| Commit message (Collapse) | Author | Age |
|---|
| | |
|
| |
|
|
|
|
| |
It's not safe to assume that we can get into the store before housekeeping runs,
so don't try. Just wait for housekeeping to run and check that all the files are
deleted afterwards.
|
| |
|
|
|
|
|
| |
Refactor all common code from testbbackupd and testbackupstore to allow other
test suites to contain multiple tests and execute selected tests more easily.
Report all test results within a suite in a standard, easy to read summary.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
Once again, the Windows issue of being unable to delete or overwrite an
open file causes issues. In this case it's only test failures. We need to
be diligent about closing open file handles and protocol sessions in tests.
|
| |
|
|
|
|
|
|
|
|
|
| |
The old assertion, that the write lock file exists before starting checking,
was erroneously passing before when no lock was held, because the lockfile
was never deleted. Now that we delete it when unlocking the account, this
started causing test failures.
Changed the way that accounts are checked for errors to use a function that
acquires a write lock first, and modified test to disconnect open clients
before starting checking the account, to fix it.
|
| |
|
|
|
| |
I needed reliable exit codes to run the tests in a loop to catch an
intermittent failure.
|
| |
|
|
|
| |
NamedLock simply didn't work before. This may cause test failures, but the
tests are already failing on Windows, and must be fixed.
|
| | |
|
| |
|
|
| |
Files need to be closed before renaming over them on Windows.
|
| | |
|
| |
|
|
|
|
| |
It was checking for raidfiles files that have different filenames in release
and debug builds, and that aren't even deleted by the test in release builds.
|
| | |
|
| |
|
|
|
| |
Should make the Travis logs shorter and more readable.
|
| |
|
|
|
|
| |
Logging everything to stdout (so that Windows users can redirect it) causes
extra output on stdout that confused the Perl script for this test.
|
| | |
|
| |
|
|
|
| |
They cause a lot of noise in the Travis build logs.
|
| |
|
|
| |
for reporting.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
This stops stale daemons from hanging around if a single test fails because
it throws an exception, which otherwise would cause the whole suite to
abort immediately without cleaning up after itself.
|
| |
|
|
|
| |
This just results in huge console noise if we can't kill a daemon for some
reason. Kill them once, after all tests have run, instead.
|
| |
|
|
|
| |
BSD tar seems to not like additional options after the first block.
|
| |
|
|
|
|
| |
Test now passes reliably with any verbosity level on NetBSD, despite
the really slow compares.
|
| |
|
|
|
|
| |
Running a compare takes far too long on NetBSD (3 seconds) and this was
messing up the timing of the test.
|
| |
|
|
|
|
|
|
| |
fail.
We didn't take into account the time taken to perform a compare as part of
the test, when deciding how long to wait for bbackupd to recover.
|
| |
|
|
| |
Hopefully will help anyone trying to debug this test in future.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
The shorter delay on error, introduced recently to speed up test runs,
combined with a guess about how long bbackupd would take to run a backup
and discover that the client store marker had changed, meant that sometimes
the daemon would have discovered the problem, aborted, waited and run again
while the test was waiting for the first failure.
Synchronising with the running daemon using bbackupctl wait_for_sync_end
should make the test timing accurate enough. We also tighten the recovery
timing checks to make sure that it's doing what it should.
|
| |
|
|
|
| |
We're using an internal daemon here, so we don't want an external one as well.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Older versions of GNU tar fail to set the timestamps on symlinks, which makes
them appear too recent/new to be backed up immediately, causing
test_bbackupd_uploads_files() for example to fail. Fixed by restoring the
timestamps manually after extracting the fixture archive.
For more details about the issue in tar, please see:
http://lists.gnu.org/archive/html/bug-tar/2009-08/msg00007.html
http://git.savannah.gnu.org/cgit/tar.git/plain/NEWS?id=release_1_24
This resulted in symlinks in fixture test files
|
| | |
|
| |
|
|
|
|
|
|
| |
This makes the tests run ~300 seconds faster, and coincidentally somehow
fixes or prevents a test failure on FreeBSD on this test, probably some
race condition.
|
| |
|
|
|
|
|
| |
It seems that sending a NULL pointer to an ostringstream on FreeBSD gives
different results depending on the type of the pointer. It could write 0
or 0x0, and if they don't match, tests will fail.
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
Normally this will be done by housekeeping anyway, so it's not that useful,
but it's good to check for it in tests, especially when testing
BackupStoreCheck's ability to repair bad things that happened to the store
when they involve references.
|
| |
|
|
|
|
|
|
|
|
| |
Add a standard method to Replyable that will be called if a recoverable
exception (a BoxException) occurs, and can return a protocol Message to be
sent to the client, such as an error code for various standard errors, or
rethrow the exception.
If you want something different, catch exceptions and return the desired
reply yourself, or you'll get the default handling.
|
| | |
|
| |
|
|
|
|
| |
Conflicts:
test/backupstore/testbackupstore.cpp
|
| | |
|
| |
|
|
| |
Use them in testbackupstorefix to reduce code duplication.
|
