...in case the zone had been XFRed before, the master serial
was not stored and after local signing, it was doomed

...in case of concurrency, it might lead to server crash on startup
this way, catalogDB is only opened from main thread in advance

Libor Peltan authored 3 years ago and Daniel Salzman committed 3 years ago

previously, adding RRSIG to the signed node confused
iteration over RRSets, leading to duplicit RRSIGs
of some NSEC(3)s

...also checking DS-TTL delay after submission

Returning KNOT_ENOTSUP makes more sense than returning KNOT_EMALF in such situations.

The test no longer works, as concurrent backups aren't possible in
Knot any more.

In case of repated server starts (enforced by occupied IP ports), an AXFR may happen
before the testing actually starts. Lower refresh value allows another AXFR of already
expired zone before the test time outs.

This change prevents unnecessary test failures under Valgrind on our new (and faster) test server.

Libor Peltan authored 3 years ago and Daniel Salzman committed 3 years ago

...is enough to take DNSKEY TTL + prop-delay

Libor Peltan authored 3 years ago and Daniel Salzman committed 3 years ago

...don't re-notify upon cold start if not necessary

This change solves the issue of undelivered NOTIFY's over UDP when the slave
is under high load with Valgrind. From now, the master, being Knot, will run
under Valgrind as well as the slave.

Daniel Salzman authored 3 years ago and Libor Peltan committed 3 years ago

fixes #715

This change reduces code complexity and information leakage.