From 636aaa2b8bdaea28284d128949119ee09612f5ee Mon Sep 17 00:00:00 2001
From: Timothy Stack <stack@flux.utah.edu>
Date: Mon, 25 Oct 2004 19:25:06 +0000
Subject: [PATCH] Changes to the "auto nice daemon" so it can work better in
Emulab.
* sensors/and/GNUmakefile.in: Emulab-specific make file. Updated to
work with a build tree separate from the source and gave it a new
version number. Files are installed under "/usr/testbed/" on ops.
* sensors/and/Makefile: Add a warning that this is not the real
makefile for us.
* sensors/and/and-OpenBSD.c: Update to work with FreeBSD and add
support for reporting process start time.
* sensors/and/and-emulab.conf.in: Emulab-specific configuration,
similar to the standard one, except it sends mail to tbops when it
does something.
* sensors/and/and-emulab.priorities: Emulab-specific priorities
database. It excludes daemon pseudo users and the event-scheduler,
otherwise, niceness levels apply to everyone.
* sensors/and/and.8.man: Add the pid file to the 'FILES' section.
* sensors/and/and.c: Added support for running a command when a
niceness level change occurs. Also writes out the pid file when
not running in test mode.
* sensors/and/and.conf.5.man: Add docs for the level commands.
* sensors/and/and.h: Add start time and child CPU time to the
and_procent struct.
* sensors/and/and.startup: Changed to use "test" instead of bash
syntax and the pid file is now used instead of killall. Also
added a "reconfig" option that HUPs the daemon.
---
sensors/and/GNUmakefile.in | 217 ++++++++++++++++++++++++++++++
sensors/and/Makefile | 11 +-
sensors/and/and-OpenBSD.c | 28 ++--
sensors/and/and-emulab.conf.in | 59 ++++++++
sensors/and/and-emulab.priorities | 66 +++++++++
sensors/and/and.8.man | 6 +
sensors/and/and.c | 214 +++++++++++++++++++++++++++--
sensors/and/and.conf.5.man | 54 ++++++++
sensors/and/and.h | 6 +-
sensors/and/and.startup | 34 +++--
10 files changed, 665 insertions(+), 30 deletions(-)
create mode 100644 sensors/and/GNUmakefile.in
create mode 100644 sensors/and/and-emulab.conf.in
create mode 100644 sensors/and/and-emulab.priorities
diff --git a/sensors/and/GNUmakefile.in b/sensors/and/GNUmakefile.in
new file mode 100644
index 0000000000..292a40b1a8
--- /dev/null
+++ b/sensors/and/GNUmakefile.in
@@ -0,0 +1,217 @@
+#
+# EMULAB-COPYRIGHT
+# Copyright (c) 2004 University of Utah and the Flux Group.
+# All rights reserved.
+#
+# This is the Emulab specific makefile.
+#
+
+#
+# Makefile for auto nice daemon
+#
+# 1999-2004 Patrick Schemitz <schemitz@users.sourceforge.net>
+# http://and.sourceforge.net/
+#
+
+SRCDIR = @srcdir@
+TESTBED_SRCDIR = @top_srcdir@
+OBJDIR = ../..
+SUBDIR = sensors/and
+
+include $(OBJDIR)/Makeconf
+
+all: and doc and-emulab.conf
+
+include $(TESTBED_SRCDIR)/GNUmakerules
+
+#
+# Init script.
+#
+INITSCRIPT=and.init
+
+#
+# Install to the ops directory.
+#
+PREFIX=$(INSTALL_DIR)/opsdir
+INSTALL_ETC=$(PREFIX)/etc
+INSTALL_INITD=$(PREFIX)/etc/rc.d
+INSTALL_SBIN=$(PREFIX)/sbin
+INSTALL_MAN=$(PREFIX)/man
+
+#
+# Version and date
+#
+VERSION=1.2.1-emulab
+DATE="25 Oct 2004"
+
+#
+# Man pages
+#
+MANPAGES=and.8 and.conf.5 and.priorities.5
+
+#
+# Determine architecture from uname(1)
+#
+ARCH=$(shell uname)
+
+#
+# Architecture-dependent settings: ANSI C compiler and linker
+#
+ifeq (${ARCH},Linux)
+ LIBS =
+else
+ifeq (${ARCH},OSF1)
+ LIBS =
+else
+ifeq (${ARCH},OpenBSD)
+ LIBS = -lkvm
+else
+ifeq (${ARCH},FreeBSD)
+ LIBS = -lkvm
+endif
+endif
+endif
+endif
+
+
+#
+# Build the auto-nice daemon.
+#
+and: and.o $(INITSCRIPT) and-$(ARCH).o
+ $(LD) and.o and-$(ARCH).o -o and $(LIBS)
+
+
+#
+# Independent part: configuration management, priority database.
+#
+# XXX Emulab note: We cannot use INSTALL_ETC for the paths since those may
+# not exist on ops.
+#
+and.o: and.c and.h
+ $(CC) -DDEFAULT_INTERVAL=60 -DDEFAULT_NICE=0 \
+ -DDEFAULT_CONFIG_FILE=\"$(INSTALL_DIR)/etc/and.conf\" \
+ -DDEFAULT_DATABASE_FILE=\"$(INSTALL_DIR)/etc/and.priorities\" \
+ -DAND_VERSION=\"$(VERSION)\" -DAND_DATE=\"$(DATE)\" -c $<
+
+
+#
+# Unix variant specific stuff
+#
+and-Linux.o: and-Linux.c and.h
+ $(CC) -c $<
+
+and-OpenBSD.o: and-OpenBSD.c and.h
+ $(CC) -c $<
+
+and-FreeBSD.o: and-OpenBSD.c and.h
+ $(CC) -c $< -o $@
+
+and-OSF1.o: and-OSF1.c and.h
+ $(CC) -c $< $@
+
+and-IRIX.o: and-OSF1.c and.h
+ $(CC) -c $< -o $@
+
+and-IRIX64.o: and-OSF1.c and.h
+ $(CC) -c $< -o $@
+
+and-SunOS.o: and-OSF1.c and.h
+ $(CC) -c $< -o $@
+
+
+
+#
+# Create script for SysV init
+#
+and.init: and.startup
+ sed s:INSTALL_SBIN:$(INSTALL_SBINDIR):g < $< > $@
+ chmod +x and.init
+
+
+#
+# Man pages
+#
+doc: $(MANPAGES)
+
+and.8: and.8.man
+ cat $< | \
+ sed s/__VERSION__/$(VERSION)/g | \
+ sed s/__DATE__/$(DATE)/g > $@
+
+and.conf.5: and.conf.5.man
+ cat $< | \
+ sed s/__VERSION__/$(VERSION)/g | \
+ sed s/__DATE__/$(DATE)/g > $@
+
+and.priorities.5: and.priorities.5.man
+ cat $< | \
+ sed s/__VERSION__/$(VERSION)/g | \
+ sed s/__DATE__/$(DATE)/g > $@
+
+#
+# Nothing to do for client install.
+#
+client client-install:
+
+#
+# Install and under $(PREFIX)/bin etc.
+#
+install: all
+ strip and
+#-mkdir $(PREFIX)
+ -mkdir -p $(DESTDIR)$(INSTALL_SBIN)
+ -mkdir -p $(DESTDIR)$(INSTALL_ETC)
+ -mkdir -p $(DESTDIR)$(INSTALL_INITD)
+ -mkdir -p $(DESTDIR)$(INSTALL_MAN)/man5
+ -mkdir -p $(DESTDIR)$(INSTALL_MAN)/man8
+ $(INSTALL) -m 0755 and $(DESTDIR)$(INSTALL_SBIN)
+ $(INSTALL) -m 0644 and-emulab.conf $(DESTDIR)$(INSTALL_ETC)/and.conf
+ $(INSTALL) -m 0644 $(SRCDIR)/and-emulab.priorities \
+ $(DESTDIR)$(INSTALL_ETC)/and.priorities
+ifneq (${INITSCRIPT},)
+ifneq (${INSTALL_INITD},)
+ @echo "Installing SysV script in $(DESTDIR)$(INSTALL_INITD)"
+ $(INSTALL) -m 0755 $(INITSCRIPT) $(DESTDIR)$(INSTALL_INITD)/and
+else
+ @echo "Installing SysV script in $(DESTDIR)$(INSTALL_SBIN)"
+ $(INSTALL) -m 0755 $(INITSCRIPT) $(DESTDIR)$(INSTALL_SBIN)
+ @echo "Installing SysV init.d finder in $(DESTDIR)$(INSTALL_SBIN)"
+ $(INSTALL) -m 0755 and-find-init.d $(DESTDIR)$(INSTALL_SBIN)
+endif
+endif
+ $(INSTALL) -m 0644 and.8 $(DESTDIR)$(INSTALL_MAN)/man8
+ $(INSTALL) -m 0644 and.conf.5 $(DESTDIR)$(INSTALL_MAN)/man5
+ $(INSTALL) -m 0644 and.priorities.5 $(DESTDIR)$(INSTALL_MAN)/man5
+
+simpleinstall: and and.init
+ strip and
+ mkdir -p $(DESTDIR)$(INSTALL_SBIN) $(DESTDIR)$(INSTALL_ETC)
+ mkdir -p $(DESTDIR)$(INSTALL_INITD)
+ mkdir -p $(DESTDIR)$(INSTALL_MAN)/man5 $(DESTDIR)$(INSTALL_MAN)/man8
+ cp and $(DESTDIR)$(INSTALL_SBIN)
+ test -e $(DESTDIR)$(INSTALL_ETC)/and.conf || \
+ cp and.conf $(DESTDIR)$(INSTALL_ETC)
+ test -e $(DESTDIR)$(INSTALL_ETC)/and.priorities || \
+ cp and.priorities $(DESTDIR)$(INSTALL_ETC)
+ifneq (${INITSCRIPT},) # on SysV only
+ cp $(INITSCRIPT) $(DESTDIR)$(INSTALL_INITD)/and
+endif
+ cp and.8 $(DESTDIR)$(INSTALL_MAN)/man8
+ cp and.conf.5 $(DESTDIR)$(INSTALL_MAN)/man5
+ cp and.priorities.5 $(DESTDIR)$(INSTALL_MAN)/man5
+
+uninstall:
+ rm -f $(DESTDIR)$(INSTALL_SBIN)/and
+ rm -f $(DESTDIR)$(INSTALL_INITD)/and
+ rm -f $(DESTDIR)$(INSTALL_ETC)/and.conf
+ rm -f $(DESTDIR)$(INSTALL_ETC)/and.priorities
+ rm -f $(DESTDIR)$(INSTALL_MAN)/man8/and.8
+ rm -f $(DESTDIR)$(INSTALL_MAN)/man5/and.conf.5
+ rm -f $(DESTDIR)$(INSTALL_MAN)/man5/and.priorities.5
+
+
+#
+# Clean up generated files.
+#
+clean:
+ rm -f *.o and and.init $(MANPAGES)
diff --git a/sensors/and/Makefile b/sensors/and/Makefile
index 5bf4f78052..fcdc16a2d5 100644
--- a/sensors/and/Makefile
+++ b/sensors/and/Makefile
@@ -6,6 +6,15 @@
#
+#
+# EMULAB PEOPLE LOOK HERE!
+#
+# This file is DEAD, edit the GNUmakefile.in if you want to change something.
+#
+# EMULAB PEOPLE LOOK HERE!
+#
+
+
#######################################################################
# Edit here to adapt to your system! #
#######################################################################
@@ -214,7 +223,7 @@ and.priorities.5: and.priorities.5.man
#
# Install and under $(PREFIX)/bin etc.
#
-install: and $(INITSCRIPT)
+install: and $(INITSCRIPT) doc
strip and
#-mkdir $(PREFIX)
-mkdir -p $(DESTDIR)$(INSTALL_SBIN)
diff --git a/sensors/and/and-OpenBSD.c b/sensors/and/and-OpenBSD.c
index fb4565041a..c26d65a6e2 100644
--- a/sensors/and/and-OpenBSD.c
+++ b/sensors/and/and-OpenBSD.c
@@ -1,7 +1,7 @@
/*
AND auto nice daemon - renice programs according to their CPU usage.
- Copyright (C) 1999-2001 Patrick Schemitz <schemitz@users.sourceforge.net>
+ Copyright (C) 1999-2004 Patrick Schemitz <schemitz@users.sourceforge.net>
http://and.sourceforge.net/
This program is free software; you can redistribute it and/or modify
@@ -94,21 +94,31 @@ struct and_procent *openbsd_getnext ()
strncpy(openbsd_proc.command,openbsd_pt[openbsd_next].kp_proc.p_comm,1023);
openbsd_proc.command[1023] = 0;
openbsd_proc.pid = openbsd_pt[openbsd_next].kp_proc.p_pid;
- openbsd_proc.ppid = openbsd_pt[openbsd_next].kp_proc.p_ppid; /* FIXME that correct? */
- openbsd_proc.nice = openbsd_pt[openbsd_next].kp_proc.p_nice-20;
+ //
openbsd_proc.uid = openbsd_pt[openbsd_next].kp_eproc.e_pcred.p_ruid;
openbsd_proc.gid = openbsd_pt[openbsd_next].kp_eproc.e_pcred.p_rgid;
+#if defined(__FreeBSD__)
+ openbsd_proc.ppid = openbsd_pt[openbsd_next].kp_eproc.e_ppid;
+ openbsd_proc.nice = openbsd_pt[openbsd_next].kp_proc.p_nice;
+ openbsd_proc.stime =
+ openbsd_pt[openbsd_next].kp_eproc.e_stats.p_start.tv_sec;
+ openbsd_proc.utime =
+ openbsd_pt[openbsd_next].kp_proc.p_runtime / (1000 * 1000);
+ openbsd_proc.ctime =
+ openbsd_pt[openbsd_next].kp_eproc.e_stats.p_cru.ru_utime.tv_sec +
+ openbsd_pt[openbsd_next].kp_eproc.e_stats.p_cru.ru_stime.tv_sec;
+#else
/* Adapted from top(1) port, as found in the misc@openbsd.org archive */
+ openbsd_proc.ppid = openbsd_pt[openbsd_next].kp_proc.p_ppid; /* FIXME that correct? */
+ openbsd_proc.nice = openbsd_pt[openbsd_next].kp_proc.p_nice-20;
openbsd_proc.utime = (openbsd_pt[openbsd_next].kp_proc.p_uticks +
openbsd_pt[openbsd_next].kp_proc.p_sticks +
openbsd_pt[openbsd_next].kp_proc.p_iticks)
/ openbsd_hz;
- /*
- printf("%-20s %5i %3i %i\n",openbsd_proc.command,openbsd_proc.pid,
- openbsd_proc.nice,openbsd_proc.utime);
- */
- and_printf(3, "OpenBSD: process %s pid: %d ppid: %d\n",
- openbsd_proc.command, openbsd_proc.pid, openbsd_proc.ppid);
+#endif
+ and_printf(3, "OpenBSD: process %s pid: %d ppid: %d cpu_secs: %d\n",
+ openbsd_proc.command, openbsd_proc.pid, openbsd_proc.ppid,
+ openbsd_proc.utime);
openbsd_next++;
return &openbsd_proc;
}
diff --git a/sensors/and/and-emulab.conf.in b/sensors/and/and-emulab.conf.in
new file mode 100644
index 0000000000..6d32fea1cb
--- /dev/null
+++ b/sensors/and/and-emulab.conf.in
@@ -0,0 +1,59 @@
+#
+# Emulab-ops configuration file for the auto nice daemon, /etc/and.conf
+#
+# Comments must have the # in the *first* column!
+#
+# Read and.conf(5) for details.
+#
+# 1999, 2000, 2004 Patrick Schemitz, schemitz@users.sourceforge.net
+#
+
+#
+# Nice level for jobs that are not in and.priorities.
+# 0 = do not renice.
+#
+defaultnice 0
+
+#
+# Time interval between renice cycles, in seconds. Default is
+# 60 seconds.
+#
+interval 120
+
+#
+# Ranges for the nice levels. Jobs with less than lv1time seconds
+# CPU time are not reniced; jobs between lv1time and lv2time seconds
+# are reniced to the first level in an.priorities; jobs between
+# lv2time and lv3time seconds to the second level; jobs with more
+# than lv3time seconds are reniced to the third level.
+#
+lv1time 90
+lv2time 1800
+lv2cmd mail -s "@THISHOMEBASE@: CPU Hog - $AND_COMMAND (30 min.)" @TBOPSEMAIL@
+lv3time 7200
+lv3cmd mail -s "@THISHOMEBASE@: CPU Hog - $AND_COMMAND (2 hrs.)" @TBOPSEMAIL@
+
+#
+# Strategy for picking the right priority entry for a user/group/job
+# triple. The strategy is a permutation of "cgu", "c"ommand, "g"roup,
+# "u"ser. The order specifies the affinity of the priority lookup
+# method. "cug" means an exact match of the command has priority
+# over an exact match of the user or group. See the documentation
+# for more details.
+#
+affinity cpug
+
+#
+# Minimum user/group id to be even considered for renicing. Processes
+# with lower user/group id are ignored. This does not affect root
+# (user id 0), which is never, ever reniced.
+#
+minuid 1
+mingid 1
+
+# NOTE: Emulab user IDs start at 10000, so if you want to exempt the
+# core developers, comment out the above lines and uncomment the ones
+# below.
+#
+# minuid 10000
+# mingid 6000
diff --git a/sensors/and/and-emulab.priorities b/sensors/and/and-emulab.priorities
new file mode 100644
index 0000000000..217f627481
--- /dev/null
+++ b/sensors/and/and-emulab.priorities
@@ -0,0 +1,66 @@
+#
+# Sample priority database for the auto-nice daemon, /etc/and.priorities
+#
+# Comments must have the # in the _first_ column!
+#
+# File format:
+# user group job parent nice1 nice2 nice3
+# - user: user name or user id or * for all
+# - group: group name or group id or * for all
+# - job: executable (without path; may be a regexp) or * for all
+# - parent: keyword "parent=" or "ancestor=" followed by the
+# executable (without path; may be a regexp), or * for all
+# - nice1, nice2, nice3: nice levels for CPU usage ranges.
+#
+# At least one of user or group must be an asterisk *.
+#
+# After /etc/and.conf:lv1time seconds matching jobs are niced to nice1,
+# after /etc/and.conf:lv2time to nice2 and after /etc/and.conf:lv3time
+# to nice3.
+#
+# Read and.priorities(5) for details.
+#
+# 1999, 2000, 2004 Patrick Schemitz, schemitz@users.sourceforge.net
+#
+
+#
+# Philosophy:
+#
+# Knock everyone except for the well known daemons, ops is not a
+# general purpose machine, that is what the experimental nodes are for.
+# If you make any changes, note that a perceptive user might rename
+# his jobs' executables to obtain higher privilegue. So the values
+# must be sensible to that.
+#
+
+#
+# Default entry -- moderate renicing at first, drop them further down
+# after that.
+#
+* * * * 5 15 19
+
+#
+# daemon entry -- for the portmapper and such.
+#
+daemon * * * 0 0 0
+
+#
+# The main elvind server.
+#
+nobody * elvind * 0 0 0
+
+#
+# Sendmail stuff
+#
+smmsp * * * 0 0 0
+
+#
+# The event scheduler, not sure about this, but for long running
+# experiments the scheduler ends up accumulating a lot of time.
+#
+# Note that we only do this for the mainline one, development
+# versions are not exempt.
+#
+* * event-sched * 0 0 0
+
+* * sshd * 0 5 15
diff --git a/sensors/and/and.8.man b/sensors/and/and.8.man
index 418177615b..508a7effff 100644
--- a/sensors/and/and.8.man
+++ b/sensors/and/and.8.man
@@ -129,6 +129,12 @@ Both files have their own manual pages.
Contains logging and status information for debugging purposes.
Used in test mode only.
+.TP 0.5i
+.B /var/run/and.pid
+Process identifier file. Stores the PID for the currently running
+.I and
+process. This file is not written in test mode.
+
.SH "SEE ALSO"
.BR and.conf (5),
.BR and.priorities (5),
diff --git a/sensors/and/and.c b/sensors/and/and.c
index 54933b785a..049cbe2272 100644
--- a/sensors/and/and.c
+++ b/sensors/and/and.c
@@ -45,9 +45,15 @@
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/types.h>
+#include <sys/wait.h>
#include <signal.h>
#include <regex.h>
#include <values.h>
+#include <errno.h>
+#include <pwd.h>
+#include <grp.h>
+#include <fcntl.h>
+#include <sys/stat.h>
#define DEBUG 0
@@ -60,6 +66,11 @@
#ifdef __GNUC__
int vsnprintf (char *str, size_t n, const char *format, va_list ap);
#define HAVE_VSNPRINTF
+int snprintf (char *str, size_t n, const char *format, ...);
+#define HAVE_SNPRINTF
+#define and_snprintf snprintf
+#else
+#define and_snprintf sprintf
#endif
#include "and.h"
@@ -86,6 +97,8 @@ int vsnprintf (char *str, size_t n, const char *format, va_list ap);
#define DEFAULT_DATABASE_FILE "/etc/and.priorities"
#endif
+#define AND_PIDFILE "/var/run/and.pid"
+
#ifndef AND_VERSION
#define AND_VERSION "1.0.7 or above (not compiled in)"
#endif
@@ -165,10 +178,12 @@ struct {
bool lock_interval;
unsigned interval;
unsigned time_mark [3];
+ char *cmd[3];
char affinity [5];
int weight [PRI_N];
int min_uid;
int min_gid;
+ char *cmd_user;
} and_config;
@@ -191,6 +206,10 @@ void set_defaults (int argc, char **argv)
and_config.min_gid = 0;
gethostname(and_config.hostname,511);
and_config.hostname[511] = 0;
+ and_config.cmd[0] = "";
+ and_config.cmd[1] = "";
+ and_config.cmd[2] = "";
+ and_config.cmd_user = "nobody";
}
@@ -286,8 +305,11 @@ void print_config ()
"default nicelevel: %2i\n"
"interval [sec]: %3u\n"
"level 0 from [sec]: %3u\n"
+ "level 0 cmd: %s\n"
"level 1 from [sec]: %3u\n"
+ "level 1 cmd: %s\n"
"level 2 from [sec]: %3u\n"
+ "level 2 cmd: %s\n"
"minimum uid: %i\n"
"minimum gid: %i\n"
"affinity: %s\n"
@@ -299,8 +321,9 @@ void print_config ()
(and_config.test?"just checkin'":"I'm serious."),
and_config.verbose,
and_config.nice_default, and_config.interval,
- and_config.time_mark[0], and_config.time_mark[1],
- and_config.time_mark[2],
+ and_config.time_mark[0], and_config.cmd[0],
+ and_config.time_mark[1], and_config.cmd[1],
+ and_config.time_mark[2], and_config.cmd[2],
and_config.min_uid, and_config.min_gid,
and_config.affinity,
and_config.weight[PRI_U], and_config.weight[PRI_G],
@@ -603,6 +626,10 @@ void read_config ()
and_printf(0,"Configuration file line %i has invalid value for lv1time: %s.\n",
line, value);
}
+ } else if (strcmp(param,"lv1cmd")==0) {
+ if (buffer[strlen(buffer) - 1] == '\n')
+ buffer[strlen(buffer) - 1] = '\0';
+ and_config.cmd[0] = strdup(&buffer[strlen(param) + 1]);
} else if (strcmp(param,"lv2time")==0) {
if (uval < UINT_MAX)
and_config.time_mark[1] = uval;
@@ -611,6 +638,10 @@ void read_config ()
and_printf(0,"Configuration file line %i has invalid value for lv2time: %s.\n",
line, value);
}
+ } else if (strcmp(param,"lv2cmd")==0) {
+ if (buffer[strlen(buffer) - 1] == '\n')
+ buffer[strlen(buffer) - 1] = '\0';
+ and_config.cmd[1] = strdup(&buffer[strlen(param) + 1]);
} else if (strcmp(param,"lv3time")==0) {
if (uval < UINT_MAX)
and_config.time_mark[2] = uval;
@@ -619,6 +650,18 @@ void read_config ()
and_printf(0,"Configuration file line %i has invalid value for lv3time: %s.\n",
line, value);
}
+ } else if (strcmp(param,"lv3cmd")==0) {
+ if (buffer[strlen(buffer) - 1] == '\n')
+ buffer[strlen(buffer) - 1] = '\0';
+ and_config.cmd[2] = strdup(&buffer[strlen(param) + 1]);
+ } else if (strcmp(param,"cmduser")==0) {
+ if (strlen(value) > 0)
+ and_config.cmd_user = value;
+ else {
+ ++bad;
+ and_printf(0,"Configuration file line %i has empty value for cmduser.\n",
+ line);
+ }
} else if (strcmp(param,"affinity")==0) {
bad_f = -1;
u = g = c = p = 0;
@@ -679,11 +722,12 @@ void read_config ()
/* Compute new nice level for given command/uid/gid/utime */
-int and_getnice (int uid, int gid, char *command, struct and_procent *parent, unsigned cpu_seconds)
+int and_getnice (int uid, int gid, char *command, struct and_procent *parent, unsigned cpu_seconds, char **cmd_out)
{
int i, level, entry, exact = -1, last;
struct and_procent *par;
int exactness [PRI_MAXENTRIES];
+ *cmd_out = NULL;
if (!command) {
and_printf(0,"Process without command string encountered. Aborting.\n");
abort();
@@ -769,13 +813,130 @@ int and_getnice (int uid, int gid, char *command, struct and_procent *parent, un
while (level >= 0 && and_config.time_mark[level] > cpu_seconds) {
--level;
}
- and_printf(2,"command=%s (%i,%i,%s) hit on entry=%i, exactness=%i, level=%i.\n",
+ and_printf(2,"command=%s (%i,%i,%s) hit on entry=%i, exactness=%i, level=%i, cs=%i.\n",
command, uid, gid, (parent!=NULL?parent->command:"(orphan)"),
- entry, exact, level);
+ entry, exact, level, cpu_seconds);
+ *cmd_out = (level >= 0) ? and_config.cmd[level] : NULL;
return (level >= 0 ? and_db.entry[entry].nl[level] : 0);
}
+int and_exec (char *cmd, struct and_procent *current, int newnice)
+{
+ int rc, retval = -1;
+
+ assert(cmd != NULL);
+ assert(strlen(cmd) > 0);
+ assert(newnice != 0);
+
+ switch (rc = fork()) {
+ case 0:
+ /* child */
+ if (getuid() == 0) {
+ char *user_name = "(unknown)", *group_name = "(unknown)";
+ int exit_value = EXIT_SUCCESS;
+ struct passwd *pw;
+ struct group *grp;
+ char buffer[2048];
+ FILE *file;
+
+ setenv("HOME", "", 1);
+ setenv("USER", and_config.cmd_user, 1);
+
+ if ((pw = getpwnam(and_config.cmd_user)) == NULL) {
+ syslog(LOG_ERR,"unknown command user: %s",and_config.cmd_user);
+ exit(1);
+ }
+
+ if (setgid(pw->pw_gid) ||
+ initgroups(and_config.cmd_user, pw->pw_gid) ||
+ setuid(pw->pw_uid)) {
+ syslog(LOG_ERR,"unable to drop privileges: %s",strerror(errno));
+ exit(1);
+ }
+
+ setenv("AND_HOST", and_config.hostname, 1);
+
+ if ((pw = getpwuid(current->uid)) != NULL)
+ user_name = pw->pw_name;
+ if ((grp = getgrgid(current->gid)) != NULL)
+ group_name = grp->gr_name;
+
+ sprintf(buffer, "%d", current->pid);
+ setenv("AND_PID", buffer, 1);
+ sprintf(buffer, "%d", current->ppid);
+ setenv("AND_PPID", buffer, 1);
+ setenv("AND_USER", user_name, 1);
+ setenv("AND_GROUP", group_name, 1);
+ setenv("AND_COMMAND", current->command, 1);
+
+ if ((file = popen(cmd, "w")) != NULL) {
+ char time_string[256];
+ time_t current_time;
+
+ current_time = time(NULL);
+ strftime(time_string, sizeof(time_string),
+ "%H:%M",
+ localtime(¤t_time));
+
+ and_snprintf(buffer, sizeof(buffer),
+ "\n"
+ "[This is an automated message from the auto nice daemon (AND)]\n"
+ "\n");
+ buffer[sizeof(buffer) - 1] = '\0';
+ fwrite(buffer, 1, strlen(buffer), file);
+ and_snprintf(buffer, sizeof(buffer),
+ "A CPU hog has been detected at %s hours on %s:\n"
+ "\n"
+ " pid\t\t%d\n"
+ " ppid\t\t%d\n"
+ " uid\t\t%d %s\n"
+ " gid\t\t%d %s\n"
+ " old nice\t%d\n"
+ " CPU seconds\t%d\n"
+ " command\t%s\n"
+ " start time\t%s"
+ "\n"
+ "Action taken:\n"
+ " %s %d\n",
+ time_string,
+ and_config.hostname,
+ current->pid,
+ current->ppid,
+ current->uid, user_name,
+ current->gid, group_name,
+ current->nice,
+ current->utime,
+ current->command,
+ ctime(¤t->stime),
+ (newnice > 0) ? "Changed nice to" : "Signalled process with",
+ (newnice > 0) ? newnice : -newnice);
+ buffer[sizeof(buffer) - 1] = '\0';
+ fwrite(buffer, 1, strlen(buffer), file);
+ if (pclose(file) == -1) {
+ syslog(LOG_ERR,"error while executing: %s",cmd);
+
+ exit_value = EXIT_FAILURE;
+ }
+ }
+ else {
+ exit_value = EXIT_FAILURE;
+ }
+
+ exit(exit_value);
+ }
+ break;
+ case -1:
+ break;
+ default:
+ /* parent */
+ retval = 0;
+ break;
+ }
+
+ return (retval);
+}
+
/**********************************************************************
@@ -812,10 +973,14 @@ struct and_procent* and_find_proc (struct and_procent *head, int ppid)
void and_loop ()
{
struct and_procent *head, *current, *new, *proc;
+ int childstatus;
int newnice;
int njobs = 0;
assert(and_getfirst != NULL);
assert(and_getnext != NULL);
+ while (wait3(&childstatus,WNOHANG,NULL) > 0) {
+ // reaped child...
+ }
head = NULL;
current = NULL;
proc = and_getfirst();
@@ -843,9 +1008,10 @@ void and_loop ()
}
current = head;
while (current != NULL) {
+ char *cmd;
njobs++;
newnice = and_getnice(current->uid,current->gid,current->command,
- current->parent,current->utime);
+ current->parent,current->utime,&cmd);
if (current->uid != 0) {
if (newnice) {
if (newnice > 0) {
@@ -858,6 +1024,9 @@ void and_loop ()
current->command);
setpriority(PRIO_PROCESS,current->pid,newnice);
}
+ if ((cmd != NULL) && (strlen(cmd) > 0)) {
+ and_exec(cmd, current, newnice);
+ }
}
} else {
if (and_config.test)
@@ -868,6 +1037,9 @@ void and_loop ()
current->command);
kill(current->pid,-newnice);
}
+ if ((cmd != NULL) && (strlen(cmd) > 0)) {
+ and_exec(cmd, current, newnice);
+ }
}
}
}
@@ -945,6 +1117,7 @@ void and_getopt (int argc, char** argv)
static int g_reload_conf;
+static bool g_looping = true;
void and_trigger_readconf (int sig)
@@ -952,6 +1125,10 @@ void and_trigger_readconf (int sig)
g_reload_conf = (sig == SIGHUP);
}
+void and_trigger_stoplooping (int sig)
+{
+ g_looping = false;
+}
void and_readconf ()
{
@@ -961,15 +1138,17 @@ void and_readconf ()
g_reload_conf = 0;
}
-
void and_worker ()
{
read_config();
read_priorities();
+ signal(SIGTERM,and_trigger_stoplooping);
+ signal(SIGINT,and_trigger_stoplooping);
+ signal(SIGQUIT,and_trigger_stoplooping);
signal(SIGHUP,and_trigger_readconf);
and_printf(0,"AND ready.\n");
g_reload_conf = 0;
- while (1) {
+ while (g_looping) {
if (g_reload_conf) {
and_readconf();
}
@@ -986,7 +1165,24 @@ int and_main (int argc, char** argv)
if (and_config.test) {
and_worker();
} else {
- if (fork() == 0) and_worker();
+ if (daemon(0,0) < 0) {
+ perror("Unable to daemonize");
+ }
+ else {
+ char pidbuf[32];
+ int pfd;
+
+ if ((pfd = open(AND_PIDFILE, O_EXCL | O_CREAT | O_WRONLY)) < 0) {
+ and_printf(0,"Could not create pid file: %s\n",AND_PIDFILE);
+ exit(1);
+ }
+ fchmod(pfd, S_IRUSR | S_IRGRP | S_IROTH);
+ sprintf(pidbuf,"%d",getpid());
+ write(pfd,pidbuf,strlen(pidbuf));
+ close(pfd);
+ and_worker();
+ unlink(AND_PIDFILE);
+ }
}
return 0;
}
diff --git a/sensors/and/and.conf.5.man b/sensors/and/and.conf.5.man
index ac57573488..bbd6bcc007 100644
--- a/sensors/and/and.conf.5.man
+++ b/sensors/and/and.conf.5.man
@@ -90,6 +90,56 @@ Defaults are
.I 3600
seconds.
+.TP 0.5i
+.B lv1cmd, lv2cmd, lv3cmd
+Commands for the nice levels. Jobs that reach a new level, and have their
+priorities changed, will cause the corresponding command to be executed. The
+commands are executed using
+.B popen(3)
+and an informational message describing the offending job is piped in. In
+addition, the following environment variables will be set, so you can reference
+them in the given command-line:
+
+.RS
+.TP 0.5i
+.B AND_HOST
+The host where
+.I and
+is running.
+
+.TP 0.5i
+.B AND_PID
+The process ID of the offending job.
+
+.TP 0.5i
+.B AND_PPID
+The parent process ID of the offending job.
+
+.TP 0.5i
+.B AND_USER
+The user name the offending job is running with.
+
+.TP 0.5i
+.B AND_GROUP
+The group name the offending job is running with.
+
+.TP 0.5i
+.B AND_COMMAND
+The command name of the offending job.
+.RE
+
+.RS
+Normally, the command is executed as "nobody", or the user specified by the
+.I cmduser
+setting. While this environment is limiting, it is enough to send mail or
+perform some other simple communication.
+.RE
+
+.TP 0.5i
+.B cmduser
+The user name to use when running level commands. The default value for
+this is "nobody", which should be sufficient for most uses.
+
.TP 0.5i
.B minuid, mingid
Minimum user id and group id to be considered for renicing.
@@ -161,6 +211,10 @@ lv2time 1800
.br
lv3time 3600
.br
+# Get someone involved if they reach level 3
+.br
+lv3cmd mail -s "CPU Hog - $AND_COMMAND" bofh@foobar.com
+.br
# Hosts foo, bar, baz are terminals and must
.br
# be more responsive, so earlier renice.
diff --git a/sensors/and/and.h b/sensors/and/and.h
index 7b1108f462..c6b0298b35 100644
--- a/sensors/and/and.h
+++ b/sensors/and/and.h
@@ -46,10 +46,12 @@ struct and_procent {
int uid;
int gid;
int nice;
- unsigned utime;
+ time_t stime; // process start time in seconds.
+ unsigned utime; // CPU time in seconds.
+ unsigned ctime; // reaped child CPU time in seconds.
char command [1024];
/* to be filled by and.c: */
- struct and_procent *parent;
+ struct and_procent *parent;
struct and_procent *next;
};
diff --git a/sensors/and/and.startup b/sensors/and/and.startup
index ea4d968b51..ba77aaddaa 100644
--- a/sensors/and/and.startup
+++ b/sensors/and/and.startup
@@ -15,14 +15,15 @@
#
AND_FLAGS=""
-test -r /etc/rc.config && . /etc/rc.config
+
+# test -r /etc/rc.config && . /etc/rc.config
case "$1" in
start)
echo -n "Starting auto nice daemon:"
- INSTALL_SBIN/and $AND_FLAGS >&/dev/null
- ps axuw | grep -v grep | grep INSTALL_SBIN/and >/dev/null
- if [ $? = 0 ]; then
+ INSTALL_SBIN/and $AND_FLAGS > /dev/null 2>&1
+ ps axuw | grep -v grep | grep INSTALL_SBIN/and > /dev/null
+ if test $? -eq 0; then
echo " done"
exit 0
else
@@ -32,8 +33,23 @@ case "$1" in
;;
stop)
echo -n "Shutting down auto nice daemon:"
- uname | grep OSF1 >/dev/null || killall INSTALL_SBIN/and
- echo " done"
+ ps axuw | grep -v grep | grep INSTALL_SBIN/and > /dev/null
+ if test $? -eq 0; then
+ kill `cat /var/run/and.pid`
+ echo " done"
+ else
+ echo " already stopped"
+ fi
+ exit 0
+ ;;
+ reconfig)
+ echo -n "Reconfiguring auto nice daemon:"
+ if test -r "/var/run/and.pid"; then
+ kill -HUP `cat /var/run/and.pid`
+ echo " done"
+ else
+ echo " no process"
+ fi
exit 0
;;
restart)
@@ -42,8 +58,8 @@ case "$1" in
;;
status)
echo -n "Checking for auto nice daemon: "
- ps axuw | grep -v grep | grep INSTALL_SBIN/and >/dev/null
- if [ $? = 0 ]; then
+ ps axuw | grep -v grep | grep INSTALL_SBIN/and > /dev/null
+ if test $? -eq 0; then
echo "running"
exit 0
else
@@ -52,6 +68,6 @@ case "$1" in
fi
;;
*)
- echo "Usage: $0 {start|stop|status|restart}"
+ echo "Usage: $0 {start|stop|status|reconfig|restart}"
exit 1
esac
--
GitLab