[mod_cluster-issues] [JBoss JIRA] (MODCLUSTER-645) Different handling of max-attempts parameter on Undertow and httpd implementation

[Radoslav Husar (JIRA)]

    [ https://issues.jboss.org/browse/MODCLUSTER-645?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13549727#comment-13549727 ] 

Radoslav Husar commented on MODCLUSTER-645:
-------------------------------------------

This is really suboptimal, because one needs to adjust the value in mod_cluster depending on which LB it is consulting. 

The mod_cluster native implementation delegates the value to mod_proxy, https://httpd.apache.org/docs/2.4/mod/mod_proxy.html#proxypass, which is defined as:
* Maximum number of failover attempts before giving up.
* default: One less than the number of workers, or 1 with a single worker.
while Undertow always defined it as total number of *attempts* as opposed to *failover attempts*.

I am not ever sure which one is 'better'. If it is failover attempts, they should have called it max-failover-attempts. 

> Different handling of max-attempts parameter on Undertow and httpd implementation
> ---------------------------------------------------------------------------------
>
>                 Key: MODCLUSTER-645
>                 URL: https://issues.jboss.org/browse/MODCLUSTER-645
>             Project: mod_cluster
>          Issue Type: Bug
>    Affects Versions: 1.3.5.Final
>            Reporter: Jan Kašík
>            Assignee: Radoslav Husar
>
> The behavior of balancer is different on httpd in comparsion with Undertow implementation when max-attempts parameter is set.
> * *Undertow*: max-attempts is total count of requests made by balancer in failover scenario. When max-attemps is set to 1 and target worker fails after first attempt made by balancer, no further attempts are made.
> * *httpd*: max-attempts is count of retries made by balancer after first attempt fails. When max-attempts is set to 1 and target worker fails after first attempt made by balancer, second worker is chosen and request is sent towards it.
> This doesn't seem right and I suggest to unify behavior across implementations.



--
This message was sent by Atlassian JIRA
(v7.5.0#75005)