Michał Kępień [Tue, 13 Mar 2018 11:04:37 +0000 (12:04 +0100)]
Add CHANGES entry
4914. [bug] A bug in zone database reference counting could lead to
a crash when multiple versions of a slave zone were
transferred from a master in close succession.
[GL #134]
Michał Kępień [Tue, 13 Mar 2018 10:28:11 +0000 (11:28 +0100)]
Fix a zone database reference counting bug in dump_done()
A typo in commit d39ab7440e introduced a bug in zone database reference
counting which leads to a crash if dumping one version of a slave zone
is not finished by the time transferring a newer version from a master
completes. Correct the typo to fix reference counting, thus preventing
crashes.
Evan Hunt [Tue, 27 Feb 2018 18:57:08 +0000 (10:57 -0800)]
migrate tests from bin/tests/dnssec-signzone to bin/tests/system/dnssec
- added tests to the dnssec system test that duplicate the ones
from bin/tests/dnssec-signzone
- changed cleanall.sh so it doesn't automatically remove all
key files, because there are now some of those that are part of the
distribution
Evan Hunt [Tue, 27 Feb 2018 20:48:10 +0000 (12:48 -0800)]
move all optional tests from bin/tests to bin/tests/optional
- some of these tests are obsolete and should be cleared up,
others overlap with ATF tests and may be removed later.
for now, let's just tidy up the bin/tests directory by
moving these files down a level.
Evan Hunt [Fri, 9 Mar 2018 00:53:16 +0000 (16:53 -0800)]
capture more information from unit tests
- in unittest step, explicitly preserve kyua.log or atf.out
- preserve kyua results database if present
- generate HTML report from kyua results if available
Michał Kępień [Fri, 9 Mar 2018 13:09:28 +0000 (14:09 +0100)]
Enable runner-specific make concurrency settings
Using fixed make concurrency settings on all runners is not flexible and
requires .gitlab-ci.yml to be modified each time tweaking these settings
is needed. Use environment variables which are expected to be set by
the runner (defaulting to 1 in case they are not set) for controlling
make concurrency.
Michał Kępień [Fri, 9 Mar 2018 13:09:26 +0000 (14:09 +0100)]
Tweak the way ccache is used during CI
Our current CI configuration causes ccache data to be zipped after each
job and also included in build artifacts, which will quickly become
infeasible as ccache data grows. Instead of asking gitlab-runner to
preserve ccache data between jobs, keep a separate ccache directory on
each runner, expecting it to be accessible at /ccache when a CI job is
run. As this requires gitlab-runner to be configured in a specific way,
do not use ccache at all in case the ccache directory is not found while
building.
Michał Kępień [Thu, 8 Mar 2018 14:40:18 +0000 (15:40 +0100)]
Tweak timestamp checking in the cds system test
Given the characteristics of the three timestamps involved in file
modification time checks in the cds system test (each one is an hour
apart from the next), reduce the resolution of these checks to 1 minute.
This will prevent intermittent false negatives caused by exceeding the
currently allowed difference of 9 seconds between file modification
times without making the test moot.
Also note that by using abs(), checkmtime.pl allows the cds system test
to pass when the modification time of the checked file is less than an
hour (or two hours for the second check) in the past. This should never
happen, so remove abs() from the condition checked by checkmtime.pl.
Michał Kępień [Tue, 6 Mar 2018 10:33:16 +0000 (11:33 +0100)]
Fix a race between "rndc reconfig" and waiting for a ./DNSKEY fetch to complete
Calling nextpart() after reconfiguring ns1 is not safe, because the
expected log message may appear in ns5/named.run before nextpart() is
run. With the TTL for ./DNSKEY set to 20 seconds, ns5 will refresh it
after 10 seconds, by which time wait_for_log() will already have failed.
This results in a false negative.
However, just calling nextpart() before reconfiguring ns1 would
introduce a different problem: if ns5 refreshed ./DNSKEY between these
two steps, the subsequent wait_for_log() call would return immediately
as it would come across the log message about a failure while refreshing
./DNSKEY instead of the expected success. This in turn would result in
a different false negative as the root key would still be uninitialized
by the time "rndc secroots" is called.
Prevent both kinds of false negatives by:
- calling nextpart() before reconfiguring ns1, in order to prevent the
first case described above,
- looking for a more specific log message, in order to prevent the
second case described above.
Also look for a more specific log message in the first part of the
relevant check, not to fix any problem, but just to emphasize that a
different fetch result is expected in that case.
With these tweaks in place, if a (failed) ./DNSKEY refresh is scheduled
between nextpart() and reconfiguring ns1, wait_for_log() will just wait
for two more seconds (one "hour"), at which point another refresh
attempt will be made that will succeed.
projects / thirdparty / bind9.git / log
