From: Martin Mares Date: Sun, 24 Aug 2008 21:56:44 +0000 (+0200) Subject: Introduced sampling of memory usage. X-Git-Tag: python-dummy-working~138 X-Git-Url: http://mj.ucw.cz/gitweb/?a=commitdiff_plain;h=785436fb49da899e77a3a977f61f564a6a71f831;p=moe.git Introduced sampling of memory usage. This patch introduces the basic stuff. So far, it samples VmPeak before and after each syscalls, which will be fixed in subsequent commits. --- diff --git a/TODO b/TODO index db24292..797b15f 100644 --- a/TODO +++ b/TODO @@ -8,6 +8,7 @@ interactive tasks: solve deadlocks? interactive tasks: logging of messages terminology: problem -> task ? box: memory usage statistics (need to search for VMPeak in /proc/$PID/status, but it disappears too early) +box: explicit breakpoints are reported as syscall #-1 Judges ~~~~~~ diff --git a/box/box.c b/box/box.c index 1394a85..2aaec6a 100644 --- a/box/box.c +++ b/box/box.c @@ -47,7 +47,11 @@ static int ticks_per_sec; static int exec_seen; static int partial_line; +static int mem_peak_kb; +static int total_ms, wall_ms; + static void die(char *msg, ...) NONRET; +static void sample_mem_peak(void); /*** Meta-files ***/ @@ -85,8 +89,6 @@ meta_printf(const char *fmt, ...) va_end(args); } -static int total_ms, wall_ms; - static void final_stats(struct rusage *rus) { @@ -99,6 +101,7 @@ final_stats(struct rusage *rus) meta_printf("time:%d.%03d\n", total_ms/1000, total_ms%1000); meta_printf("time-wall:%d.%03d\n", wall_ms/1000, wall_ms%1000); + meta_printf("mem:%llu\n", (unsigned long long) mem_peak_kb * 1024); } /*** Messages and exits ***/ @@ -108,6 +111,7 @@ box_exit(int rc) { if (box_pid > 0) { + sample_mem_peak(); if (is_ptraced) ptrace(PTRACE_KILL, box_pid); kill(-box_pid, SIGKILL); @@ -705,6 +709,27 @@ signal_int(int unused UNUSED) err("SG: Interrupted"); } +#define PROC_BUF_SIZE 4096 +static void +read_proc_file(char *buf, char *name, int *fdp) +{ + int c; + + if (!*fdp) + { + sprintf(buf, "/proc/%d/%s", (int) box_pid, name); + *fdp = open(buf, O_RDONLY); + if (*fdp < 0) + die("open(%s): %m", buf); + } + lseek(*fdp, 0, SEEK_SET); + if ((c = read(*fdp, buf, PROC_BUF_SIZE-1)) < 0) + die("read on /proc/$pid/%s: %m", name); + if (c >= PROC_BUF_SIZE-1) + die("/proc/$pid/%s too long", name); + buf[c] = 0; +} + static void check_timeout(void) { @@ -722,35 +747,23 @@ check_timeout(void) } if (timeout) { - char buf[4096], *x; - int c, utime, stime, ms; - static int proc_status_fd; - if (!proc_status_fd) - { - sprintf(buf, "/proc/%d/stat", (int) box_pid); - proc_status_fd = open(buf, O_RDONLY); - if (proc_status_fd < 0) - die("open(%s): %m", buf); - } - lseek(proc_status_fd, 0, SEEK_SET); - if ((c = read(proc_status_fd, buf, sizeof(buf)-1)) < 0) - die("read on /proc/$pid/stat: %m"); - if (c >= (int) sizeof(buf) - 1) - die("/proc/$pid/stat too long"); - buf[c] = 0; + char buf[PROC_BUF_SIZE], *x; + int utime, stime, ms; + static int proc_stat_fd; + read_proc_file(buf, "stat", &proc_stat_fd); x = buf; while (*x && *x != ' ') x++; while (*x == ' ') x++; if (*x++ != '(') - die("proc syntax error 1"); + die("proc stat syntax error 1"); while (*x && (*x != ')' || x[1] != ' ')) x++; while (*x == ')' || *x == ' ') x++; if (sscanf(x, "%*c %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d %d %d", &utime, &stime) != 2) - die("proc syntax error 2"); + die("proc stat syntax error 2"); ms = (utime + stime) * 1000 / ticks_per_sec; if (verbose > 1) fprintf(stderr, "[time check: %d msec]\n", ms); @@ -759,6 +772,51 @@ check_timeout(void) } } +static void +sample_mem_peak(void) +{ + /* + * We want to find out the peak memory usage of the process, which is + * maintained by the kernel, but unforunately it gets lost when the + * process exits (it is not reported in struct rusage). Therefore we + * have to sample it whenever we suspect that the process is about + * to exit. + */ + char buf[PROC_BUF_SIZE], *x; + static int proc_status_fd; + read_proc_file(buf, "status", &proc_status_fd); + + x = buf; + while (*x) + { + char *key = x; + while (*x && *x != ':' && *x != '\n') + x++; + if (!*x || *x == '\n') + break; + *x++ = 0; + while (*x == ' ' || *x == '\t') + x++; + + char *val = x; + while (*x && *x != '\n') + x++; + if (!*x) + break; + *x++ = 0; + + if (!strcmp(key, "VmPeak")) + { + int peak = atoi(val); + if (peak > mem_peak_kb) + mem_peak_kb = peak; + } + } + + if (verbose > 1) + msg("[mem-peak: %u KB]\n", mem_peak_kb); +} + static void boxkeeper(void) { @@ -820,9 +878,10 @@ boxkeeper(void) if (wall_timeout && wall_ms > wall_timeout) err("TO: Time limit exceeded (wall clock)"); flush_line(); - fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d syscalls)\n", + fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d MB, %d syscalls)\n", total_ms/1000, total_ms%1000, wall_ms/1000, wall_ms%1000, + (mem_peak_kb + 1023) / 1024, syscall_count); box_exit(0); } @@ -830,6 +889,7 @@ boxkeeper(void) { box_pid = 0; meta_printf("exitsig:%d\n", WTERMSIG(stat)); + final_stats(&rus); err("SG: Caught fatal signal %d%s", WTERMSIG(stat), (syscall_count ? "" : " during startup")); } if (WIFSTOPPED(stat)) @@ -872,11 +932,13 @@ boxkeeper(void) } else /* Syscall return */ msg("= %ld\n", u.regs.eax); + sample_mem_peak(); ptrace(PTRACE_SYSCALL, box_pid, 0, 0); } else if (sig != SIGSTOP && sig != SIGXCPU && sig != SIGXFSZ) { msg(">> Signal %d\n", sig); + sample_mem_peak(); /* Signal might be fatal, so update mem-peak */ ptrace(PTRACE_SYSCALL, box_pid, 0, sig); } else