1RRDCREATE(1) rrdtool RRDCREATE(1)
2
3
4
6 rrdcreate - Set up a new Round Robin Database
7
9 rrdtool create filename [--start|-b start time] [--step|-s step]
10 [DS:ds-name:DST:dst arguments] [RRA:CF:cf arguments]
11
13 The create function of RRDtool lets you set up new Round Robin Database
14 (RRD) files. The file is created at its final, full size and filled
15 with *UNKNOWN* data.
16
17 filename
18 The name of the RRD you want to create. RRD files should end with the
19 extension .rrd. However, RRDtool will accept any filename.
20
21 --start|-b start time (default: now - 10s)
22 Specifies the time in seconds since 1970-01-01 UTC when the first value
23 should be added to the RRD. RRDtool will not accept any data timed
24 before or at the time specified.
25
26 See also AT-STYLE TIME SPECIFICATION section in the rrdfetch
27 documentation for other ways to specify time.
28
29 --step|-s step (default: 300 seconds)
30 Specifies the base interval in seconds with which data will be fed into
31 the RRD.
32
33 DS:ds-name:DST:dst arguments
34 A single RRD can accept input from several data sources (DS), for
35 example incoming and outgoing traffic on a specific communication line.
36 With the DS configuration option you must define some basic properties
37 of each data source you want to store in the RRD.
38
39 ds-name is the name you will use to reference this particular data
40 source from an RRD. A ds-name must be 1 to 19 characters long in the
41 characters [a-zA-Z0-9_].
42
43 DST defines the Data Source Type. The remaining arguments of a data
44 source entry depend on the data source type. For GAUGE, COUNTER,
45 DERIVE, and ABSOLUTE the format for a data source entry is:
46
47 DS:ds-name:GAUGE | COUNTER | DERIVE | ABSOLUTE:heartbeat:min:max
48
49 For COMPUTE data sources, the format is:
50
51 DS:ds-name:COMPUTE:rpn-expression
52
53 In order to decide which data source type to use, review the
54 definitions that follow. Also consult the section on "HOW TO MEASURE"
55 for further insight.
56
57 GAUGE
58 is for things like temperatures or number of people in a room or
59 the value of a RedHat share.
60
61 COUNTER
62 is for continuous incrementing counters like the ifInOctets counter
63 in a router. The COUNTER data source assumes that the counter never
64 decreases, except when a counter overflows. The update function
65 takes the overflow into account. The counter is stored as a per-
66 second rate. When the counter overflows, RRDtool checks if the
67 overflow happened at the 32bit or 64bit border and acts accordingly
68 by adding an appropriate value to the result.
69
70 DERIVE
71 will store the derivative of the line going from the last to the
72 current value of the data source. This can be useful for gauges,
73 for example, to measure the rate of people entering or leaving a
74 room. Internally, derive works exactly like COUNTER but without
75 overflow checks. So if your counter does not reset at 32 or 64 bit
76 you might want to use DERIVE and combine it with a MIN value of 0.
77
78 NOTE on COUNTER vs DERIVE
79
80 by Don Baarda <don.baarda@baesystems.com>
81
82 If you cannot tolerate ever mistaking the occasional counter reset
83 for a legitimate counter wrap, and would prefer "Unknowns" for all
84 legitimate counter wraps and resets, always use DERIVE with min=0.
85 Otherwise, using COUNTER with a suitable max will return correct
86 values for all legitimate counter wraps, mark some counter resets
87 as "Unknown", but can mistake some counter resets for a legitimate
88 counter wrap.
89
90 For a 5 minute step and 32-bit counter, the probability of
91 mistaking a counter reset for a legitimate wrap is arguably about
92 0.8% per 1Mbps of maximum bandwidth. Note that this equates to 80%
93 for 100Mbps interfaces, so for high bandwidth interfaces and a
94 32bit counter, DERIVE with min=0 is probably preferable. If you are
95 using a 64bit counter, just about any max setting will eliminate
96 the possibility of mistaking a reset for a counter wrap.
97
98 ABSOLUTE
99 is for counters which get reset upon reading. This is used for fast
100 counters which tend to overflow. So instead of reading them
101 normally you reset them after every read to make sure you have a
102 maximum time available before the next overflow. Another usage is
103 for things you count like number of messages since the last update.
104
105 COMPUTE
106 is for storing the result of a formula applied to other data
107 sources in the RRD. This data source is not supplied a value on
108 update, but rather its Primary Data Points (PDPs) are computed from
109 the PDPs of the data sources according to the rpn-expression that
110 defines the formula. Consolidation functions are then applied
111 normally to the PDPs of the COMPUTE data source (that is the rpn-
112 expression is only applied to generate PDPs). In database software,
113 such data sets are referred to as "virtual" or "computed" columns.
114
115 heartbeat defines the maximum number of seconds that may pass between
116 two updates of this data source before the value of the data source is
117 assumed to be *UNKNOWN*.
118
119 min and max define the expected range values for data supplied by a
120 data source. If min and/or max any value outside the defined range will
121 be regarded as *UNKNOWN*. If you do not know or care about min and max,
122 set them to U for unknown. Note that min and max always refer to the
123 processed values of the DS. For a traffic-COUNTER type DS this would be
124 the maximum and minimum data-rate expected from the device.
125
126 If information on minimal/maximal expected values is available, always
127 set the min and/or max properties. This will help RRDtool in doing a
128 simple sanity check on the data supplied when running update.
129
130 rpn-expression defines the formula used to compute the PDPs of a
131 COMPUTE data source from other data sources in the same <RRD>. It is
132 similar to defining a CDEF argument for the graph command. Please refer
133 to that manual page for a list and description of RPN operations
134 supported. For COMPUTE data sources, the following RPN operations are
135 not supported: COUNT, PREV, TIME, and LTIME. In addition, in defining
136 the RPN expression, the COMPUTE data source may only refer to the names
137 of data source listed previously in the create command. This is similar
138 to the restriction that CDEFs must refer only to DEFs and CDEFs
139 previously defined in the same graph command.
140
141 RRA:CF:cf arguments
142 The purpose of an RRD is to store data in the round robin archives
143 (RRA). An archive consists of a number of data values or statistics for
144 each of the defined data-sources (DS) and is defined with an RRA line.
145
146 When data is entered into an RRD, it is first fit into time slots of
147 the length defined with the -s option, thus becoming a primary data
148 point.
149
150 The data is also processed with the consolidation function (CF) of the
151 archive. There are several consolidation functions that consolidate
152 primary data points via an aggregate function: AVERAGE, MIN, MAX, LAST.
153
154 AVERAGE
155 the average of the data points is stored.
156
157 MIN the smallest of the data points is stored.
158
159 MAX the largest of the data points is stored.
160
161 LAST
162 the last data points is used.
163
164 Note that data aggregation inevitably leads to loss of precision and
165 information. The trick is to pick the aggregate function such that the
166 interesting properties of your data is kept across the aggregation
167 process.
168
169 The format of RRA line for these consolidation functions is:
170
171 RRA:AVERAGE | MIN | MAX | LAST:xff:steps:rows
172
173 xff The xfiles factor defines what part of a consolidation interval may
174 be made up from *UNKNOWN* data while the consolidated value is still
175 regarded as known. It is given as the ratio of allowed *UNKNOWN* PDPs
176 to the number of PDPs in the interval. Thus, it ranges from 0 to 1
177 (exclusive).
178
179 steps defines how many of these primary data points are used to build a
180 consolidated data point which then goes into the archive.
181
182 rows defines how many generations of data values are kept in an RRA.
183 Obviously, this has to be greater than zero.
184
186 In addition to the aggregate functions, there are a set of specialized
187 functions that enable RRDtool to provide data smoothing (via the Holt-
188 Winters forecasting algorithm), confidence bands, and the flagging
189 aberrant behavior in the data source time series:
190
191 · RRA:HWPREDICT:rows:alpha:beta:seasonal period[:rra-num]
192
193 · RRA:MHWPREDICT:rows:alpha:beta:seasonal period[:rra-num]
194
195 · RRA:SEASONAL:seasonal period:gamma:rra-
196 num[:smoothing-window=fraction]
197
198 · RRA:DEVSEASONAL:seasonal period:gamma:rra-
199 num[:smoothing-window=fraction]
200
201 · RRA:DEVPREDICT:rows:rra-num
202
203 · RRA:FAILURES:rows:threshold:window length:rra-num
204
205 These RRAs differ from the true consolidation functions in several
206 ways. First, each of the RRAs is updated once for every primary data
207 point. Second, these RRAs are interdependent. To generate real-time
208 confidence bounds, a matched set of SEASONAL, DEVSEASONAL, DEVPREDICT,
209 and either HWPREDICT or MHWPREDICT must exist. Generating smoothed
210 values of the primary data points requires a SEASONAL RRA and either an
211 HWPREDICT or MHWPREDICT RRA. Aberrant behavior detection requires
212 FAILURES, DEVSEASONAL, SEASONAL, and either HWPREDICT or MHWPREDICT.
213
214 The predicted, or smoothed, values are stored in the HWPREDICT or
215 MHWPREDICT RRA. HWPREDICT and MHWPREDICT are actually two variations on
216 the Holt-Winters method. They are interchangeable. Both attempt to
217 decompose data into three components: a baseline, a trend, and a
218 seasonal coefficient. HWPREDICT adds its seasonal coefficient to the
219 baseline to form a prediction, whereas MHWPREDICT multiplies its
220 seasonal coefficient by the baseline to form a prediction. The
221 difference is noticeable when the baseline changes significantly in the
222 course of a season; HWPREDICT will predict the seasonality to stay
223 constant as the baseline changes, but MHWPREDICT will predict the
224 seasonality to grow or shrink in proportion to the baseline. The proper
225 choice of method depends on the thing being modeled. For simplicity,
226 the rest of this discussion will refer to HWPREDICT, but MHWPREDICT may
227 be substituted in its place.
228
229 The predicted deviations are stored in DEVPREDICT (think a standard
230 deviation which can be scaled to yield a confidence band). The FAILURES
231 RRA stores binary indicators. A 1 marks the indexed observation as
232 failure; that is, the number of confidence bounds violations in the
233 preceding window of observations met or exceeded a specified threshold.
234 An example of using these RRAs to graph confidence bounds and failures
235 appears in rrdgraph.
236
237 The SEASONAL and DEVSEASONAL RRAs store the seasonal coefficients for
238 the Holt-Winters forecasting algorithm and the seasonal deviations,
239 respectively. There is one entry per observation time point in the
240 seasonal cycle. For example, if primary data points are generated every
241 five minutes and the seasonal cycle is 1 day, both SEASONAL and
242 DEVSEASONAL will have 288 rows.
243
244 In order to simplify the creation for the novice user, in addition to
245 supporting explicit creation of the HWPREDICT, SEASONAL, DEVPREDICT,
246 DEVSEASONAL, and FAILURES RRAs, the RRDtool create command supports
247 implicit creation of the other four when HWPREDICT is specified alone
248 and the final argument rra-num is omitted.
249
250 rows specifies the length of the RRA prior to wrap around. Remember
251 that there is a one-to-one correspondence between primary data points
252 and entries in these RRAs. For the HWPREDICT CF, rows should be larger
253 than the seasonal period. If the DEVPREDICT RRA is implicitly created,
254 the default number of rows is the same as the HWPREDICT rows argument.
255 If the FAILURES RRA is implicitly created, rows will be set to the
256 seasonal period argument of the HWPREDICT RRA. Of course, the RRDtool
257 resize command is available if these defaults are not sufficient and
258 the creator wishes to avoid explicit creations of the other specialized
259 function RRAs.
260
261 seasonal period specifies the number of primary data points in a
262 seasonal cycle. If SEASONAL and DEVSEASONAL are implicitly created,
263 this argument for those RRAs is set automatically to the value
264 specified by HWPREDICT. If they are explicitly created, the creator
265 should verify that all three seasonal period arguments agree.
266
267 alpha is the adaption parameter of the intercept (or baseline)
268 coefficient in the Holt-Winters forecasting algorithm. See rrdtool for
269 a description of this algorithm. alpha must lie between 0 and 1. A
270 value closer to 1 means that more recent observations carry greater
271 weight in predicting the baseline component of the forecast. A value
272 closer to 0 means that past history carries greater weight in
273 predicting the baseline component.
274
275 beta is the adaption parameter of the slope (or linear trend)
276 coefficient in the Holt-Winters forecasting algorithm. beta must lie
277 between 0 and 1 and plays the same role as alpha with respect to the
278 predicted linear trend.
279
280 gamma is the adaption parameter of the seasonal coefficients in the
281 Holt-Winters forecasting algorithm (HWPREDICT) or the adaption
282 parameter in the exponential smoothing update of the seasonal
283 deviations. It must lie between 0 and 1. If the SEASONAL and
284 DEVSEASONAL RRAs are created implicitly, they will both have the same
285 value for gamma: the value specified for the HWPREDICT alpha argument.
286 Note that because there is one seasonal coefficient (or deviation) for
287 each time point during the seasonal cycle, the adaptation rate is much
288 slower than the baseline. Each seasonal coefficient is only updated (or
289 adapts) when the observed value occurs at the offset in the seasonal
290 cycle corresponding to that coefficient.
291
292 If SEASONAL and DEVSEASONAL RRAs are created explicitly, gamma need not
293 be the same for both. Note that gamma can also be changed via the
294 RRDtool tune command.
295
296 smoothing-window specifies the fraction of a season that should be
297 averaged around each point. By default, the value of smoothing-window
298 is 0.05, which means each value in SEASONAL and DEVSEASONAL will be
299 occasionally replaced by averaging it with its (seasonal period*0.05)
300 nearest neighbors. Setting smoothing-window to zero will disable the
301 running-average smoother altogether.
302
303 rra-num provides the links between related RRAs. If HWPREDICT is
304 specified alone and the other RRAs are created implicitly, then there
305 is no need to worry about this argument. If RRAs are created
306 explicitly, then carefully pay attention to this argument. For each RRA
307 which includes this argument, there is a dependency between that RRA
308 and another RRA. The rra-num argument is the 1-based index in the order
309 of RRA creation (that is, the order they appear in the create command).
310 The dependent RRA for each RRA requiring the rra-num argument is listed
311 here:
312
313 · HWPREDICT rra-num is the index of the SEASONAL RRA.
314
315 · SEASONAL rra-num is the index of the HWPREDICT RRA.
316
317 · DEVPREDICT rra-num is the index of the DEVSEASONAL RRA.
318
319 · DEVSEASONAL rra-num is the index of the HWPREDICT RRA.
320
321 · FAILURES rra-num is the index of the DEVSEASONAL RRA.
322
323 threshold is the minimum number of violations (observed values outside
324 the confidence bounds) within a window that constitutes a failure. If
325 the FAILURES RRA is implicitly created, the default value is 7.
326
327 window length is the number of time points in the window. Specify an
328 integer greater than or equal to the threshold and less than or equal
329 to 28. The time interval this window represents depends on the
330 interval between primary data points. If the FAILURES RRA is implicitly
331 created, the default value is 9.
332
334 Here is an explanation by Don Baarda on the inner workings of RRDtool.
335 It may help you to sort out why all this *UNKNOWN* data is popping up
336 in your databases:
337
338 RRDtool gets fed samples/updates at arbitrary times. From these it
339 builds Primary Data Points (PDPs) on every "step" interval. The PDPs
340 are then accumulated into the RRAs.
341
342 The "heartbeat" defines the maximum acceptable interval between
343 samples/updates. If the interval between samples is less than
344 "heartbeat", then an average rate is calculated and applied for that
345 interval. If the interval between samples is longer than "heartbeat",
346 then that entire interval is considered "unknown". Note that there are
347 other things that can make a sample interval "unknown", such as the
348 rate exceeding limits, or a sample that was explicitly marked as
349 unknown.
350
351 The known rates during a PDP's "step" interval are used to calculate an
352 average rate for that PDP. If the total "unknown" time accounts for
353 more than half the "step", the entire PDP is marked as "unknown". This
354 means that a mixture of known and "unknown" sample times in a single
355 PDP "step" may or may not add up to enough "known" time to warrent for
356 a known PDP.
357
358 The "heartbeat" can be short (unusual) or long (typical) relative to
359 the "step" interval between PDPs. A short "heartbeat" means you require
360 multiple samples per PDP, and if you don't get them mark the PDP
361 unknown. A long heartbeat can span multiple "steps", which means it is
362 acceptable to have multiple PDPs calculated from a single sample. An
363 extreme example of this might be a "step" of 5 minutes and a
364 "heartbeat" of one day, in which case a single sample every day will
365 result in all the PDPs for that entire day period being set to the same
366 average rate. -- Don Baarda <don.baarda@baesystems.com>
367
368 time|
369 axis|
370 begin__|00|
371 |01|
372 u|02|----* sample1, restart "hb"-timer
373 u|03| /
374 u|04| /
375 u|05| /
376 u|06|/ "hbt" expired
377 u|07|
378 |08|----* sample2, restart "hb"
379 |09| /
380 |10| /
381 u|11|----* sample3, restart "hb"
382 u|12| /
383 u|13| /
384 step1_u|14| /
385 u|15|/ "swt" expired
386 u|16|
387 |17|----* sample4, restart "hb", create "pdp" for step1 =
388 |18| / = unknown due to 10 "u" labled secs > 0.5 * step
389 |19| /
390 |20| /
391 |21|----* sample5, restart "hb"
392 |22| /
393 |23| /
394 |24|----* sample6, restart "hb"
395 |25| /
396 |26| /
397 |27|----* sample7, restart "hb"
398 step2__|28| /
399 |22| /
400 |23|----* sample8, restart "hb", create "pdp" for step1, create "cdp"
401 |24| /
402 |25| /
403
404 graphics by vladimir.lavrov@desy.de.
405
407 Here are a few hints on how to measure:
408
409 Temperature
410 Usually you have some type of meter you can read to get the
411 temperature. The temperature is not really connected with a time.
412 The only connection is that the temperature reading happened at a
413 certain time. You can use the GAUGE data source type for this.
414 RRDtool will then record your reading together with the time.
415
416 Mail Messages
417 Assume you have a method to count the number of messages
418 transported by your mailserver in a certain amount of time, giving
419 you data like '5 messages in the last 65 seconds'. If you look at
420 the count of 5 like an ABSOLUTE data type you can simply update the
421 RRD with the number 5 and the end time of your monitoring period.
422 RRDtool will then record the number of messages per second. If at
423 some later stage you want to know the number of messages
424 transported in a day, you can get the average messages per second
425 from RRDtool for the day in question and multiply this number with
426 the number of seconds in a day. Because all math is run with
427 Doubles, the precision should be acceptable.
428
429 It's always a Rate
430 RRDtool stores rates in amount/second for COUNTER, DERIVE and
431 ABSOLUTE data. When you plot the data, you will get on the y axis
432 amount/second which you might be tempted to convert to an absolute
433 amount by multiplying by the delta-time between the points. RRDtool
434 plots continuous data, and as such is not appropriate for plotting
435 absolute amounts as for example "total bytes" sent and received in
436 a router. What you probably want is plot rates that you can scale
437 to bytes/hour, for example, or plot absolute amounts with another
438 tool that draws bar-plots, where the delta-time is clear on the
439 plot for each point (such that when you read the graph you see for
440 example GB on the y axis, days on the x axis and one bar for each
441 day).
442
444 rrdtool create temperature.rrd --step 300 \
445 DS:temp:GAUGE:600:-273:5000 \
446 RRA:AVERAGE:0.5:1:1200 \
447 RRA:MIN:0.5:12:2400 \
448 RRA:MAX:0.5:12:2400 \
449 RRA:AVERAGE:0.5:12:2400
450
451 This sets up an RRD called temperature.rrd which accepts one
452 temperature value every 300 seconds. If no new data is supplied for
453 more than 600 seconds, the temperature becomes *UNKNOWN*. The minimum
454 acceptable value is -273 and the maximum is 5'000.
455
456 A few archive areas are also defined. The first stores the temperatures
457 supplied for 100 hours (1'200 * 300 seconds = 100 hours). The second
458 RRA stores the minimum temperature recorded over every hour (12 * 300
459 seconds = 1 hour), for 100 days (2'400 hours). The third and the fourth
460 RRA's do the same for the maximum and average temperature,
461 respectively.
462
464 rrdtool create monitor.rrd --step 300 \
465 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
466 RRA:AVERAGE:0.5:1:2016 \
467 RRA:HWPREDICT:1440:0.1:0.0035:288
468
469 This example is a monitor of a router interface. The first RRA tracks
470 the traffic flow in octets; the second RRA generates the specialized
471 functions RRAs for aberrant behavior detection. Note that the rra-num
472 argument of HWPREDICT is missing, so the other RRAs will implicitly be
473 created with default parameter values. In this example, the forecasting
474 algorithm baseline adapts quickly; in fact the most recent one hour of
475 observations (each at 5 minute intervals) accounts for 75% of the
476 baseline prediction. The linear trend forecast adapts much more slowly.
477 Observations made during the last day (at 288 observations per day)
478 account for only 65% of the predicted linear trend. Note: these
479 computations rely on an exponential smoothing formula described in the
480 LISA 2000 paper.
481
482 The seasonal cycle is one day (288 data points at 300 second
483 intervals), and the seasonal adaption parameter will be set to 0.1. The
484 RRD file will store 5 days (1'440 data points) of forecasts and
485 deviation predictions before wrap around. The file will store 1 day (a
486 seasonal cycle) of 0-1 indicators in the FAILURES RRA.
487
488 The same RRD file and RRAs are created with the following command,
489 which explicitly creates all specialized function RRAs.
490
491 rrdtool create monitor.rrd --step 300 \
492 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
493 RRA:AVERAGE:0.5:1:2016 \
494 RRA:HWPREDICT:1440:0.1:0.0035:288:3 \
495 RRA:SEASONAL:288:0.1:2 \
496 RRA:DEVPREDICT:1440:5 \
497 RRA:DEVSEASONAL:288:0.1:2 \
498 RRA:FAILURES:288:7:9:5
499
500 Of course, explicit creation need not replicate implicit create, a
501 number of arguments could be changed.
502
504 rrdtool create proxy.rrd --step 300 \
505 DS:Total:DERIVE:1800:0:U \
506 DS:Duration:DERIVE:1800:0:U \
507 DS:AvgReqDur:COMPUTE:Duration,Requests,0,EQ,1,Requests,IF,/ \
508 RRA:AVERAGE:0.5:1:2016
509
510 This example is monitoring the average request duration during each 300
511 sec interval for requests processed by a web proxy during the interval.
512 In this case, the proxy exposes two counters, the number of requests
513 processed since boot and the total cumulative duration of all processed
514 requests. Clearly these counters both have some rollover point, but
515 using the DERIVE data source also handles the reset that occurs when
516 the web proxy is stopped and restarted.
517
518 In the RRD, the first data source stores the requests per second rate
519 during the interval. The second data source stores the total duration
520 of all requests processed during the interval divided by 300. The
521 COMPUTE data source divides each PDP of the AccumDuration by the
522 corresponding PDP of TotalRequests and stores the average request
523 duration. The remainder of the RPN expression handles the divide by
524 zero case.
525
527 Tobias Oetiker <tobi@oetiker.ch>
528
529
530
5311.3.8 2008-06-11 RRDCREATE(1)