[bind10-dev] IXFR to AXFR fallback (Re: b10-xfrin fails to receive IXFR while b10-xfrout is processing AXFR)

[JINMEI Tatuya / 神明達哉]

At Wed, 03 Apr 2013 23:39:29 -0700,
JINMEI Tatuya <jinmei at isc.org> wrote:

> > It seems that b10-xfrout locks database while putting AXFR; thus
> > b10-xfrin fails to acquire lock of database to update zone.
[...]
> Falling back to AXFR doesn't seem to be the reasonable behavior in
> this case, though.

While working on http://bind10.isc.org/ticket/2911 I changed my mind.
Disabling fallback-to-AXFR may not always be a good idea as long as
there's a possibility of success after fallback.  Although it's quite
unlikely to help in the DB lock case (retry itself may help, whether
AXFR or IXFR), it may not always be that obvious.  The overhead of
fallback should be normally small except for some special cases like
very large zones, so it should be safer to do fallback by default
regardless of the type of error.  I've checked BIND 9's
implementation, and it also performs fallback almost all error cases.

So, instead of doing #2911 as originally proposed, I propose to add a
configuration knob to disable the fallback:
  Xfrin/zones[n]/disable_axfr_fallback
This is set to false by default, and if set to true fallback from IXFR
to AXFR will be suppressed for that zone.

Comments?

---
JINMEI, Tatuya