3 rrdcreate - Set up a new Round Robin Database
7 B<rrdtool> B<create> I<filename>
8 S<[B<--start>|B<-b> I<start time>]>
9 S<[B<--step>|B<-s> I<step>]>
10 S<[B<DS:>I<ds-name>B<:>I<DST>B<:>I<dst arguments>]>
11 S<[B<RRA:>I<CF>B<:>I<cf arguments>]>
15 The create function of RRDtool lets you set up new Round Robin
16 Database (B<RRD>) files. The file is created at its final, full size
17 and filled with I<*UNKNOWN*> data.
23 The name of the B<RRD> you want to create. B<RRD> files should end
24 with the extension F<.rrd>. However, B<RRDtool> will accept any
27 =item B<--start>|B<-b> I<start time> (default: now - 10s)
29 Specifies the time in seconds since 1970-01-01 UTC when the first
30 value should be added to the B<RRD>. B<RRDtool> will not accept
31 any data timed before or at the time specified.
33 See also AT-STYLE TIME SPECIFICATION section in the
34 I<rrdfetch> documentation for other ways to specify time.
36 =item B<--step>|B<-s> I<step> (default: 300 seconds)
38 Specifies the base interval in seconds with which data will be fed
41 =item B<DS:>I<ds-name>B<:>I<DST>B<:>I<dst arguments>
43 A single B<RRD> can accept input from several data sources (B<DS>),
44 for example incoming and outgoing traffic on a specific communication
45 line. With the B<DS> configuration option you must define some basic
46 properties of each data source you want to store in the B<RRD>.
48 I<ds-name> is the name you will use to reference this particular data
49 source from an B<RRD>. A I<ds-name> must be 1 to 19 characters long in
50 the characters [a-zA-Z0-9_].
52 I<DST> defines the Data Source Type. The remaining arguments of a
53 data source entry depend on the data source type. For GAUGE, COUNTER,
54 DERIVE, and ABSOLUTE the format for a data source entry is:
56 B<DS:>I<ds-name>B<:>I<GAUGE | COUNTER | DERIVE | ABSOLUTE>B<:>I<heartbeat>B<:>I<min>B<:>I<max>
58 For COMPUTE data sources, the format is:
60 B<DS:>I<ds-name>B<:>I<COMPUTE>B<:>I<rpn-expression>
62 In order to decide which data source type to use, review the
63 definitions that follow. Also consult the section on "HOW TO MEASURE"
70 is for things like temperatures or number of people in a room or the
71 value of a RedHat share.
75 is for continuous incrementing counters like the ifInOctets counter in
76 a router. The B<COUNTER> data source assumes that the counter never
77 decreases, except when a counter overflows. The update function takes
78 the overflow into account. The counter is stored as a per-second
79 rate. When the counter overflows, RRDtool checks if the overflow
80 happened at the 32bit or 64bit border and acts accordingly by adding
81 an appropriate value to the result.
85 will store the derivative of the line going from the last to the
86 current value of the data source. This can be useful for gauges, for
87 example, to measure the rate of people entering or leaving a
88 room. Internally, derive works exactly like COUNTER but without
89 overflow checks. So if your counter does not reset at 32 or 64 bit you
90 might want to use DERIVE and combine it with a MIN value of 0.
94 =item NOTE on COUNTER vs DERIVE
96 by Don Baarda E<lt>don.baarda@baesystems.comE<gt>
98 If you cannot tolerate ever mistaking the occasional counter reset for a
99 legitimate counter wrap, and would prefer "Unknowns" for all legitimate
100 counter wraps and resets, always use DERIVE with min=0. Otherwise, using
101 COUNTER with a suitable max will return correct values for all legitimate
102 counter wraps, mark some counter resets as "Unknown", but can mistake some
103 counter resets for a legitimate counter wrap.
105 For a 5 minute step and 32-bit counter, the probability of mistaking a
106 counter reset for a legitimate wrap is arguably about 0.8% per 1Mbps of
107 maximum bandwidth. Note that this equates to 80% for 100Mbps interfaces, so
108 for high bandwidth interfaces and a 32bit counter, DERIVE with min=0 is
109 probably preferable. If you are using a 64bit counter, just about any max
110 setting will eliminate the possibility of mistaking a reset for a counter
117 is for counters which get reset upon reading. This is used for fast counters
118 which tend to overflow. So instead of reading them normally you reset them
119 after every read to make sure you have a maximum time available before the
120 next overflow. Another usage is for things you count like number of messages
121 since the last update.
125 is for storing the result of a formula applied to other data sources
126 in the B<RRD>. This data source is not supplied a value on update, but
127 rather its Primary Data Points (PDPs) are computed from the PDPs of
128 the data sources according to the rpn-expression that defines the
129 formula. Consolidation functions are then applied normally to the PDPs
130 of the COMPUTE data source (that is the rpn-expression is only applied
131 to generate PDPs). In database software, such data sets are referred
132 to as "virtual" or "computed" columns.
136 I<heartbeat> defines the maximum number of seconds that may pass
137 between two updates of this data source before the value of the
138 data source is assumed to be I<*UNKNOWN*>.
140 I<min> and I<max> define the expected range values for data supplied by a
141 data source. If I<min> and/or I<max> any value outside the defined range
142 will be regarded as I<*UNKNOWN*>. If you do not know or care about min and
143 max, set them to U for unknown. Note that min and max always refer to the
144 processed values of the DS. For a traffic-B<COUNTER> type DS this would be
145 the maximum and minimum data-rate expected from the device.
147 I<If information on minimal/maximal expected values is available,
148 always set the min and/or max properties. This will help RRDtool in
149 doing a simple sanity check on the data supplied when running update.>
151 I<rpn-expression> defines the formula used to compute the PDPs of a
152 COMPUTE data source from other data sources in the same <RRD>. It is
153 similar to defining a B<CDEF> argument for the graph command. Please
154 refer to that manual page for a list and description of RPN operations
155 supported. For COMPUTE data sources, the following RPN operations are
156 not supported: COUNT, PREV, TIME, and LTIME. In addition, in defining
157 the RPN expression, the COMPUTE data source may only refer to the
158 names of data source listed previously in the create command. This is
159 similar to the restriction that B<CDEF>s must refer only to B<DEF>s
160 and B<CDEF>s previously defined in the same graph command.
162 =item B<RRA:>I<CF>B<:>I<cf arguments>
165 The purpose of an B<RRD> is to store data in the round robin archives
166 (B<RRA>). An archive consists of a number of data values or statistics for
167 each of the defined data-sources (B<DS>) and is defined with an B<RRA> line.
169 When data is entered into an B<RRD>, it is first fit into time slots
170 of the length defined with the B<-s> option, thus becoming a I<primary
173 The data is also processed with the consolidation function (I<CF>) of
174 the archive. There are several consolidation functions that
175 consolidate primary data points via an aggregate function: B<AVERAGE>,
176 B<MIN>, B<MAX>, B<LAST>. The format of B<RRA> line for these
177 consolidation functions is:
179 B<RRA:>I<AVERAGE | MIN | MAX | LAST>B<:>I<xff>B<:>I<steps>B<:>I<rows>
181 I<xff> The xfiles factor defines what part of a consolidation interval may
182 be made up from I<*UNKNOWN*> data while the consolidated value is still
183 regarded as known. It is given as the ratio of allowed I<*UNKNOWN*> PDPs
184 to the number of PDPs in the interval. Thus, it ranges from 0 to 1 (exclusive).
187 I<steps> defines how many of these I<primary data points> are used to build
188 a I<consolidated data point> which then goes into the archive.
190 I<rows> defines how many generations of data values are kept in an B<RRA>.
194 =head1 Aberrant Behavior Detection with Holt-Winters Forecasting
196 In addition to the aggregate functions, there are a set of specialized
197 functions that enable B<RRDtool> to provide data smoothing (via the
198 Holt-Winters forecasting algorithm), confidence bands, and the
199 flagging aberrant behavior in the data source time series:
205 B<RRA:>I<HWPREDICT>B<:>I<rows>B<:>I<alpha>B<:>I<beta>B<:>I<seasonal period>[B<:>I<rra-num>]
209 B<RRA:>I<SEASONAL>B<:>I<seasonal period>B<:>I<gamma>B<:>I<rra-num>
213 B<RRA:>I<DEVSEASONAL>B<:>I<seasonal period>B<:>I<gamma>B<:>I<rra-num>
217 B<RRA:>I<DEVPREDICT>B<:>I<rows>B<:>I<rra-num>
221 B<RRA:>I<FAILURES>B<:>I<rows>B<:>I<threshold>B<:>I<window length>B<:>I<rra-num>
225 These B<RRAs> differ from the true consolidation functions in several ways.
226 First, each of the B<RRA>s is updated once for every primary data point.
227 Second, these B<RRAs> are interdependent. To generate real-time confidence
228 bounds, a matched set of HWPREDICT, SEASONAL, DEVSEASONAL, and
229 DEVPREDICT must exist. Generating smoothed values of the primary data points
230 requires both a HWPREDICT B<RRA> and SEASONAL B<RRA>. Aberrant behavior
231 detection requires FAILURES, HWPREDICT, DEVSEASONAL, and SEASONAL.
233 The actual predicted, or smoothed, values are stored in the HWPREDICT
234 B<RRA>. The predicted deviations are stored in DEVPREDICT (think a standard
235 deviation which can be scaled to yield a confidence band). The FAILURES
236 B<RRA> stores binary indicators. A 1 marks the indexed observation as
237 failure; that is, the number of confidence bounds violations in the
238 preceding window of observations met or exceeded a specified threshold. An
239 example of using these B<RRAs> to graph confidence bounds and failures
240 appears in L<rrdgraph>.
242 The SEASONAL and DEVSEASONAL B<RRAs> store the seasonal coefficients for the
243 Holt-Winters forecasting algorithm and the seasonal deviations, respectively.
244 There is one entry per observation time point in the seasonal cycle. For
245 example, if primary data points are generated every five minutes and the
246 seasonal cycle is 1 day, both SEASONAL and DEVSEASONAL will have 288 rows.
248 In order to simplify the creation for the novice user, in addition to
249 supporting explicit creation of the HWPREDICT, SEASONAL, DEVPREDICT,
250 DEVSEASONAL, and FAILURES B<RRAs>, the B<RRDtool> create command supports
251 implicit creation of the other four when HWPREDICT is specified alone and
252 the final argument I<rra-num> is omitted.
254 I<rows> specifies the length of the B<RRA> prior to wrap around. Remember
255 that there is a one-to-one correspondence between primary data points and
256 entries in these RRAs. For the HWPREDICT CF, I<rows> should be larger than
257 the I<seasonal period>. If the DEVPREDICT B<RRA> is implicitly created, the
258 default number of rows is the same as the HWPREDICT I<rows> argument. If the
259 FAILURES B<RRA> is implicitly created, I<rows> will be set to the I<seasonal
260 period> argument of the HWPREDICT B<RRA>. Of course, the B<RRDtool>
261 I<resize> command is available if these defaults are not sufficient and the
262 creator wishes to avoid explicit creations of the other specialized function
265 I<seasonal period> specifies the number of primary data points in a seasonal
266 cycle. If SEASONAL and DEVSEASONAL are implicitly created, this argument for
267 those B<RRAs> is set automatically to the value specified by HWPREDICT. If
268 they are explicitly created, the creator should verify that all three
269 I<seasonal period> arguments agree.
271 I<alpha> is the adaption parameter of the intercept (or baseline)
272 coefficient in the Holt-Winters forecasting algorithm. See L<rrdtool> for a
273 description of this algorithm. I<alpha> must lie between 0 and 1. A value
274 closer to 1 means that more recent observations carry greater weight in
275 predicting the baseline component of the forecast. A value closer to 0 means
276 that past history carries greater weight in predicting the baseline
279 I<beta> is the adaption parameter of the slope (or linear trend) coefficient
280 in the Holt-Winters forecasting algorithm. I<beta> must lie between 0 and 1
281 and plays the same role as I<alpha> with respect to the predicted linear
284 I<gamma> is the adaption parameter of the seasonal coefficients in the
285 Holt-Winters forecasting algorithm (HWPREDICT) or the adaption parameter in
286 the exponential smoothing update of the seasonal deviations. It must lie
287 between 0 and 1. If the SEASONAL and DEVSEASONAL B<RRAs> are created
288 implicitly, they will both have the same value for I<gamma>: the value
289 specified for the HWPREDICT I<alpha> argument. Note that because there is
290 one seasonal coefficient (or deviation) for each time point during the
291 seasonal cycle, the adaptation rate is much slower than the baseline. Each
292 seasonal coefficient is only updated (or adapts) when the observed value
293 occurs at the offset in the seasonal cycle corresponding to that
296 If SEASONAL and DEVSEASONAL B<RRAs> are created explicitly, I<gamma> need not
297 be the same for both. Note that I<gamma> can also be changed via the
298 B<RRDtool> I<tune> command.
300 I<rra-num> provides the links between related B<RRAs>. If HWPREDICT is
301 specified alone and the other B<RRAs> are created implicitly, then
302 there is no need to worry about this argument. If B<RRAs> are created
303 explicitly, then carefully pay attention to this argument. For each
304 B<RRA> which includes this argument, there is a dependency between
305 that B<RRA> and another B<RRA>. The I<rra-num> argument is the 1-based
306 index in the order of B<RRA> creation (that is, the order they appear
307 in the I<create> command). The dependent B<RRA> for each B<RRA>
308 requiring the I<rra-num> argument is listed here:
314 HWPREDICT I<rra-num> is the index of the SEASONAL B<RRA>.
318 SEASONAL I<rra-num> is the index of the HWPREDICT B<RRA>.
322 DEVPREDICT I<rra-num> is the index of the DEVSEASONAL B<RRA>.
326 DEVSEASONAL I<rra-num> is the index of the HWPREDICT B<RRA>.
330 FAILURES I<rra-num> is the index of the DEVSEASONAL B<RRA>.
334 I<threshold> is the minimum number of violations (observed values outside
335 the confidence bounds) within a window that constitutes a failure. If the
336 FAILURES B<RRA> is implicitly created, the default value is 7.
338 I<window length> is the number of time points in the window. Specify an
339 integer greater than or equal to the threshold and less than or equal to 28.
340 The time interval this window represents depends on the interval between
341 primary data points. If the FAILURES B<RRA> is implicitly created, the
344 =head1 The HEARTBEAT and the STEP
346 Here is an explanation by Don Baarda on the inner workings of RRDtool.
347 It may help you to sort out why all this *UNKNOWN* data is popping
348 up in your databases:
350 RRDtool gets fed samples at arbitrary times. From these it builds Primary
351 Data Points (PDPs) at exact times on every "step" interval. The PDPs are
352 then accumulated into RRAs.
354 The "heartbeat" defines the maximum acceptable interval between
355 samples. If the interval between samples is less than "heartbeat",
356 then an average rate is calculated and applied for that interval. If
357 the interval between samples is longer than "heartbeat", then that
358 entire interval is considered "unknown". Note that there are other
359 things that can make a sample interval "unknown", such as the rate
360 exceeding limits, or even an "unknown" input sample.
362 The known rates during a PDP's "step" interval are used to calculate
363 an average rate for that PDP. Also, if the total "unknown" time during
364 the "step" interval exceeds the "heartbeat", the entire PDP is marked
365 as "unknown". This means that a mixture of known and "unknown" sample
366 times in a single PDP "step" may or may not add up to enough "unknown"
367 time to exceed "heartbeat" and hence mark the whole PDP "unknown". So
368 "heartbeat" is not only the maximum acceptable interval between
369 samples, but also the maximum acceptable amount of "unknown" time per
370 PDP (obviously this is only significant if you have "heartbeat" less
373 The "heartbeat" can be short (unusual) or long (typical) relative to
374 the "step" interval between PDPs. A short "heartbeat" means you
375 require multiple samples per PDP, and if you don't get them mark the
376 PDP unknown. A long heartbeat can span multiple "steps", which means
377 it is acceptable to have multiple PDPs calculated from a single
378 sample. An extreme example of this might be a "step" of 5 minutes and a
379 "heartbeat" of one day, in which case a single sample every day will
380 result in all the PDPs for that entire day period being set to the
381 same average rate. I<-- Don Baarda E<lt>don.baarda@baesystems.comE<gt>>
387 u|02|----* sample1, restart "hb"-timer
393 |08|----* sample2, restart "hb"
396 u|11|----* sample3, restart "hb"
402 |17|----* sample4, restart "hb", create "pdp" for step1 =
403 |18| / = unknown due to 10 "u" labled secs > "hb"
406 |21|----* sample5, restart "hb"
409 |24|----* sample6, restart "hb"
412 |27|----* sample7, restart "hb"
415 |23|----* sample8, restart "hb", create "pdp" for step1, create "cdp"
419 graphics by I<vladimir.lavrov@desy.de>.
422 =head1 HOW TO MEASURE
424 Here are a few hints on how to measure:
431 Usually you have some type of meter you can read to get the temperature.
432 The temperature is not really connected with a time. The only connection is
433 that the temperature reading happened at a certain time. You can use the
434 B<GAUGE> data source type for this. RRDtool will then record your reading
435 together with the time.
439 Assume you have a method to count the number of messages transported by
440 your mailserver in a certain amount of time, giving you data like '5
441 messages in the last 65 seconds'. If you look at the count of 5 like an
442 B<ABSOLUTE> data type you can simply update the RRD with the number 5 and the
443 end time of your monitoring period. RRDtool will then record the number of
444 messages per second. If at some later stage you want to know the number of
445 messages transported in a day, you can get the average messages per second
446 from RRDtool for the day in question and multiply this number with the
447 number of seconds in a day. Because all math is run with Doubles, the
448 precision should be acceptable.
450 =item It's always a Rate
452 RRDtool stores rates in amount/second for COUNTER, DERIVE and ABSOLUTE
453 data. When you plot the data, you will get on the y axis
454 amount/second which you might be tempted to convert to an absolute
455 amount by multiplying by the delta-time between the points. RRDtool
456 plots continuous data, and as such is not appropriate for plotting
457 absolute amounts as for example "total bytes" sent and received in a
458 router. What you probably want is plot rates that you can scale to
459 bytes/hour, for example, or plot absolute amounts with another tool
460 that draws bar-plots, where the delta-time is clear on the plot for
461 each point (such that when you read the graph you see for example GB
462 on the y axis, days on the x axis and one bar for each day).
469 rrdtool create temperature.rrd --step 300 \
470 DS:temp:GAUGE:600:-273:5000 \
471 RRA:AVERAGE:0.5:1:1200 \
472 RRA:MIN:0.5:12:2400 \
473 RRA:MAX:0.5:12:2400 \
474 RRA:AVERAGE:0.5:12:2400
476 This sets up an B<RRD> called F<temperature.rrd> which accepts one
477 temperature value every 300 seconds. If no new data is supplied for
478 more than 600 seconds, the temperature becomes I<*UNKNOWN*>. The
479 minimum acceptable value is -273 and the maximum is 5'000.
481 A few archive areas are also defined. The first stores the
482 temperatures supplied for 100 hours (1'200 * 300 seconds = 100
483 hours). The second RRA stores the minimum temperature recorded over
484 every hour (12 * 300 seconds = 1 hour), for 100 days (2'400 hours). The
485 third and the fourth RRA's do the same for the maximum and
486 average temperature, respectively.
490 rrdtool create monitor.rrd --step 300 \
491 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
492 RRA:AVERAGE:0.5:1:2016 \
493 RRA:HWPREDICT:1440:0.1:0.0035:288
495 This example is a monitor of a router interface. The first B<RRA> tracks the
496 traffic flow in octets; the second B<RRA> generates the specialized
497 functions B<RRAs> for aberrant behavior detection. Note that the I<rra-num>
498 argument of HWPREDICT is missing, so the other B<RRAs> will implicitly be
499 created with default parameter values. In this example, the forecasting
500 algorithm baseline adapts quickly; in fact the most recent one hour of
501 observations (each at 5 minute intervals) accounts for 75% of the baseline
502 prediction. The linear trend forecast adapts much more slowly. Observations
503 made during the last day (at 288 observations per day) account for only
504 65% of the predicted linear trend. Note: these computations rely on an
505 exponential smoothing formula described in the LISA 2000 paper.
507 The seasonal cycle is one day (288 data points at 300 second intervals), and
508 the seasonal adaption parameter will be set to 0.1. The RRD file will store 5
509 days (1'440 data points) of forecasts and deviation predictions before wrap
510 around. The file will store 1 day (a seasonal cycle) of 0-1 indicators in
513 The same RRD file and B<RRAs> are created with the following command,
514 which explicitly creates all specialized function B<RRAs>.
516 rrdtool create monitor.rrd --step 300 \
517 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
518 RRA:AVERAGE:0.5:1:2016 \
519 RRA:HWPREDICT:1440:0.1:0.0035:288:3 \
520 RRA:SEASONAL:288:0.1:2 \
521 RRA:DEVPREDICT:1440:5 \
522 RRA:DEVSEASONAL:288:0.1:2 \
523 RRA:FAILURES:288:7:9:5
525 Of course, explicit creation need not replicate implicit create, a
526 number of arguments could be changed.
530 rrdtool create proxy.rrd --step 300 \
531 DS:Total:DERIVE:1800:0:U \
532 DS:Duration:DERIVE:1800:0:U \
533 DS:AvgReqDur:COMPUTE:Duration,Requests,0,EQ,1,Requests,IF,/ \
534 RRA:AVERAGE:0.5:1:2016
536 This example is monitoring the average request duration during each 300 sec
537 interval for requests processed by a web proxy during the interval.
538 In this case, the proxy exposes two counters, the number of requests
539 processed since boot and the total cumulative duration of all processed
540 requests. Clearly these counters both have some rollover point, but using the
541 DERIVE data source also handles the reset that occurs when the web proxy is
542 stopped and restarted.
544 In the B<RRD>, the first data source stores the requests per second rate
545 during the interval. The second data source stores the total duration of all
546 requests processed during the interval divided by 300. The COMPUTE data source
547 divides each PDP of the AccumDuration by the corresponding PDP of
548 TotalRequests and stores the average request duration. The remainder of the
549 RPN expression handles the divide by zero case.
553 Tobias Oetiker E<lt>tobi@oetiker.chE<gt>