| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
| |
Instead of using the number 3 directly, let's introduce and use
WATCHDOG_MAX_TICKS.
Reviewed-by: Petr Čech <pcech@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When a time is changed into the past during sssd runtime
(e.g. on boot during time correction), it is possible that
we never hit watchdog tevent timer since it is based on
system time.
This patch adds a past-time shift detection mechanism. If a time
shift is detected we restart watchdog.
Resolves:
https://fedorahosted.org/sssd/ticket/3154
Reviewed-by: Lukáš Slebodník <lslebodn@redhat.com>
Reviewed-by: Fabiano Fidêncio <fidencio@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Older versions of gcc does not like initialisation of struct sigevent
because the first member of structure is union (sigval_t)
src/util/util_watchdog.c: In function 'setup_watchdog':
src/util/util_watchdog.c:77:12:
warning: missing braces around initializer [-Wmissing-braces]
struct sigevent sev = { 0 };
^
src/util/util_watchdog.c:77:12:
warning: (near initialization for 'sev.sigev_value') [-Wmissing-braces]
src/util/util_watchdog.c:77:12:
warning: initialization makes integer from pointer without a cast
src/util/util_watchdog.c:77:12:
warning: (near initialization for 'sev.sigev_value.sival_int')
Reviewed-by: Sumit Bose <sbose@redhat.com>
|
|
|
The watchdog uses a kernel timer to issue a signal to the process.
It checks if the ticker is not being reset by the main event loop, which
would indicate that the process got stuck.
At the same time it sets a tevent timer to clear the watchdog ticker, so
that the watchdog handler is kept happy.
If the watchdog detects that the timer event failed to reset the watchdog for
three times in a row then the process is killed.
Normally the monitor will detect the child terminated and will rescheduled it.
Related:
https://fedorahosted.org/sssd/ticket/2921
Reviewed-by: Jakub Hrozek <jhrozek@redhat.com>
Reviewed-by: Pavel Březina <pbrezina@redhat.com>
|
