@@ -5,4 +5,14 @@ cd @PTEST_PATH@/tests/
# Clear old data
rm -rf testsuite.dir testsuite.log
-./testsuite --am-fmt
+# Report free space: the sparse tests create multi-GB files and the
+# ptest image is small, so this helps diagnose ENOSPC failures.
+df -h .
+
+# sparse05 (listing sparse files bigger than 2^33 B) writes ~10 GiB of
+# real data (1000 fragments of 10 MiB each), which cannot fit in the
+# ptest image. When genfile hits ENOSPC the test is normally reported
+# as skipped, but the filesystem is left full for a moment and the
+# autotest bookkeeping (status files, mkdir of the next test group)
+# intermittently fails instead, taking the next test down with it.
+./testsuite --am-fmt -k '!sparse05'
The "listing sparse files bigger than 2^33 B" test (sparse05) generates a file with 1000 data fragments of 10 MiB each, i.e. about 10 GiB of real data, which cannot fit in the ptest image (1.4 GB rootfs). Normally genfile fails with ENOSPC and the test is reported as skipped, but the filesystem stays full until the test directory is cleaned up. When the autotest bookkeeping (status files, mkdir for the next test group) hits that window, the test is reported as FAIL and the next test cannot even start, matching the intermittent autobuilder failure. Skip sparse05 explicitly and print the available disk space at the start of the run to make similar failures easier to diagnose. Tested on qemux86-64 with musl: 200 tests passed, 46 skipped, 0 failed, no ENOSPC errors. [YOCTO #16419] Signed-off-by: Enzo Frese <freseenzo@gmail.com> --- meta/recipes-extended/tar/tar/run-ptest | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)