I am testing IXFR for servers I did not install nor have easy access to. version.server. says it is 3.2.6. I know there are IXFR changes since then per the NEWS file and from git log. I don't see same behavior on my different version different systems but they are also configured differently. The knot.conf zones are not configured with "zonefile-load: difference" and the response effectively has the entire zone as if was AXFR and not the changes. If I pass the IXFR SOA SERIAL to latest it has no changes (answer has has the SOA only with same serial). I used dnspython to output the response from doing IXFR queries (IXFR question with SOA authoritative set with the serial in the query). I noticed the output abruptly stops when "dig" doesn't stop. So I used tcpdump many times to compare knot, named, and my other knot. I found an odd behavior in this knot 3.2.6 response which dig ignores and my dnspython fails. After the expect record it has 1) OPT record with the requestors pay load size (class 1232) and edns rcode and flags (all zeros ttl), then 00 rdlength and 00 rdata field. 2) then 28 bytes I don't understand such as: 40 11be dc80 0000 0101 fa00 0000 01 or 40 20be dc80 0000 0102 0300 0000 01 or 40 0fe1 6a81 0000 0102 0500 0000 00 or 12 8de1 6a81 0000 0100 9200 0000 00 3) then an IXFR record following other labels ... 0363 6f6d 00 three characters "com" and end of domain 00 fb IXFR record type 00 01 INternet class and then ends there, with NO ttl, rdlength, nor rdata. 4) followed by next label length, label ... etc with rrtype, class, ttl, rdlength, rdata and so on. This odd OPT, bytes I don't know, partial/broken IXFR record, may be repeated a few times. I assume these were interspersed where IXFR's SOA records should be. I couldn't find an RFC that suggested using interspersed OPT nor IXFR records. I find it odd that OPT record is in my ANSWER section. I find it odd that the IXFR record is incomplete. And I don't know what the other bytes are in-between. This recognizable to anyone? The IXFR works fine as seen with dig or when I use named as my secondary but I assume the named is ignoring the junk parts too. --