf...@gmail.com> wrote: > > Yeah, there is no mechanism that would wait within a transaction for a > parent to become available again. Within a transaction when a parent is > marked down, ATS will try to select another available parent to complete the > transaction or go direct to the origin if the config allows it. The closest > thing to what you’re asking for is the > proxy.config.http.parent_proxy.retry_time. A down parent will be retried > once that time has expired and will be marked up if the next transaction > succeeds > > John. > >> On Feb 20, 2023, at 4:18 PM, Nick Dunkin <nick.dun...@vecima.com.INVALID> >> wrote: >> >> Hi John, >> >> Ok, that’s what I thought. >> >> It’s the 502s that I’m trying to avoid in this instance. >> >> As far as I can tell, the retry mechanisms on a parent do not support any >> kind of “back off” mechanism, so you can have 100+ retries, but with an >> offline parent those “connection attempts” are exhausted in milliseconds. >> >> I would like to be able to allow a parent to be temporarily “unavailable” >> for anything up to ~2 seconds, and make the downstream client wait for the >> parent resource to be available again. It’s a rare edge case, but it’s a >> real one. >> >> The context here Is ABR video streaming, where a client has enough buffer to >> very occasionally wait a couple of extra seconds for its content, but an >> HTTP 502 would cause it to abort its stream. >> >> Thanks for your assistance, >> >> Nick >> >> From: John Rushford <jjrushf...@gmail.com> >> Date: Monday, February 20, 2023 at 6:03 PM >> To: dev@trafficserver.apache.org <dev@trafficserver.apache.org> >> Cc: Nick Dunkin <nick.dun...@vecima.com.invalid> >> Subject: Re: Configuring sleep times between parent connection retires? >> No a parent is not retried within the same transaction. If the configured >> number of retry counts is not exceeded, another parent would be chosen if >> there are other available parents. If there are no other available parents >> and the configs allow it, the transaction may go direct to the origin, >> go_direct in parent.config. If there are no other available parents and >> go_direct == false, the transaction fails and the client sees a 502 >> >> Sent from my iPhone >> >>>> On Feb 20, 2023, at 3:18 PM, Nick Dunkin <nick.dun...@vecima.com.invalid> >>>> wrote: >>> >>> Hi John, >>> >>> Thank you for the reply. >>> >>> Are you suggesting that I could have a parent marked down, with a very >>> short retry_time (~1 second), and it would be retried within the same >>> transaction? >>> >>> I would have thought that at the point it’s marked down, the transaction >>> would end? >>> >>> Thanks >>> >>> Nick >>> >>> From: John J. Rushford <jjrushf...@gmail.com> >>> Date: Monday, February 20, 2023 at 3:21 PM >>> To: dev@trafficserver.apache.org <dev@trafficserver.apache.org>, Nick >>> Dunkin <nick.dun...@vecima.com.INVALID> >>> Subject: Re: Configuring sleep times between parent connection retires? >>> Nick, >>> >>> See the Parent Proxy Configurations, >>> https://docs.trafficserver.apache.org/admin-guide/files/records.config.en.html#parent-proxy-configuration >>> >>> You may need to do some tweaking. There is a sliding window, >>> proxy.config.http.parent_proxy.retry_time. When the failure threshold >>> ,proxy.config.http.parent_proxy.fail_threshold¶, on a parent is reached >>> the parent is marked down and will not be used until the retry window >>> has elapsed. when the window has elapsed, a failed parent will be >>> retried but, if the retry fails, no further attempts will be made to >>> that parent until the retry window has elapsed again. If the retry >>> succeeds, the parent is marked up. >>> >>> thanks >>> >>> John >>> >>>> On 2/20/23 12:02, Nick Dunkin wrote: >>>> >>>> Hi, >>>> >>>> If a parent cache is offline for a brief period of time (say < 3 >>>> seconds) I would like to know if there is way to configure a “sleep” >>>> between connection retries on a particular parent? >>>> >>>> The ability to configure the “connection attempts” is not useful here, >>>> as when the parent is completely offline the algorithm can run through >>>> many hundreds of connection attempts in milliseconds (confirmed from >>>> logs). >>>> >>>> The use case here is that, for us, it is preferable to make the client >>>> occasionally wait 1-3 seconds, than to return them a HTTP 502. >>>> >>>> I’ve started looking at the code, but I’d like to confirm that this is >>>> not already supported somehow. >>>> >>>> Many thanks >>>> >>>> Nick >>>> >>>> *Nick Dunkin* >>>> >>>> Director, Software Architecture >>>> >>>> Manager – Architecture and New Product Introduction >>>> >>>> *o: **+1 678.258.4071* >>>> >>>> *e:*nick.dun...@vecima.com <mailto:nick.dun...@vecima.com> >>>> >>>> cidimage001.png@01D6CC8C.6FC5A580 >>>> >
