1 collectd - System information collection daemon
2 =================================================
8 collectd is a small daemon which collects system information periodically
9 and provides mechanisms to store and monitor the values in a variety of
16 * collectd is able to collect the following data:
19 Apache server utilization: Number of bytes transfered, number of
20 requests handled and detailed scoreboard statistics
23 APC UPS Daemon: UPS charge, load, input/output/battery voltage, etc.
26 Sensors in Macs running Mac OS X / Darwin: Temperature, fanspeed and
30 Batterycharge, -current and voltage of ACPI and PMU based laptop
34 CPU utilization: Time spent in the system, user, nice, idle, and related
38 CPU frequency (For laptops with speed step or a similar technology)
41 Mountpoint usage (Basically the values `df(1)' delivers)
44 Disk utilization: Sectors read/written, number of read/write actions,
45 average time an IO-operation took to complete.
48 DNS traffic: Query types, response codes, opcodes and traffic/octets
52 Email statistics: Count, traffic, spam scores and checks.
53 See collectd-email(5).
56 Amount of entropy available to the system.
59 Values gathered by a custom program or script.
63 Harddisk temperatures using hddtempd.
66 Interface traffic: Number of octets, packets and errors for each
70 Iptables' counters: Number of bytes that were matched by a certain
74 IPVS connection statistics (number of connections, octets and packets
75 for each service and destination).
76 See http://www.linuxvirtualserver.org/software/index.html.
79 IRQ counters: Frequency in which certain interrupts occur.
82 System load average over the last 1, 5 and 15 minutes.
85 CPU, disk and network I/O statistics from virtual machines.
88 Motherboard sensors: temperature, fanspeed and voltage information,
92 Statistics of the memcached distributed caching system.
93 <http://www.danga.com/memcached/>
96 Memory utilization: Memory occupied by running processes, page cache,
97 buffer cache and free.
100 Information provided by serial multimeters, such as the `Metex
104 MySQL server statistics: Commands issued, handlers triggered, thread
105 usage, query cache utilization and traffic/octets sent and received.
108 Very detailed Linux network interface and routing statistics. You can get
109 (detailed) information on interfaces, qdiscs, classes, and, if you can
110 make use of it, filters.
113 Receive values that were collected by other hosts. Large setups will
114 want to collect the data on one dedicated machine, and this is the
115 plugin of choice for that.
118 NFS Procedures: Which NFS command were called how often. Only NFSv2 and
122 Collects statistics from `nginx' (speak: engine X), a HTTP and mail
126 NTP daemon statistics: Local clock drift, offset to peers, etc.
129 Network UPS tools: UPS current, voltage, power, charge, utilisation,
130 temperature, etc. See upsd(8).
133 The perl plugin implements a Perl-interpreter into collectd. You can
134 write your own plugins in Perl and return arbitrary values using this
135 API. See collectd-perl(5).
138 Network latency: Time to reach the default gateway or another given
142 Process counts: Number of running, sleeping, zombie, ... processes.
145 System sensors, accessed using lm_sensors: Voltages, temperatures and
149 RX and TX of serial interfaces. Linux only; needs root privileges.
152 Read values from SNMP (Simple Network Management Protocol) enabled
153 network devices such as switches, routers, thermometers, rack monitoring
154 servers, etc. See collectd-snmp(5).
157 Pages swapped out onto harddisk or whatever is called `swap' by the OS..
160 Follows (tails) logfiles, parses them by lines and submits matched
164 Bytes and operations read and written on tape devices. Solaris only.
167 Number of TCP connections to specific local and remote ports.
170 Users currently logged in.
173 Virtual memory statistics, e. g. the number of page-ins/-outs or the
174 number of pagefaults.
177 System resources used by Linux VServers.
178 See <http://linux-vserver.org/>.
181 Link quality of wireless cards. Linux only.
184 Bitrate and frequency of music played with XMMS.
186 * Output can be written or send to various destinations by the following
190 Write to comma separated values (CSV) files. This needs lots of
191 diskspace but is extremely portable and can be analysed with almost
192 every program that can analyse anything. Even Microsoft's Excel..
195 Send the data to a remote host to save the data somehow. This is useful
196 for large setups where the data should be saved by a dedicated machine.
199 Of course the values are propagated to plugins written in Perl, too, so
200 you can easily do weird stuff with the plugins we didn't dare think of
201 ;) See collectd-perl(5).
204 Output to round-robin-database (RRD) files using librrd. See rrdtool(1).
205 This is likely the most popular destination for such values. Since
206 updates to RRD-files are somewhat expensive this plugin can cache
207 updates to the files and write a bunch of updates at once, which lessens
211 One can query the values from the unixsock plugin whenever they're
212 needed. Please read collectd-unixsock(5) for a description on how that's
215 * Logging is, as everything in collectd, provided by plugins. The following
216 plugins keep up informed about what's going on:
219 Writes logmessages to a file or STDOUT/STDERR.
222 Log messages are propagated to plugins written in Perl as well.
223 See collectd-perl(5).
226 Logs to the standard UNIX logging mechanism, syslog.
228 * Notifications can be handled by the following plugins:
231 Execute a program or script to handle the notification.
232 See collectd-exec(5).
235 Writes the notification message to a file or STDOUT/STDERR.
238 Send the notification to a remote host to handle it somehow.
241 Notifications are propagated to plugins written in Perl as well.
242 See collectd-perl(5).
244 * Miscellaneous plugins:
247 Sets the hostname to an unique identifier. This is meant for setups
248 where each client may migrate to another physical host, possibly going
249 through one or more name changes in the process.
251 * Performance: Since collectd is running as a daemon it doesn't spend much
252 time starting up again and again. With the exception of the exec plugin no
253 processes are forked. Caching in output plugins, such as the rrdtool and
254 network plugins, makes sure your resources are used efficiently. Also,
255 since collectd is programmed multithreaded it benefits from hyperthreading
256 and multicore processors and makes sure that the daemon isn't idle if only
257 one plugins waits for an IO-operation to complete.
259 * Once set up, hardly any maintenance is necessary. Setup is kept as easy
260 as possible and the default values should be okay for most users.
266 * collectd's configuration file can be found at `sysconfdir'/collectd.conf.
267 Run `collectd -h' for a list of builtin defaults. See `collectd.conf(5)'
268 for a list of options and a syntax description.
270 * When the `csv' or `rrdtool' plugins are loaded they'll write the values to
271 files. The usual place for these files is beneath `/var/lib/collectd'.
273 * When using some of the plugins, collectd needs to run as user root, since
274 only root can do certain things, such as craft ICMP packages needed to ping
275 other hosts. collectd should NOT be installed setuid root since it can be
276 used to overwrite valuable files!
278 * Sample scripts to generate graphs reside in `contrib/' in the source
279 package or somewhere near `/usr/share/doc/collectd' in most distributions.
280 Please be aware that those script are meant as a starting point for your
281 own experiments.. Some of them require the `RRDs' Perl module.
282 (`librrds-perl' on Debian) If you have written a more sophisticated
283 solution please share it with us.
285 * The RRAs of the automatically created RRD files depend on the `step'
286 and `heartbeat' settings given. If change these settings you may need to
287 re-create the files, losing all data. Please be aware of that when changing
288 the values and read the rrdtool(1) manpage thoroughly.
291 collectd and chkrootkit
292 -----------------------
294 If you are using the `dns' plugin chkrootkit(1) will report collectd as a
295 packet sniffer ("<iface>: PACKET SNIFFER(/usr/sbin/collectd[<pid>])"). The
296 plugin captures all UDP packets on port 53 to analyze the DNS traffic. In
297 this case, collectd is a legitimate sniffer and the report should be
298 considered to be a false positive. However, you might want to check that
299 this really is collectd and not some other, illegitimate sniffer.
305 To compile collectd from source you will need:
307 * Usual suspects: C compiler, linker, preprocessor, make, ...
309 * A POSIX-threads (pthread) implementation.
310 Since gathering some statistics is slow (network connections, slow devices,
311 etc) the collectd is parallelized. The POSIX threads interface is being
312 used and should be found in various implementations for hopefully all
316 If you want to use the `apache' and/or `nginx' plugins.
319 If present, the uuid plugin will check for UUID from HAL.
322 For querying iptables counters.
324 * libmysqlclient (optional)
325 Unsurprisingly used by the `mysql' plugin.
327 * libnetlink (optional)
328 Used, obviously, for the `netlink' plugin.
330 * libnetsnmp (optional)
331 For the `snmp' plugin.
333 * liboping (optional, if not found a version shipped with this distribution
335 Used by the `ping' plugin to send and receive ICMP packets.
338 Used to capture packets by the `dns' plugin.
341 Obviously used by the `perl' plugin. The library has to be compiled with
342 ithread support (introduced in Perl 5.6.0).
344 * librrd (optional; headers and library; rrdtool 1.0 and 1.2 both work fine)
345 If built without `librrd' the resulting binary will be `client only', i.e.
346 will send its values via multicast and not create any RRD files itself.
347 Alternatively you can chose to write CSV-files (Comma Separated Values)
350 * libsensors (optional)
351 To read from `lm_sensors', see the `sensors' plugin.
353 * libstatgrab may be used to collect statistics on systems other than Linux
354 and/or Solaris. Note that CPU- and disk-statistics, while being provided
355 by this library, are not supported in collectd right now..
356 <http://www.i-scream.org/libstatgrab/>
358 * libupsclient/nut (optional)
359 For the `nut' plugin which queries nut's `upsd'.
363 * librt, libsocket, libkstat, libdevinfo
364 Various standard Solaris libraries which provide system functions.
366 * CoreFoundation.framework and IOKit.framework
367 For compiling on Darwin in general and the `apple_sensors' plugin in
371 Collect statistics from virtual machines.
374 Parse XML data provided by libvirt.
377 Configuring / Compiling / Installing
378 ------------------------------------
380 To configure, build and install collectd with the default settings, run
381 `./configure && make && make install'. For detailed, generic instructions
382 see INSTALL. For a complete list of configure options and their description,
383 run `./configure --help'.
385 By default, the configure script will check for all build dependencies and
386 disable all plugins whose requirements cannot be fulfilled (any other plugin
387 will be enabled). To enable a plugin, install missing dependencies (see
388 section `Prerequisites' above) and rerun `configure'. If you specify the
389 `--enable-<plugin>' configure option, you can force the plugin to be built.
390 This will most likely fail though unless you're working in a very unusual
391 setup and you really know what you're doing.
393 By default, collectd will be installed into `/opt/collectd'. You can adjust
394 this setting by specifying the `--prefix' configure option - see INSTALL for
395 details. If you pass DESTDIR=<path> to `make install', <path> will be
396 prefixed to all installation directories. This might be useful when creating
397 packages for collectd.
403 To compile correctly collectd needs to be able to initialize static
404 variables to NAN (Not A Number). Some C libraries, especially the GNU
405 libc, have a problem with that.
407 Luckily, with GCC it's possible to work around that problem: One can define
408 NAN as being (0.0 / 0.0) and `isnan' as `f != f'. However, to test this
409 ``implementation'' the configure script needs to compile and run a short
410 test program. Obviously running a test program when doing a cross-
411 compilation is, well, challenging.
413 If you run into this problem, you can use the `--with-nan-emulation'
414 configure option to force the use of this implementation. We can't promise
415 that the compiled binary actually behaves as it should, but since NANs
416 are likely never passed to the libm you have a good chance to be lucky.
422 For questions, bugreports, development information and basically all other
423 concerns please send an email to collectd's mailinglist at
424 <collectd at verplant.org>.
426 For live discussion and more personal contact visit us in IRC, we're in
427 channel #collectd on freenode.
433 Florian octo Forster <octo at verplant.org>,
434 Sebastian tokkee Harl <sh at tokkee.org>,
435 and many contributors (see `AUTHORS').
437 Please send bugreports and patches to the mailinglist, see `Contact' above.