iotests: fix three spurious local failures
cover iotests: fix three spurious local failures
Three independent fixes found while chasing a clean './check -qcow2' run on a Python 3.14 host. Unrelated to each other, sent as one series since they came out of the same session.
Denis V. Lunev (3):
iotests: run the test pool with the 'fork' start method
iotests: skip FUSE tests when FUSE is not usable iotests/migrate-bitmaps-postcopy-test: drop the postcopy timing
assertion
tests/qemu-iotests/common.rc | 14 ++++++++++++++ tests/qemu-iotests/testrunner.py | 6 ++++-- tests/qemu-iotests/tests/file-io-error | 4 +--- tests/qemu-iotests/tests/fuse-allow-other | 2 ++ tests/qemu-iotests/tests/fuse-mmap-shared | 9 ++++++--- .../tests/migrate-bitmaps-postcopy-test | 1 - 6 files changed, 27 insertions(+), 9 deletions(-)
Signature · 1 line
2.53.0
PATCH 1/3 iotests: run the test pool with the 'fork' start method
run_tests_pool() shares the runner via the class attribute TestRunner.shared_self, relying on worker processes to inherit it. That only works with the 'fork' start method. Python 3.14 switched the Linux default to 'forkserver', so workers see shared_self as None and parallel runs abort with:
assert runner is not None
AssertionError
Only reproduces with Python 3.14+ and 'check -jN' (N > 1); meson runs one test per process and never calls run_tests_pool(), so CI is unaffected.
Request get_context('fork') explicitly; it is available on all supported Python versions and a no-op before 3.14.
--- tests/qemu-iotests/testrunner.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
tests/qemu-iotests/testrunner.py +2 -2
@@ -26,7 +26,7 @@ import json import shutil import sys -from multiprocessing import Pool +from multiprocessing import Pool, get_context from typing import List, Optional, Any, Sequence, Dict from testenv import TestEnv
@@ -125,7 +125,7 @@ def run_tests_pool(self, tests: List[str], assert TestRunner.shared_self is None TestRunner.shared_self = self - with Pool(jobs) as p: + with get_context('fork').Pool(jobs) as p: results = p.starmap(self.proc_run_test, zip(tests, [test_field_width] * len(tests)))
Signature · 1 line
2.53.0
Re: [PATCH 1/3] iotests: run the test pool with the 'fork' start method
Am 10.07.2026 um 18:13 hat Denis V. Lunev geschrieben:
Quoted 37 lines
run_tests_pool()shares the runner via the class attribute TestRunner.shared_self, relying on worker processes to inherit it. That only works with the 'fork' start method. Python 3.14 switched the Linux default to 'forkserver', so workers seeshared_selfas None and parallel runs abort with:assert runner is not None
AssertionErrorOnly reproduces with Python 3.14+ and 'check -jN' (N > 1); meson runs one test per process and never calls
run_tests_pool(), so CI is unaffected.Request
get_context('fork')explicitly; it is available on all supported Python versions and a no-op before 3.14.Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Hanna Reitz <[email protected]>--- tests/qemu-iotests/testrunner.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)tests/qemu-iotests/testrunner.py +1
-1@@ -26,7 +26,7 @@ import json import shutil import sys -from multiprocessing import Pool +from multiprocessing import Pool, get_context from typing import List, Optional, Any, Sequence, Dict from testenv import TestEnv
tests/qemu-iotests/testrunner.py def run_tests_pool(self, tests: List[str], · line 130
Quoted 7 lines
tests/qemu-iotests/testrunner.py +1
-1@@ -125,7 +125,7 @@ def run_tests_pool(self, tests: List[str], assert TestRunner.shared_self is None TestRunner.shared_self = self - with Pool(jobs) as p: + with get_context('fork').Pool(jobs) as p: results = p.starmap(self.proc_run_test, zip(tests, [test_field_width] * len(tests)))
Makes sense, but the Pool import is now unused:
=== pylint ===
+************* Module testrunner
+testrunner.py:29:0: W0611: Unused Pool imported from multiprocessing (unused-import)
=== mypy ===
Kevin
PATCH 2/3 iotests: skip FUSE tests when FUSE is not usable
In reply to Denis V. Lunevfile-io-error, fuse-allow-other and fuse-mmap-shared skip only when FUSE is not compiled in. When FUSE is built in but unusable at run time (no /dev/fuse access, fusermount lacking permissions), the export fails to mount with "Failed to mount FUSE session to export" and the tests report a spurious failure instead of skipping, like NBD tests already do for missing NBD support.
Add _notrun_on_no_fuse() to common.rc and use it in the shell tests. fuse-mmap-shared is Python, so it gets an equivalent inline check.
--- tests/qemu-iotests/common.rc | 14 ++++++++++++++ tests/qemu-iotests/tests/file-io-error | 4 +--- tests/qemu-iotests/tests/fuse-allow-other | 2 ++ tests/qemu-iotests/tests/fuse-mmap-shared | 9 ++++++--- 4 files changed, 23 insertions(+), 6 deletions(-)
tests/qemu-iotests/common.rc +14 -0
@@ -981,6 +981,20 @@ _require_drivers() done } +# Skip if FUSE is unusable: not compiled in, or the export failed to +# mount. $1 is the failing 'block-export-add' reply. +_notrun_on_no_fuse() +{ + case "$1" in + *"Parameter 'type' does not accept value 'fuse'"*) + _notrun "No FUSE support" + ;; + *"Failed to mount FUSE session"*) + _notrun "FUSE not usable in this environment" + ;; + esac +} + # Check that we have a file system that allows huge (but very sparse) files # _require_large_file()
tests/qemu-iotests/tests/file-io-error +1 -3
@@ -89,9 +89,7 @@ output=$(_send_qemu_cmd $QEMU_HANDLE \ 'return' \ | grep -v 'option allow_other only allowed if') -if echo "$output" | grep -q "Parameter 'type' does not accept value 'fuse'"; then - _notrun 'No FUSE support' -fi +_notrun_on_no_fuse "$output" echo "$output" echo
tests/qemu-iotests/tests/fuse-allow-other +2 -0
@@ -77,6 +77,8 @@ fuse_export_add() _notrun "allow_other not supported" fi + _notrun_on_no_fuse "$output" + echo "$output" }
tests/qemu-iotests/tests/fuse-mmap-shared +6 -3
@@ -28,9 +28,12 @@ def test_fuse_support(mount_point): }) test_qsd.stop() if 'error' in res: - assert (res['error']['desc'] == - "Parameter 'type' does not accept value 'fuse'") - iotests.notrun('No FUSE support') + desc = res['error']['desc'] + if desc == "Parameter 'type' does not accept value 'fuse'": + iotests.notrun('No FUSE support') + if 'Failed to mount FUSE session' in desc: + iotests.notrun('FUSE not usable in this environment') + assert False, desc # Shared mmap when using direct IO is only supported for Linux kernels >= 6.6 # with commit e78662e818f94 ("fuse: add a new fuse init flag to relax
Signature · 1 line
2.53.0
Re: [PATCH 2/3] iotests: skip FUSE tests when FUSE is not usable
Am 10.07.2026 um 18:13 hat Denis V. Lunev geschrieben:
Quoted 25 lines
file-io-error, fuse-allow-other and fuse-mmap-shared skip only when FUSE is not compiled in. When FUSE is built in but unusable at run time (no
/dev/fuseaccess, fusermount lacking permissions), the export fails to mount with "Failed to mount FUSE session to export" and the tests report a spurious failure instead of skipping, like NBD tests already do for missing NBD support.Add
_notrun_on_no_fuse()to common.rc and use it in the shell tests. fuse-mmap-shared is Python, so it gets an equivalent inline check.Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Hanna Reitz <[email protected]>--- tests/qemu-iotests/common.rc | 14 ++++++++++++++ tests/qemu-iotests/tests/file-io-error | 4 +--- tests/qemu-iotests/tests/fuse-allow-other | 2 ++ tests/qemu-iotests/tests/fuse-mmap-shared | 9 ++++++--- 4 files changed, 23 insertions(+), 6 deletions(-)tests/qemu-iotests/common.rc +0
-0
tests/qemu-iotests/common.rc _require_drivers() · line 986
Quoted 6 lines
tests/qemu-iotests/common.rc +3
-0@@ -981,6 +981,20 @@ _require_drivers() done } +# Skip if FUSE is unusable: not compiled in, or the export failed to +# mount. $1 is the failing 'block-export-add' reply. +_notrun_on_no_fuse()
I'm not completely happy with the name, though this is not a blocker.
My first thought was to keep the name consistent with the other functions in common.rc and have something like _require_fuse. On second thoughts, this is probably confusing because it's different from the existing _require_* functions in that it takes an error message from the test case instead of doing its own probing.
So maybe at least remove the double negative? _notrun_on_fuse_error sounds slightly better to me.
tests/qemu-iotests/common.rc _require_drivers() · line 997
Quoted 11 lines
tests/qemu-iotests/common.rc +11
-0+{ + case "$1" in + *"Parameter 'type' does not accept value 'fuse'"*) + _notrun "No FUSE support" + ;; + *"Failed to mount FUSE session"*) + _notrun "FUSE not usable in this environment" + ;; + esac +} +
Adding these checks is a good idea, of course.
Kevin
PATCH 3/3 iotests/migrate-bitmaps-postcopy-test: drop the postcopy timing assertion
In reply to Denis V. Lunevtest_postcopy_success asserted downtime * 10 < postcopy_time to confirm the run spent meaningful time in postcopy. This is an unnormalized wall-clock heuristic tuned to old hardware (commit e80a4150a5), and fails on fast hosts where the bitmap payload now transfers in under a second while switch-over downtime stays fixed (e.g. downtime=0.6s, postcopy_time=0.8s).
It is also a weak proxy for the regression it guards against (postcopy degenerating to precopy): that is already covered by the bitmap sha256, mid-migration update and persistence checks, none of which depend on this timing.
Drop the assertion; keep printing the timings under debug.
--- tests/qemu-iotests/tests/migrate-bitmaps-postcopy-test | 1 - 1 file changed, 1 deletion(-)
tests/qemu-iotests/tests/migrate-bitmaps-postcopy-test +0 -1
@@ -186,7 +186,6 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): downtime = event_dist(event_stop, event_resume) postcopy_time = event_dist(event_resume, event_complete) - assert downtime * 10 < postcopy_time if debug: print('downtime:', downtime) print('postcopy_time:', postcopy_time)
Signature · 1 line
2.53.0
Re: [PATCH 3/3] iotests/migrate-bitmaps-postcopy-test: drop the postcopy timing assertion
Am 10.07.2026 um 18:13 hat Denis V. Lunev geschrieben:
Quoted 12 lines
test_postcopy_successasserteddowntime * 10 < postcopy_timeto confirm the run spent meaningful time in postcopy. This is an unnormalized wall-clock heuristic tuned to old hardware (commite80a4150a5), and fails on fast hosts where the bitmap payload now transfers in under a second while switch-over downtime stays fixed (e.g. downtime=0.6s, postcopy_time=0.8s).It is also a weak proxy for the regression it guards against (postcopy degenerating to precopy): that is already covered by the bitmap sha256, mid-migration update and persistence checks, none of which depend on this timing.
Quoted 5 lines
Drop the assertion; keep printing the timings under debug.
Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Hanna Reitz <[email protected]>
I agree that this assertion failure doesn't mean that the test failed. What it might mean is that maybe we didn't test the code as well as we would have liked to.
Should we try to slow down/throttle the migration artificially to trigger the cases we want to test here more reliably?
Kevin
Quoted 18 lines
tests/qemu-iotests/tests/migrate-bitmaps-postcopy-test | 1 - 1 file changed, 1 deletion(-)tests/qemu-iotests/tests/migrate-bitmaps-postcopy-test +0
-1@@ -186,7 +186,6 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): downtime = event_dist(event_stop, event_resume) postcopy_time = event_dist(event_resume, event_complete) - assert downtime * 10 < postcopy_time if debug: print('downtime:', downtime) print('postcopy_time:', postcopy_time)Signature · 1 line
2.53.0
Re: [PATCH 3/3] iotests/migrate-bitmaps-postcopy-test: drop the postcopy timing assertion
Quoted 15 lines · Kevin Wolf
Am 10.07.2026 um 18:13 hat Denis V. Lunev geschrieben:
Quoted 14 lines
test_postcopy_successasserteddowntime * 10 < postcopy_timeto confirm the run spent meaningful time in postcopy. This is an unnormalized wall-clock heuristic tuned to old hardware (commite80a4150a5), and fails on fast hosts where the bitmap payload now transfers in under a second while switch-over downtime stays fixed (e.g. downtime=0.6s, postcopy_time=0.8s).It is also a weak proxy for the regression it guards against (postcopy degenerating to precopy): that is already covered by the bitmap sha256, mid-migration update and persistence checks, none of which depend on this timing.
Drop the assertion; keep printing the timings under debug.
Quoted 7 lines
Quoted 3 lines
Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Hanna Reitz <[email protected]>I agree that this assertion failure doesn't mean that the test failed. What it might mean is that maybe we didn't test the code as well as we would have liked to.
Yes, the aim was to guarantee, that we don't simply pass the whole bitmap in downtime, not starting any postcopy at all.
Possible alternative is to do query-migrate exactly after RESUME, and check that remaining is significantly more than zero: this data is transferred in postcopy for sure.
Quoted 25 lines
Should we try to slow down/throttle the migration artificially to trigger the cases we want to test here more reliably?
Kevin
Quoted 18 lines
tests/qemu-iotests/tests/migrate-bitmaps-postcopy-test | 1 - 1 file changed, 1 deletion(-)tests/qemu-iotests/tests/migrate-bitmaps-postcopy-test +0
-1@@ -186,7 +186,6 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): downtime = event_dist(event_stop, event_resume) postcopy_time = event_dist(event_resume, event_complete) - assert downtime * 10 < postcopy_time if debug: print('downtime:', downtime) print('postcopy_time:', postcopy_time)Signature · 1 line
2.53.0
Signature · 2 lines
Best regards, Vladimir