This still is an issue in at-3.1.16
"Abrupt reboot of a server, from a crash for example, can leave
malformed at job files, which can cause atd to go into high cpu
utilization after boot.
Logs can be flooded with such errors as:
# grep atd /var/log/messages|tail
Aug 26 11:24:29 isp2264 atd[30648]: Job 62696 (a0f4e8016e2e2b) -
userid 250
does not match file uid 0
Aug 26 11:24:30 isp2264 atd[30649]: Job 62696 (a0f4e8016e2e2b) -
userid 250
does not match file uid 0
Aug 26 11:24:30 isp2264 atd[30652]: Job 62696 (a0f4e8016e2e2b) -
userid 250
does not match file uid 0
Aug 26 11:24:30 isp2264 atd[30654]: Job 62696 (a0f4e8016e2e2b) -
userid 250
does not match file uid 0
(...)"
I'm attaching a patch that reschedules failed jobs with a pause in order
to avoid error messages flood. It also tries to prevent creation of the
corrupted files by using 'O_SYNC' flag with open() system call.
Best regards
Kristyna Streitova
Index: at-3.1.13/at.c
===================================================================
--- at-3.1.13.orig/at.c
+++ at-3.1.13/at.c
@@ -319,7 +319,8 @@ writefile(time_t runtimer, char queue)
* bit. Yes, this is a kluge.
*/
cmask = umask(S_IRUSR | S_IWUSR | S_IXUSR);
- if ((fd = open(atfile, O_CREAT | O_EXCL | O_TRUNC | O_WRONLY, S_IRUSR)) == -1)
+ if ((fd = open(atfile,
+ O_CREAT | O_EXCL | O_TRUNC | O_WRONLY | O_SYNC, S_IRUSR)) == -1)
perr("Cannot create atjob file %.500s", atfile);
if ((fd2 = dup(fd)) < 0)
Index: at-3.1.13/atd.c
===================================================================
--- at-3.1.13.orig/atd.c
+++ at-3.1.13/atd.c
@@ -103,6 +103,7 @@ int selinux_enabled=0;
#define BATCH_INTERVAL_DEFAULT 60
#define CHECK_INTERVAL 3600
+#define RETRY_INTERVAL CHECK_INTERVAL
/* Global variables */
@@ -845,12 +846,17 @@ run_loop()
/* Something went wrong the last time this was executed.
* Let's remove the lockfile and reschedule.
+ *
+ * To prevent pointless CPU heating with permanent errors,
+ * next execution is scheduled with RETRY_INTERVAL inserted.
*/
strncpy(lock_name, dirent->d_name, sizeof(lock_name)-1);
lock_name[sizeof(lock_name)-1] = 0;
lock_name[0] = '=';
unlink(lock_name);
- next_job = now;
+ if (next_job > now + RETRY_INTERVAL) {
+ next_job = now + RETRY_INTERVAL;
+ }
nothing_to_do = 0;
}
continue;
