1RRDCREATE(1) rrdtool RRDCREATE(1)
2
6 rrdcreate - Set up a new Round Robin Database
7
9 rrdtool create filename [--start|-b start time] [--step|-s step]
10 [--no-overwrite] [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 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 --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 See also AT-STYLE TIME SPECIFICATION section in the rrdfetch
27 documentation for other ways to specify time.
28 --step|-s step (default: 300 seconds)
30 Specifies the base interval in seconds with which data will be fed into
31 the RRD.
32 --no-overwrite
34 Do not clobber an existing file of the same name.
35 DS:ds-name:DST:dst arguments
37 A single RRD can accept input from several data sources (DS), for
38 example incoming and outgoing traffic on a specific communication line.
39 With the DS configuration option you must define some basic properties
40 of each data source you want to store in the RRD.
41 ds-name is the name you will use to reference this particular data
43 source from an RRD. A ds-name must be 1 to 19 characters long in the
44 characters [a-zA-Z0-9_].
45 DST defines the Data Source Type. The remaining arguments of a data
47 source entry depend on the data source type. For GAUGE, COUNTER,
48 DERIVE, and ABSOLUTE the format for a data source entry is:
49 DS:ds-name:GAUGE | COUNTER | DERIVE | ABSOLUTE:heartbeat:min:max
51 For COMPUTE data sources, the format is:
53 DS:ds-name:COMPUTE:rpn-expression
55 In order to decide which data source type to use, review the
57 definitions that follow. Also consult the section on "HOW TO MEASURE"
58 for further insight.
59 GAUGE
61 is for things like temperatures or number of people in a room or
62 the value of a RedHat share.
63 COUNTER
65 is for continuous incrementing counters like the ifInOctets counter
66 in a router. The COUNTER data source assumes that the counter never
67 decreases, except when a counter overflows. The update function
68 takes the overflow into account. The counter is stored as a per-
69 second rate. When the counter overflows, RRDtool checks if the
70 overflow happened at the 32bit or 64bit border and acts accordingly
71 by adding an appropriate value to the result.
72 DERIVE
74 will store the derivative of the line going from the last to the
75 current value of the data source. This can be useful for gauges,
76 for example, to measure the rate of people entering or leaving a
77 room. Internally, derive works exactly like COUNTER but without
78 overflow checks. So if your counter does not reset at 32 or 64 bit
79 you might want to use DERIVE and combine it with a MIN value of 0.
80 NOTE on COUNTER vs DERIVE
82 by Don Baarda <don.baarda@baesystems.com>
84 If you cannot tolerate ever mistaking the occasional counter reset
86 for a legitimate counter wrap, and would prefer "Unknowns" for all
87 legitimate counter wraps and resets, always use DERIVE with min=0.
88 Otherwise, using COUNTER with a suitable max will return correct
89 values for all legitimate counter wraps, mark some counter resets
90 as "Unknown", but can mistake some counter resets for a legitimate
91 counter wrap.
92 For a 5 minute step and 32-bit counter, the probability of
94 mistaking a counter reset for a legitimate wrap is arguably about
95 0.8% per 1Mbps of maximum bandwidth. Note that this equates to 80%
96 for 100Mbps interfaces, so for high bandwidth interfaces and a
97 32bit counter, DERIVE with min=0 is probably preferable. If you are
98 using a 64bit counter, just about any max setting will eliminate
99 the possibility of mistaking a reset for a counter wrap.
100 ABSOLUTE
102 is for counters which get reset upon reading. This is used for fast
103 counters which tend to overflow. So instead of reading them
104 normally you reset them after every read to make sure you have a
105 maximum time available before the next overflow. Another usage is
106 for things you count like number of messages since the last update.
107 COMPUTE
109 is for storing the result of a formula applied to other data
110 sources in the RRD. This data source is not supplied a value on
111 update, but rather its Primary Data Points (PDPs) are computed from
112 the PDPs of the data sources according to the rpn-expression that
113 defines the formula. Consolidation functions are then applied
114 normally to the PDPs of the COMPUTE data source (that is the rpn-
115 expression is only applied to generate PDPs). In database software,
116 such data sets are referred to as "virtual" or "computed" columns.
117 heartbeat defines the maximum number of seconds that may pass between
119 two updates of this data source before the value of the data source is
120 assumed to be *UNKNOWN*.
121 min and max define the expected range values for data supplied by a
123 data source. If min and/or max any value outside the defined range will
124 be regarded as *UNKNOWN*. If you do not know or care about min and max,
125 set them to U for unknown. Note that min and max always refer to the
126 processed values of the DS. For a traffic-COUNTER type DS this would be
127 the maximum and minimum data-rate expected from the device.
128 If information on minimal/maximal expected values is available, always
130 set the min and/or max properties. This will help RRDtool in doing a
131 simple sanity check on the data supplied when running update.
132 rpn-expression defines the formula used to compute the PDPs of a
134 COMPUTE data source from other data sources in the same <RRD>. It is
135 similar to defining a CDEF argument for the graph command. Please refer
136 to that manual page for a list and description of RPN operations
137 supported. For COMPUTE data sources, the following RPN operations are
138 not supported: COUNT, PREV, TIME, and LTIME. In addition, in defining
139 the RPN expression, the COMPUTE data source may only refer to the names
140 of data source listed previously in the create command. This is similar
141 to the restriction that CDEFs must refer only to DEFs and CDEFs
142 previously defined in the same graph command.
143 RRA:CF:cf arguments
145 The purpose of an RRD is to store data in the round robin archives
146 (RRA). An archive consists of a number of data values or statistics for
147 each of the defined data-sources (DS) and is defined with an RRA line.
148 When data is entered into an RRD, it is first fit into time slots of
150 the length defined with the -s option, thus becoming a primary data
151 point.
152 The data is also processed with the consolidation function (CF) of the
154 archive. There are several consolidation functions that consolidate
155 primary data points via an aggregate function: AVERAGE, MIN, MAX, LAST.
156 AVERAGE
158 the average of the data points is stored.
159 MIN the smallest of the data points is stored.
161 MAX the largest of the data points is stored.
163 LAST
165 the last data points is used.
166 Note that data aggregation inevitably leads to loss of precision and
168 information. The trick is to pick the aggregate function such that the
169 interesting properties of your data is kept across the aggregation
170 process.
171 The format of RRA line for these consolidation functions is:
173 RRA:AVERAGE | MIN | MAX | LAST:xff:steps:rows
175 xff The xfiles factor defines what part of a consolidation interval may
177 be made up from *UNKNOWN* data while the consolidated value is still
178 regarded as known. It is given as the ratio of allowed *UNKNOWN* PDPs
179 to the number of PDPs in the interval. Thus, it ranges from 0 to 1
180 (exclusive).
181 steps defines how many of these primary data points are used to build a
183 consolidated data point which then goes into the archive.
184 rows defines how many generations of data values are kept in an RRA.
186 Obviously, this has to be greater than zero.
187
189 In addition to the aggregate functions, there are a set of specialized
190 functions that enable RRDtool to provide data smoothing (via the Holt-
191 Winters forecasting algorithm), confidence bands, and the flagging
192 aberrant behavior in the data source time series:
193 · RRA:HWPREDICT:rows:alpha:beta:seasonal period[:rra-num]
195 · RRA:MHWPREDICT:rows:alpha:beta:seasonal period[:rra-num]
197 · RRA:SEASONAL:seasonal period:gamma:rra-
199 num[:smoothing-window=fraction]
200 · RRA:DEVSEASONAL:seasonal period:gamma:rra-
202 num[:smoothing-window=fraction]
203 · RRA:DEVPREDICT:rows:rra-num
205 · RRA:FAILURES:rows:threshold:window length:rra-num
207 These RRAs differ from the true consolidation functions in several
209 ways. First, each of the RRAs is updated once for every primary data
210 point. Second, these RRAs are interdependent. To generate real-time
211 confidence bounds, a matched set of SEASONAL, DEVSEASONAL, DEVPREDICT,
212 and either HWPREDICT or MHWPREDICT must exist. Generating smoothed
213 values of the primary data points requires a SEASONAL RRA and either an
214 HWPREDICT or MHWPREDICT RRA. Aberrant behavior detection requires
215 FAILURES, DEVSEASONAL, SEASONAL, and either HWPREDICT or MHWPREDICT.
216 The predicted, or smoothed, values are stored in the HWPREDICT or
218 MHWPREDICT RRA. HWPREDICT and MHWPREDICT are actually two variations on
219 the Holt-Winters method. They are interchangeable. Both attempt to
220 decompose data into three components: a baseline, a trend, and a
221 seasonal coefficient. HWPREDICT adds its seasonal coefficient to the
222 baseline to form a prediction, whereas MHWPREDICT multiplies its
223 seasonal coefficient by the baseline to form a prediction. The
224 difference is noticeable when the baseline changes significantly in the
225 course of a season; HWPREDICT will predict the seasonality to stay
226 constant as the baseline changes, but MHWPREDICT will predict the
227 seasonality to grow or shrink in proportion to the baseline. The proper
228 choice of method depends on the thing being modeled. For simplicity,
229 the rest of this discussion will refer to HWPREDICT, but MHWPREDICT may
230 be substituted in its place.
231 The predicted deviations are stored in DEVPREDICT (think a standard
233 deviation which can be scaled to yield a confidence band). The FAILURES
234 RRA stores binary indicators. A 1 marks the indexed observation as
235 failure; that is, the number of confidence bounds violations in the
236 preceding window of observations met or exceeded a specified threshold.
237 An example of using these RRAs to graph confidence bounds and failures
238 appears in rrdgraph.
239 The SEASONAL and DEVSEASONAL RRAs store the seasonal coefficients for
241 the Holt-Winters forecasting algorithm and the seasonal deviations,
242 respectively. There is one entry per observation time point in the
243 seasonal cycle. For example, if primary data points are generated every
244 five minutes and the seasonal cycle is 1 day, both SEASONAL and
245 DEVSEASONAL will have 288 rows.
246 In order to simplify the creation for the novice user, in addition to
248 supporting explicit creation of the HWPREDICT, SEASONAL, DEVPREDICT,
249 DEVSEASONAL, and FAILURES RRAs, the RRDtool create command supports
250 implicit creation of the other four when HWPREDICT is specified alone
251 and the final argument rra-num is omitted.
252 rows specifies the length of the RRA prior to wrap around. Remember
254 that there is a one-to-one correspondence between primary data points
255 and entries in these RRAs. For the HWPREDICT CF, rows should be larger
256 than the seasonal period. If the DEVPREDICT RRA is implicitly created,
257 the default number of rows is the same as the HWPREDICT rows argument.
258 If the FAILURES RRA is implicitly created, rows will be set to the
259 seasonal period argument of the HWPREDICT RRA. Of course, the RRDtool
260 resize command is available if these defaults are not sufficient and
261 the creator wishes to avoid explicit creations of the other specialized
262 function RRAs.
263 seasonal period specifies the number of primary data points in a
265 seasonal cycle. If SEASONAL and DEVSEASONAL are implicitly created,
266 this argument for those RRAs is set automatically to the value
267 specified by HWPREDICT. If they are explicitly created, the creator
268 should verify that all three seasonal period arguments agree.
269 alpha is the adaption parameter of the intercept (or baseline)
271 coefficient in the Holt-Winters forecasting algorithm. See rrdtool for
272 a description of this algorithm. alpha must lie between 0 and 1. A
273 value closer to 1 means that more recent observations carry greater
274 weight in predicting the baseline component of the forecast. A value
275 closer to 0 means that past history carries greater weight in
276 predicting the baseline component.
277 beta is the adaption parameter of the slope (or linear trend)
279 coefficient in the Holt-Winters forecasting algorithm. beta must lie
280 between 0 and 1 and plays the same role as alpha with respect to the
281 predicted linear trend.
282 gamma is the adaption parameter of the seasonal coefficients in the
284 Holt-Winters forecasting algorithm (HWPREDICT) or the adaption
285 parameter in the exponential smoothing update of the seasonal
286 deviations. It must lie between 0 and 1. If the SEASONAL and
287 DEVSEASONAL RRAs are created implicitly, they will both have the same
288 value for gamma: the value specified for the HWPREDICT alpha argument.
289 Note that because there is one seasonal coefficient (or deviation) for
290 each time point during the seasonal cycle, the adaptation rate is much
291 slower than the baseline. Each seasonal coefficient is only updated (or
292 adapts) when the observed value occurs at the offset in the seasonal
293 cycle corresponding to that coefficient.
294 If SEASONAL and DEVSEASONAL RRAs are created explicitly, gamma need not
296 be the same for both. Note that gamma can also be changed via the
297 RRDtool tune command.
298 smoothing-window specifies the fraction of a season that should be
300 averaged around each point. By default, the value of smoothing-window
301 is 0.05, which means each value in SEASONAL and DEVSEASONAL will be
302 occasionally replaced by averaging it with its (seasonal period*0.05)
303 nearest neighbors. Setting smoothing-window to zero will disable the
304 running-average smoother altogether.
305 rra-num provides the links between related RRAs. If HWPREDICT is
307 specified alone and the other RRAs are created implicitly, then there
308 is no need to worry about this argument. If RRAs are created
309 explicitly, then carefully pay attention to this argument. For each RRA
310 which includes this argument, there is a dependency between that RRA
311 and another RRA. The rra-num argument is the 1-based index in the order
312 of RRA creation (that is, the order they appear in the create command).
313 The dependent RRA for each RRA requiring the rra-num argument is listed
314 here:
315 · HWPREDICT rra-num is the index of the SEASONAL RRA.
317 · SEASONAL rra-num is the index of the HWPREDICT RRA.
319 · DEVPREDICT rra-num is the index of the DEVSEASONAL RRA.
321 · DEVSEASONAL rra-num is the index of the HWPREDICT RRA.
323 · FAILURES rra-num is the index of the DEVSEASONAL RRA.
325 threshold is the minimum number of violations (observed values outside
327 the confidence bounds) within a window that constitutes a failure. If
328 the FAILURES RRA is implicitly created, the default value is 7.
329 window length is the number of time points in the window. Specify an
331 integer greater than or equal to the threshold and less than or equal
332 to 28. The time interval this window represents depends on the
333 interval between primary data points. If the FAILURES RRA is implicitly
334 created, the default value is 9.
335
337 Here is an explanation by Don Baarda on the inner workings of RRDtool.
338 It may help you to sort out why all this *UNKNOWN* data is popping up
339 in your databases:
340 RRDtool gets fed samples/updates at arbitrary times. From these it
342 builds Primary Data Points (PDPs) on every "step" interval. The PDPs
343 are then accumulated into the RRAs.
344 The "heartbeat" defines the maximum acceptable interval between
346 samples/updates. If the interval between samples is less than
347 "heartbeat", then an average rate is calculated and applied for that
348 interval. If the interval between samples is longer than "heartbeat",
349 then that entire interval is considered "unknown". Note that there are
350 other things that can make a sample interval "unknown", such as the
351 rate exceeding limits, or a sample that was explicitly marked as
352 unknown.
353 The known rates during a PDP's "step" interval are used to calculate an
355 average rate for that PDP. If the total "unknown" time accounts for
356 more than half the "step", the entire PDP is marked as "unknown". This
357 means that a mixture of known and "unknown" sample times in a single
358 PDP "step" may or may not add up to enough "known" time to warrant a
359 known PDP.
360 The "heartbeat" can be short (unusual) or long (typical) relative to
362 the "step" interval between PDPs. A short "heartbeat" means you require
363 multiple samples per PDP, and if you don't get them mark the PDP
364 unknown. A long heartbeat can span multiple "steps", which means it is
365 acceptable to have multiple PDPs calculated from a single sample. An
366 extreme example of this might be a "step" of 5 minutes and a
367 "heartbeat" of one day, in which case a single sample every day will
368 result in all the PDPs for that entire day period being set to the same
369 average rate. -- Don Baarda <don.baarda@baesystems.com>
370 time|
372 axis|
373 begin__|00|
374 |01|
375 u|02|----* sample1, restart "hb"-timer
376 u|03| /
377 u|04| /
378 u|05| /
379 u|06|/ "hbt" expired
380 u|07|
381 |08|----* sample2, restart "hb"
382 |09| /
383 |10| /
384 u|11|----* sample3, restart "hb"
385 u|12| /
386 u|13| /
387 step1_u|14| /
388 u|15|/ "swt" expired
389 u|16|
390 |17|----* sample4, restart "hb", create "pdp" for step1 =
391 |18| / = unknown due to 10 "u" labled secs > 0.5 * step
392 |19| /
393 |20| /
394 |21|----* sample5, restart "hb"
395 |22| /
396 |23| /
397 |24|----* sample6, restart "hb"
398 |25| /
399 |26| /
400 |27|----* sample7, restart "hb"
401 step2__|28| /
402 |22| /
403 |23|----* sample8, restart "hb", create "pdp" for step1, create "cdp"
404 |24| /
405 |25| /
406 graphics by vladimir.lavrov@desy.de.
408
410 Here are a few hints on how to measure:
411 Temperature
413 Usually you have some type of meter you can read to get the
414 temperature. The temperature is not really connected with a time.
415 The only connection is that the temperature reading happened at a
416 certain time. You can use the GAUGE data source type for this.
417 RRDtool will then record your reading together with the time.
418 Mail Messages
420 Assume you have a method to count the number of messages
421 transported by your mail server in a certain amount of time, giving
422 you data like '5 messages in the last 65 seconds'. If you look at
423 the count of 5 like an ABSOLUTE data type you can simply update the
424 RRD with the number 5 and the end time of your monitoring period.
425 RRDtool will then record the number of messages per second. If at
426 some later stage you want to know the number of messages
427 transported in a day, you can get the average messages per second
428 from RRDtool for the day in question and multiply this number with
429 the number of seconds in a day. Because all math is run with
430 Doubles, the precision should be acceptable.
431 It's always a Rate
433 RRDtool stores rates in amount/second for COUNTER, DERIVE and
434 ABSOLUTE data. When you plot the data, you will get on the y axis
435 amount/second which you might be tempted to convert to an absolute
436 amount by multiplying by the delta-time between the points. RRDtool
437 plots continuous data, and as such is not appropriate for plotting
438 absolute amounts as for example "total bytes" sent and received in
439 a router. What you probably want is plot rates that you can scale
440 to bytes/hour, for example, or plot absolute amounts with another
441 tool that draws bar-plots, where the delta-time is clear on the
442 plot for each point (such that when you read the graph you see for
443 example GB on the y axis, days on the x axis and one bar for each
444 day).
445
447 rrdtool create temperature.rrd --step 300 \
448 DS:temp:GAUGE:600:-273:5000 \
449 RRA:AVERAGE:0.5:1:1200 \
450 RRA:MIN:0.5:12:2400 \
451 RRA:MAX:0.5:12:2400 \
452 RRA:AVERAGE:0.5:12:2400
453 This sets up an RRD called temperature.rrd which accepts one
455 temperature value every 300 seconds. If no new data is supplied for
456 more than 600 seconds, the temperature becomes *UNKNOWN*. The minimum
457 acceptable value is -273 and the maximum is 5'000.
458 A few archive areas are also defined. The first stores the temperatures
460 supplied for 100 hours (1'200 * 300 seconds = 100 hours). The second
461 RRA stores the minimum temperature recorded over every hour (12 * 300
462 seconds = 1 hour), for 100 days (2'400 hours). The third and the fourth
463 RRA's do the same for the maximum and average temperature,
464 respectively.
465
467 rrdtool create monitor.rrd --step 300 \
468 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
469 RRA:AVERAGE:0.5:1:2016 \
470 RRA:HWPREDICT:1440:0.1:0.0035:288
471 This example is a monitor of a router interface. The first RRA tracks
473 the traffic flow in octets; the second RRA generates the specialized
474 functions RRAs for aberrant behavior detection. Note that the rra-num
475 argument of HWPREDICT is missing, so the other RRAs will implicitly be
476 created with default parameter values. In this example, the forecasting
477 algorithm baseline adapts quickly; in fact the most recent one hour of
478 observations (each at 5 minute intervals) accounts for 75% of the
479 baseline prediction. The linear trend forecast adapts much more slowly.
480 Observations made during the last day (at 288 observations per day)
481 account for only 65% of the predicted linear trend. Note: these
482 computations rely on an exponential smoothing formula described in the
483 LISA 2000 paper.
484 The seasonal cycle is one day (288 data points at 300 second
486 intervals), and the seasonal adaption parameter will be set to 0.1. The
487 RRD file will store 5 days (1'440 data points) of forecasts and
488 deviation predictions before wrap around. The file will store 1 day (a
489 seasonal cycle) of 0-1 indicators in the FAILURES RRA.
490 The same RRD file and RRAs are created with the following command,
492 which explicitly creates all specialized function RRAs.
493 rrdtool create monitor.rrd --step 300 \
495 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
496 RRA:AVERAGE:0.5:1:2016 \
497 RRA:HWPREDICT:1440:0.1:0.0035:288:3 \
498 RRA:SEASONAL:288:0.1:2 \
499 RRA:DEVPREDICT:1440:5 \
500 RRA:DEVSEASONAL:288:0.1:2 \
501 RRA:FAILURES:288:7:9:5
502 Of course, explicit creation need not replicate implicit create, a
504 number of arguments could be changed.
505
507 rrdtool create proxy.rrd --step 300 \
508 DS:Total:DERIVE:1800:0:U \
509 DS:Duration:DERIVE:1800:0:U \
510 DS:AvgReqDur:COMPUTE:Duration,Requests,0,EQ,1,Requests,IF,/ \
511 RRA:AVERAGE:0.5:1:2016
512 This example is monitoring the average request duration during each 300
514 sec interval for requests processed by a web proxy during the interval.
515 In this case, the proxy exposes two counters, the number of requests
516 processed since boot and the total cumulative duration of all processed
517 requests. Clearly these counters both have some rollover point, but
518 using the DERIVE data source also handles the reset that occurs when
519 the web proxy is stopped and restarted.
520 In the RRD, the first data source stores the requests per second rate
522 during the interval. The second data source stores the total duration
523 of all requests processed during the interval divided by 300. The
524 COMPUTE data source divides each PDP of the AccumDuration by the
525 corresponding PDP of TotalRequests and stores the average request
526 duration. The remainder of the RPN expression handles the divide by
527 zero case.
528
530 Tobias Oetiker <tobi@oetiker.ch>
5311.4.4 2010-03-08 RRDCREATE(1)