[Pacemaker] [Problem] The "dampen" parameter of the attrd_updater command is ignored, and an attribute is updated.

[renayama19661014 at ybb.ne.jp]

Hi Andrew,

> Perhaps try:
> 
> diff --git a/attrd/commands.c b/attrd/commands.c
> index 7f1b4b0..7342e23 100644
> --- a/attrd/commands.c
> +++ b/attrd/commands.c
> @@ -464,6 +464,15 @@ attrd_peer_update(crm_node_t *peer, xmlNode *xml, bool filter)
>  
>      a->changed |= changed;
>  
> +    if(changed) {
> +        if(a->timer) {
> +            crm_trace("Delayed write out (%dms) for %s", a->timeout_ms, a->id);
> +            mainloop_timer_start(a->timer);
> +        } else {
> +            write_or_elect_attribute(a);
> +        }
> +    }
> +
>      /* this only involves cluster nodes. */
>      if(v->nodeid == 0 && (v->is_remote == FALSE)) {
>          if(crm_element_value_int(xml, F_ATTRD_HOST_ID, (int*)&v->nodeid) == 0) {
> @@ -476,15 +485,6 @@ attrd_peer_update(crm_node_t *peer, xmlNode *xml, bool filter)
>              }
>          }
>      }
> -
> -    if(changed) {
> -        if(a->timer) {
> -            crm_trace("Delayed write out (%dms) for %s", a->timeout_ms, a->id);
> -            mainloop_timer_start(a->timer);
> -        } else {
> -            write_or_elect_attribute(a);
> -        }
> -    }
>  }
>  
>  void

Okay!
I confirm movement.

Many Thanks!
Hideo Yamauchi.

--- On Wed, 2014/5/28, Andrew Beekhof <andrew at beekhof.net> wrote:

> 
> On 28 May 2014, at 4:10 pm, Andrew Beekhof <andrew at beekhof.net> wrote:
> 
> > 
> > On 28 May 2014, at 3:04 pm, renayama19661014 at ybb.ne.jp wrote:
> > 
> >> Hi Andrew,
> >> 
> >>>> I'd expect that block to hit this clause though:
> >>>> 
> >>>>     } else if(mainloop_timer_running(a->timer)) {
> >>>>        crm_info("Write out of '%s' delayed: timer is running", a->id);
> >>>>        return;
> >>> 
> >>> Which point of the source code does the suggested code mentioned above revise?
> >>> (Which line of the source code is it?)
> >> 
> >> Is it the next cord that you pointed?
> > 
> > right
> > 
> >> 
> >> void
> >> write_attribute(attribute_t *a)
> >> {
> >>   int updates = 0;
> >> (snip)
> >>   } else if(mainloop_timer_running(a->timer)) {
> >>       crm_info("Write out of '%s' delayed: timer is running", a->id);
> >>       return;
> >>   }
> >> (snip)
> >> 
> >> At the time of phenomenon of the problem, the timer does not yet block it by this processing because it does not start.
> > 
> > Thats the curious part
> 
> Perhaps try:
> 
> diff --git a/attrd/commands.c b/attrd/commands.c
> index 7f1b4b0..7342e23 100644
> --- a/attrd/commands.c
> +++ b/attrd/commands.c
> @@ -464,6 +464,15 @@ attrd_peer_update(crm_node_t *peer, xmlNode *xml, bool filter)
>  
>      a->changed |= changed;
>  
> +    if(changed) {
> +        if(a->timer) {
> +            crm_trace("Delayed write out (%dms) for %s", a->timeout_ms, a->id);
> +            mainloop_timer_start(a->timer);
> +        } else {
> +            write_or_elect_attribute(a);
> +        }
> +    }
> +
>      /* this only involves cluster nodes. */
>      if(v->nodeid == 0 && (v->is_remote == FALSE)) {
>          if(crm_element_value_int(xml, F_ATTRD_HOST_ID, (int*)&v->nodeid) == 0) {
> @@ -476,15 +485,6 @@ attrd_peer_update(crm_node_t *peer, xmlNode *xml, bool filter)
>              }
>          }
>      }
> -
> -    if(changed) {
> -        if(a->timer) {
> -            crm_trace("Delayed write out (%dms) for %s", a->timeout_ms, a->id);
> -            mainloop_timer_start(a->timer);
> -        } else {
> -            write_or_elect_attribute(a);
> -        }
> -    }
>  }
>  
>  void
> 
> 
> 
>