Fix: mi test: possible race between listing UST events and testapp start
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 1 line · Jonathan Rajotte
Signed-off-by Jonathan Rajotte <[email protected]>
More description of the change is needed IMO.
Quoted 9 lines
--- tests/regression/tools/mi/test_mi | 5 +++++ 1 file changed, 5 insertions(+)tests/regression/tools/mi/test_mi +0
-0
tests/regression/tools/mi/test_mi function test_list_ust_event () · line 515
Quoted 9 lines
tests/regression/tools/mi/test_mi +5
-0@@ -507,6 +507,11 @@ function test_list_ust_event () #Begin testing $TESTAPP_BIN $NR_USEC_WAIT & 2>/dev/null + pid=$! + + #Wait for TESTAPP to run + while ! kill -0 $pid 2> /dev/null; do :; done + list_lttng_with_opts "-u -f"
Is this an attempt to give the test program enough time to register tracepoints?
If so, I agree there is a race, but I don't think this is the right way to address it. I would not expect the while ! kill loop to repeat at all; if you remove the stderr redirection do you see any output, ever?
The exit status from kill $! is not a reliable indication that the shell has execve'd the program in the background, and even if it were, it would not be a reliable indication of how far the program has progressed (e.g. whether it has got far enough along to register tracepoints).
If a test program needs to run in the background, yet needs to reach a certain state before the rest of the test should proceed, then it should have a way to indicate to the test harness that it has reached the desired state. The shell's rudimentary facilities for managing concurrency are not sufficient.
A message that is not in the archive (sent off-list or not yet received): [email protected]
Hey
Quoted 16 lines · Nathan Lynch
Quoted 1 line · Jonathan Rajotte
Signed-off-by Jonathan Rajotte <[email protected]>More description of the change is needed IMO.
Review
Quoted 5 lines
--- tests/regression/tools/mi/test_mi | 5 +++++ 1 file changed, 5 insertions(+)(quoted diff) +0
-0b/tests/regression/tools/mi/test_mi
Quoted 5 lines
index fa46b51..7acec18 100755
tests/regression/tools/mi/test_mi +0
-0@@ -507,6 +507,11 @@ function test_list_ust_event ()
Quoted 12 lines
tests/regression/tools/mi/test_mi function test_list_ust_event () · line 515
Quoted 8 lines
tests/regression/tools/mi/test_mi +5
-0#Begin testing $TESTAPP_BIN $NR_USEC_WAIT & 2>/dev/null + pid=$! + + #Wait for TESTAPP to run + while ! kill -0 $pid 2> /dev/null; do :; done + list_lttng_with_opts "-u -f"Is this an attempt to give the test program enough time to register tracepoints?
A poor one... if one at all.... for which I needed feedback (missing rfc tag here) because I could not simply find any simple way. See [1]
Quoted 2 lines
If so, I agree there is a race, but I don't think this is the right way to address it.
I would not expect the while ! kill loop to repeat at
Quoted 1 line
all;
The presence of ":" assure the execution of the while loop. A simple test on a process that does not exist yield a lot of printing:
while ! kill -0 500; do :; done
Anyway it does not solve the problem. ;) and I'm not sure you were talking about that.
if you remove the stderr redirection do you see any output, ever?
The exit status from kill $! is not a reliable indication that the shell
Quoted 2 lines
has execve'd the program in the background, and even if it were, it would not be a reliable indication of how far the program has progressed
+1
(e.g. whether it has got far enough along to register tracepoints).
Quoted 9 lines
If a test program needs to run in the background, yet needs to reach a certain state before the rest of the test should proceed, then it should have a way to indicate to the test harness that it has reached the desired state. The shell's rudimentary facilities for managing concurrency are not sufficient.
I'm pushing limits here regarding simplicity: how about a trap and signal exchange between the testapp... might give it a shot. Or a simple sleep...
Thanks, I'll think about it a little more.
[1]:http://meta.wikimedia.org/wiki/Cunningham%27s_Law Bonus xkcd: http://xkcd.com/386/
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
Quoted 10 lines · Jonathan Rajotte
Signed-off-by Jonathan Rajotte <[email protected]>--- tests/regression/tools/mi/test_mi | 5 +++++ 1 file changed, 5 insertions(+)tests/regression/tools/mi/test_mi +0
-0
tests/regression/tools/mi/test_mi function test_list_ust_event () · line 513
Quoted 7 lines
tests/regression/tools/mi/test_mi +4
-0@@ -507,6 +507,11 @@ function test_list_ust_event () #Begin testing $TESTAPP_BIN $NR_USEC_WAIT & 2>/dev/null + pid=$! + + #Wait for TESTAPP to run + while ! kill -0 $pid 2> /dev/null; do :; done
So of what I understand here, this tests the existance of the PID but wondering if it really fixes the race. Even if the PID indeed exists, still it might not have registered to the session daemon.
I think one way to fix that is to have a "lttng list -u | grep $PID" test and if the output is not empty, the application did registered. Also, you can use the "create file" option of gen-ust-events. You can pass a fourth argument to the app, once a tracepoint is hit, it will create the file you gave with the full path. You can then loop on that. There are some tests that already uses that, you can check it out.
Finally, that test should actually spawn the application with a long life time and kill it at the end so we don't race between very long commands and the application actually quiting gracefully.
Cheers!
David
Quoted 12 lines
tests/regression/tools/mi/test_mi +1
-0+ list_lttng_with_opts "-u -f" $XML_VALIDATE $OUTPUT_DEST ok $? "Mi test: list ust event xsd validation"Signature · 7 lines
1.9.1 _______________________________________________ lttng-dev mailing list [email protected] http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev