How to iterate VMOD objects on VCL_EVENT_COLD?
dridi at varni.sh
Tue Dec 15 20:34:49 CET 2015
> Each object implements a connection pool maintaining multiple opened connections in its internal state. I would like to be able to release those connections when the VCL using the VMOD is not active:
> - I think that's not doable in 4.0. The only alternative is waiting for the VCL to be discarded and then close connections when the __fini() callback is executed for each instance.
In 4.0 you don't go through the __fini functions until the VCL is
actually discarded. There's no way to reliably release resources
when the VCL is not active, only to release them when the VCL is
> - In 4.1 we have the VMOD event function. The VMOD could listen for VCL_EVENT_COLD events and then close established connections. In order to do that the full list of instances would be required. Is it possible to get access to that list somehow? Should it be manually implemented by the VMOD? Is there any plans to propagate VCL temperature events also to object instantiated by a VMOD?
This has been discussed and events are not propagated to VMOD objects
so you need to keep track of them and do the event propagation
yourself. From what I remember, there are no plans to send events to
objects (they still follow an init/fini style) or other data
structures such as private pointers (PRIV_VCL). Be aware that VCL
temperature is racy in 4.1.0 and that some of the issues have been
fixed in master and others are awaiting a review.
During the review, we found new holes in the design that need to be
discussed: what do we do with VCLs that fail to load or warm up when
the child is started? What do we do if the active VCL fails?
 that I know of
More information about the varnish-misc