Trying to mitigate a file system issue

Description

The test were failing on the default gitlab runners used by normal gitlab users from https://gitlab.com/username/buildstream that are not pushing directly to https://gitlab.com/BuildStream/buildstream and thus using the special runners.

This was because:

• The cache was sorting objects by time stamp but due to the mtime limitation they all had the same time stamp so when sorted they were not in the correct order.
• When elements were build and then a change made so that the cached item and the change had the same time stamp from mtime then buildstream could not tell that the cached item was older than the change and so could not tell that it needed to be rebuilt.

Changes

This patch adds a sleep to force the mtime values to be different and there for to have the intended behaviour, these are all in the tests as this is fundamentally a issue with the tests rather than with buildstream, although it is a limitation.

Other more complex approaches could be taken if the community thinks the additional time would be well spent.

Related items

This merge request, when approved, will close issue/bug: #484 (closed)

---

tests/artifactcache/expiry.py

@@ -5,7 +5,7 @@ import pytest
 from buildstream import _yaml
 from buildstream._exceptions import ErrorDomain, LoadErrorReason
 
-from tests.testutils import cli, create_element_size
+from tests.testutils import cli, create_element_size, wait_for_cache_granularity
 
 
 DATA_DIR = os.path.join(
@@ -108,6 +108,8 @@ def test_expiry_order(cli, datafiles, tmpdir):
     res = cli.run(project=project, args=['build', 'target2.bst'])
     res.assert_success()
 
+    wait_for_cache_granularity()
+
     # Now extract dep.bst
     res = cli.run(project=project, args=['checkout', 'dep.bst', checkout])
     res.assert_success()