list -k seems corrupted sometimes
When doing repeated short lived live kernel tracing sessions, the output of lttng list -k is sometimes corrupted (the loglevel value seems random) or some events are missing.
lttng-modules : 2.4.0-rc4 309370e252e189867745fb31fe5c15c9ff494f34
lttng-tools : 2.4.0-rc5 1d6448a074accdc6cb32e2f724f9768b006a6d67
Steps to reproduce :
- Compile lttngtop
- in utils/lttngtrace, check where the "$SUDO lttng list
k" is called and add an exit in the if statement below that (to have a clean exit when the problem appears) while true; do ./utils/lttngtrace ls; done
After less than 2 minutes, you should see the error message printed in loop.
It will stay in the corrupted state until you let lttngtrace continue (removing the exit in the if statement).
A copy of the corrupted lttng list -k is attached to this bug.
No change in the list of loaded modules has been noticed.
The lttng-sessiond, consumer and relay don't restart between runs of lttngtrace.
Basically, lttngtrace creates a live session with a set of events and contexts; attach to it with a client and destroys the session.
There might be something related to the list of enabled events and/or contexts, because reproducing this with only sched_switch enabled does not seem to reproduce the problem.
Updated by Mathieu Desnoyers about 6 years ago
- Status changed from New to Resolved
commit 53efb85a242809ed5ed21e9ab40effa696ecbc6f Author: Mathieu Desnoyers <firstname.lastname@example.org> Date: Thu Feb 20 10:23:46 2014 -0500 Fix: miscellaneous memory handling fixes Issues found with Valgrind: - Uninitialized memory sent through sendmsg, - Uninitialized realloc memory, - Erroneous pointer arithmetic in jul_list_events(), - Memory leaks in kernel_consumer_add_channel(), kernel_consumer_add_metadata(), - Memory leak (snapshot_destroy missing hash table cleanup), - Uninitialized poll/epoll memory on realloc. Signed-off-by: Mathieu Desnoyers <email@example.com> Signed-off-by: David Goulet <firstname.lastname@example.org>