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 [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
19 the 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
23 value should be added to the RRD. RRDtool will not accept any data
24 timed before or at the time specified.
25 See also AT-STYLE TIME SPECIFICATION section in the rrdfetch docu‐
27 mentation 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
31 into the RRD.
32 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
36 line. With the DS configuration option you must define some basic
37 properties of each data source you want to store in the RRD.
38 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
41 the characters [a-zA-Z0-9_].
42 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 DS:ds-name:GAUGE | COUNTER | DERIVE | ABSOLUTE:heartbeat:min:max
48 For COMPUTE data sources, the format is:
50 DS:ds-name:COMPUTE:rpn-expression
52 In order to decide which data source type to use, review the defi‐
54 nitions that follow. Also consult the section on "HOW TO MEASURE"
55 for further insight.
56 GAUGE
58 is for things like temperatures or number of people in a room
59 or the value of a RedHat share.
60 COUNTER
62 is for continuous incrementing counters like the ifInOctets
63 counter in a router. The COUNTER data source assumes that the
64 counter never decreases, except when a counter overflows. The
65 update function takes the overflow into account. The counter
66 is stored as a per-second rate. When the counter overflows,
67 RRDtool checks if the overflow happened at the 32bit or 64bit
68 border and acts accordingly by adding an appropriate value to
69 the result.
70 DERIVE
72 will store the derivative of the line going from the last to
73 the current value of the data source. This can be useful for
74 gauges, for example, to measure the rate of people entering or
75 leaving a room. Internally, derive works exactly like COUNTER
76 but without overflow checks. So if your counter does not reset
77 at 32 or 64 bit you might want to use DERIVE and combine it
78 with a MIN value of 0.
79 NOTE on COUNTER vs DERIVE
81 by Don Baarda <don.baarda@baesystems.com>
83 If you cannot tolerate ever mistaking the occasional counter
85 reset for a legitimate counter wrap, and would prefer
86 "Unknowns" for all legitimate counter wraps and resets, always
87 use DERIVE with min=0. Otherwise, using COUNTER with a suitable
88 max will return correct values for all legitimate counter
89 wraps, mark some counter resets as "Unknown", but can mistake
90 some counter resets for a legitimate counter wrap.
91 For a 5 minute step and 32-bit counter, the probability of mis‐
93 taking a counter reset for a legitimate wrap is arguably about
94 0.8% per 1Mbps of maximum bandwidth. Note that this equates to
95 80% for 100Mbps interfaces, so for high bandwidth interfaces
96 and a 32bit counter, DERIVE with min=0 is probably preferable.
97 If you are using a 64bit counter, just about any max setting
98 will eliminate the possibility of mistaking a reset for a
99 counter wrap.
100 ABSOLUTE
102 is for counters which get reset upon reading. This is used for
103 fast counters which tend to overflow. So instead of reading
104 them normally you reset them after every read to make sure you
105 have a maximum time available before the next overflow. Another
106 usage is for things you count like number of messages since the
107 last update.
108 COMPUTE
110 is for storing the result of a formula applied to other data
111 sources in the RRD. This data source is not supplied a value on
112 update, but rather its Primary Data Points (PDPs) are computed
113 from the PDPs of the data sources according to the rpn-expres‐
114 sion that defines the formula. Consolidation functions are then
115 applied normally to the PDPs of the COMPUTE data source (that
116 is the rpn-expression is only applied to generate PDPs). In
117 database software, such data sets are referred to as "virtual"
118 or "computed" columns.
119 heartbeat defines the maximum number of seconds that may pass
121 between two updates of this data source before the value of the
122 data source is assumed to be *UNKNOWN*.
123 min and max define the expected range values for data supplied by a
125 data source. If min and/or max any value outside the defined range
126 will be regarded as *UNKNOWN*. If you do not know or care about min
127 and max, set them to U for unknown. Note that min and max always
128 refer to the processed values of the DS. For a traffic-COUNTER type
129 DS this would be the maximum and minimum data-rate expected from
130 the device.
131 If information on minimal/maximal expected values is available,
133 always set the min and/or max properties. This will help RRDtool in
134 doing a simple sanity check on the data supplied when running
135 update.
136 rpn-expression defines the formula used to compute the PDPs of a
138 COMPUTE data source from other data sources in the same <RRD>. It
139 is similar to defining a CDEF argument for the graph command.
140 Please refer to that manual page for a list and description of RPN
141 operations supported. For COMPUTE data sources, the following RPN
142 operations are not supported: COUNT, PREV, TIME, and LTIME. In
143 addition, in defining the RPN expression, the COMPUTE data source
144 may only refer to the names of data source listed previously in the
145 create command. This is similar to the restriction that CDEFs must
146 refer only to DEFs and CDEFs previously defined in the same graph
147 command.
148 RRA:CF:cf arguments
150 The purpose of an RRD is to store data in the round robin archives
151 (RRA). An archive consists of a number of data values or statistics
152 for each of the defined data-sources (DS) and is defined with an
153 RRA line.
154 When data is entered into an RRD, it is first fit into time slots
156 of the length defined with the -s option, thus becoming a primary
157 data point.
158 The data is also processed with the consolidation function (CF) of
160 the archive. There are several consolidation functions that consol‐
161 idate primary data points via an aggregate function: AVERAGE, MIN,
162 MAX, LAST. The format of RRA line for these consolidation functions
163 is:
164 RRA:AVERAGE | MIN | MAX | LAST:xff:steps:rows
166 xff The xfiles factor defines what part of a consolidation interval
168 may be made up from *UNKNOWN* data while the consolidated value is
169 still regarded as known. It is given as the ratio of allowed
170 *UNKNOWN* PDPs to the number of PDPs in the interval. Thus, it
171 ranges from 0 to 1 (exclusive).
172 steps defines how many of these primary data points are used to
174 build a consolidated data point which then goes into the archive.
175 rows defines how many generations of data values are kept in an
177 RRA.
178
180 In addition to the aggregate functions, there are a set of specialized
181 functions that enable RRDtool to provide data smoothing (via the Holt-
182 Winters forecasting algorithm), confidence bands, and the flagging
183 aberrant behavior in the data source time series:
184 · RRA:HWPREDICT:rows:alpha:beta:seasonal period[:rra-num]
186 · RRA:SEASONAL:seasonal period:gamma:rra-num
188 · RRA:DEVSEASONAL:seasonal period:gamma:rra-num
190 · RRA:DEVPREDICT:rows:rra-num
192 · RRA:FAILURES:rows:threshold:window length:rra-num
194 These RRAs differ from the true consolidation functions in several
196 ways. First, each of the RRAs is updated once for every primary data
197 point. Second, these RRAs are interdependent. To generate real-time
198 confidence bounds, a matched set of HWPREDICT, SEASONAL, DEVSEASONAL,
199 and DEVPREDICT must exist. Generating smoothed values of the primary
200 data points requires both a HWPREDICT RRA and SEASONAL RRA. Aberrant
201 behavior detection requires FAILURES, HWPREDICT, DEVSEASONAL, and SEA‐
202 SONAL.
203 The actual predicted, or smoothed, values are stored in the HWPREDICT
205 RRA. The predicted deviations are stored in DEVPREDICT (think a stan‐
206 dard deviation which can be scaled to yield a confidence band). The
207 FAILURES RRA stores binary indicators. A 1 marks the indexed observa‐
208 tion as failure; that is, the number of confidence bounds violations in
209 the preceding window of observations met or exceeded a specified
210 threshold. An example of using these RRAs to graph confidence bounds
211 and failures appears in rrdgraph.
212 The SEASONAL and DEVSEASONAL RRAs store the seasonal coefficients for
214 the Holt-Winters forecasting algorithm and the seasonal deviations,
215 respectively. There is one entry per observation time point in the
216 seasonal cycle. For example, if primary data points are generated every
217 five minutes and the seasonal cycle is 1 day, both SEASONAL and DEVSEA‐
218 SONAL will have 288 rows.
219 In order to simplify the creation for the novice user, in addition to
221 supporting explicit creation of the HWPREDICT, SEASONAL, DEVPREDICT,
222 DEVSEASONAL, and FAILURES RRAs, the RRDtool create command supports
223 implicit creation of the other four when HWPREDICT is specified alone
224 and the final argument rra-num is omitted.
225 rows specifies the length of the RRA prior to wrap around. Remember
227 that there is a one-to-one correspondence between primary data points
228 and entries in these RRAs. For the HWPREDICT CF, rows should be larger
229 than the seasonal period. If the DEVPREDICT RRA is implicitly created,
230 the default number of rows is the same as the HWPREDICT rows argument.
231 If the FAILURES RRA is implicitly created, rows will be set to the sea‐
232 sonal period argument of the HWPREDICT RRA. Of course, the RRDtool
233 resize command is available if these defaults are not sufficient and
234 the creator wishes to avoid explicit creations of the other specialized
235 function RRAs.
236 seasonal period specifies the number of primary data points in a sea‐
238 sonal cycle. If SEASONAL and DEVSEASONAL are implicitly created, this
239 argument for those RRAs is set automatically to the value specified by
240 HWPREDICT. If they are explicitly created, the creator should verify
241 that all three seasonal period arguments agree.
242 alpha is the adaption parameter of the intercept (or baseline) coeffi‐
244 cient in the Holt-Winters forecasting algorithm. See rrdtool for a
245 description of this algorithm. alpha must lie between 0 and 1. A value
246 closer to 1 means that more recent observations carry greater weight in
247 predicting the baseline component of the forecast. A value closer to 0
248 means that past history carries greater weight in predicting the base‐
249 line component.
250 beta is the adaption parameter of the slope (or linear trend) coeffi‐
252 cient in the Holt-Winters forecasting algorithm. beta must lie between
253 0 and 1 and plays the same role as alpha with respect to the predicted
254 linear trend.
255 gamma is the adaption parameter of the seasonal coefficients in the
257 Holt-Winters forecasting algorithm (HWPREDICT) or the adaption parame‐
258 ter in the exponential smoothing update of the seasonal deviations. It
259 must lie between 0 and 1. If the SEASONAL and DEVSEASONAL RRAs are cre‐
260 ated implicitly, they will both have the same value for gamma: the
261 value specified for the HWPREDICT alpha argument. Note that because
262 there is one seasonal coefficient (or deviation) for each time point
263 during the seasonal cycle, the adaptation rate is much slower than the
264 baseline. Each seasonal coefficient is only updated (or adapts) when
265 the observed value occurs at the offset in the seasonal cycle corre‐
266 sponding to that coefficient.
267 If SEASONAL and DEVSEASONAL RRAs are created explicitly, gamma need not
269 be the same for both. Note that gamma can also be changed via the RRD‐
270 tool tune command.
271 rra-num provides the links between related RRAs. If HWPREDICT is speci‐
273 fied alone and the other RRAs are created implicitly, then there is no
274 need to worry about this argument. If RRAs are created explicitly, then
275 carefully pay attention to this argument. For each RRA which includes
276 this argument, there is a dependency between that RRA and another RRA.
277 The rra-num argument is the 1-based index in the order of RRA creation
278 (that is, the order they appear in the create command). The dependent
279 RRA for each RRA requiring the rra-num argument is listed here:
280 · HWPREDICT rra-num is the index of the SEASONAL RRA.
282 · SEASONAL rra-num is the index of the HWPREDICT RRA.
284 · DEVPREDICT rra-num is the index of the DEVSEASONAL RRA.
286 · DEVSEASONAL rra-num is the index of the HWPREDICT RRA.
288 · FAILURES rra-num is the index of the DEVSEASONAL RRA.
290 threshold is the minimum number of violations (observed values outside
292 the confidence bounds) within a window that constitutes a failure. If
293 the FAILURES RRA is implicitly created, the default value is 7.
294 window length is the number of time points in the window. Specify an
296 integer greater than or equal to the threshold and less than or equal
297 to 28. The time interval this window represents depends on the inter‐
298 val between primary data points. If the FAILURES RRA is implicitly cre‐
299 ated, the default value is 9.
300
302 Here is an explanation by Don Baarda on the inner workings of RRDtool.
303 It may help you to sort out why all this *UNKNOWN* data is popping up
304 in your databases:
305 RRDtool gets fed samples at arbitrary times. From these it builds Pri‐
307 mary Data Points (PDPs) at exact times on every "step" interval. The
308 PDPs are then accumulated into RRAs.
309 The "heartbeat" defines the maximum acceptable interval between sam‐
311 ples. If the interval between samples is less than "heartbeat", then an
312 average rate is calculated and applied for that interval. If the inter‐
313 val between samples is longer than "heartbeat", then that entire inter‐
314 val is considered "unknown". Note that there are other things that can
315 make a sample interval "unknown", such as the rate exceeding limits, or
316 even an "unknown" input sample.
317 The known rates during a PDP's "step" interval are used to calculate an
319 average rate for that PDP. Also, if the total "unknown" time during the
320 "step" interval exceeds the "heartbeat", the entire PDP is marked as
321 "unknown". This means that a mixture of known and "unknown" sample
322 times in a single PDP "step" may or may not add up to enough "unknown"
323 time to exceed "heartbeat" and hence mark the whole PDP "unknown". So
324 "heartbeat" is not only the maximum acceptable interval between sam‐
325 ples, but also the maximum acceptable amount of "unknown" time per PDP
326 (obviously this is only significant if you have "heartbeat" less than
327 "step").
328 The "heartbeat" can be short (unusual) or long (typical) relative to
330 the "step" interval between PDPs. A short "heartbeat" means you require
331 multiple samples per PDP, and if you don't get them mark the PDP
332 unknown. A long heartbeat can span multiple "steps", which means it is
333 acceptable to have multiple PDPs calculated from a single sample. An
334 extreme example of this might be a "step" of 5 minutes and a "heart‐
335 beat" of one day, in which case a single sample every day will result
336 in all the PDPs for that entire day period being set to the same aver‐
337 age rate. -- Don Baarda <don.baarda@baesystems.com>
338 time|
340 axis|
341 begin__|00|
342 |01|
343 u|02|----* sample1, restart "hb"-timer
344 u|03| /
345 u|04| /
346 u|05| /
347 u|06|/ "hbt" expired
348 u|07|
349 |08|----* sample2, restart "hb"
350 |09| /
351 |10| /
352 u|11|----* sample3, restart "hb"
353 u|12| /
354 u|13| /
355 step1_u|14| /
356 u|15|/ "swt" expired
357 u|16|
358 |17|----* sample4, restart "hb", create "pdp" for step1 =
359 |18| / = unknown due to 10 "u" labled secs > "hb"
360 |19| /
361 |20| /
362 |21|----* sample5, restart "hb"
363 |22| /
364 |23| /
365 |24|----* sample6, restart "hb"
366 |25| /
367 |26| /
368 |27|----* sample7, restart "hb"
369 step2__|28| /
370 |22| /
371 |23|----* sample8, restart "hb", create "pdp" for step1, create "cdp"
372 |24| /
373 |25| /
374 graphics by vladimir.lavrov@desy.de.
376
378 Here are a few hints on how to measure:
379 Temperature
381 Usually you have some type of meter you can read to get the temper‐
382 ature. The temperature is not really connected with a time. The
383 only connection is that the temperature reading happened at a cer‐
384 tain time. You can use the GAUGE data source type for this. RRDtool
385 will then record your reading together with the time.
386 Mail Messages
388 Assume you have a method to count the number of messages trans‐
389 ported by your mailserver in a certain amount of time, giving you
390 data like '5 messages in the last 65 seconds'. If you look at the
391 count of 5 like an ABSOLUTE data type you can simply update the RRD
392 with the number 5 and the end time of your monitoring period. RRD‐
393 tool will then record the number of messages per second. If at some
394 later stage you want to know the number of messages transported in
395 a day, you can get the average messages per second from RRDtool for
396 the day in question and multiply this number with the number of
397 seconds in a day. Because all math is run with Doubles, the preci‐
398 sion should be acceptable.
399 It's always a Rate
401 RRDtool stores rates in amount/second for COUNTER, DERIVE and ABSO‐
402 LUTE data. When you plot the data, you will get on the y axis
403 amount/second which you might be tempted to convert to an absolute
404 amount by multiplying by the delta-time between the points. RRDtool
405 plots continuous data, and as such is not appropriate for plotting
406 absolute amounts as for example "total bytes" sent and received in
407 a router. What you probably want is plot rates that you can scale
408 to bytes/hour, for example, or plot absolute amounts with another
409 tool that draws bar-plots, where the delta-time is clear on the
410 plot for each point (such that when you read the graph you see for
411 example GB on the y axis, days on the x axis and one bar for each
412 day).
413
415 rrdtool create temperature.rrd --step 300 \
416 DS:temp:GAUGE:600:-273:5000 \
417 RRA:AVERAGE:0.5:1:1200 \
418 RRA:MIN:0.5:12:2400 \
419 RRA:MAX:0.5:12:2400 \
420 RRA:AVERAGE:0.5:12:2400
421 This sets up an RRD called temperature.rrd which accepts one tempera‐
423 ture value every 300 seconds. If no new data is supplied for more than
424 600 seconds, the temperature becomes *UNKNOWN*. The minimum acceptable
425 value is -273 and the maximum is 5'000.
426 A few archive areas are also defined. The first stores the temperatures
428 supplied for 100 hours (1'200 * 300 seconds = 100 hours). The second
429 RRA stores the minimum temperature recorded over every hour (12 * 300
430 seconds = 1 hour), for 100 days (2'400 hours). The third and the fourth
431 RRA's do the same for the maximum and average temperature, respec‐
432 tively.
433
435 rrdtool create monitor.rrd --step 300 \
436 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
437 RRA:AVERAGE:0.5:1:2016 \
438 RRA:HWPREDICT:1440:0.1:0.0035:288
439 This example is a monitor of a router interface. The first RRA tracks
441 the traffic flow in octets; the second RRA generates the specialized
442 functions RRAs for aberrant behavior detection. Note that the rra-num
443 argument of HWPREDICT is missing, so the other RRAs will implicitly be
444 created with default parameter values. In this example, the forecasting
445 algorithm baseline adapts quickly; in fact the most recent one hour of
446 observations (each at 5 minute intervals) accounts for 75% of the base‐
447 line prediction. The linear trend forecast adapts much more slowly.
448 Observations made during the last day (at 288 observations per day)
449 account for only 65% of the predicted linear trend. Note: these compu‐
450 tations rely on an exponential smoothing formula described in the LISA
451 2000 paper.
452 The seasonal cycle is one day (288 data points at 300 second inter‐
454 vals), and the seasonal adaption parameter will be set to 0.1. The RRD
455 file will store 5 days (1'440 data points) of forecasts and deviation
456 predictions before wrap around. The file will store 1 day (a seasonal
457 cycle) of 0-1 indicators in the FAILURES RRA.
458 The same RRD file and RRAs are created with the following command,
460 which explicitly creates all specialized function RRAs.
461 rrdtool create monitor.rrd --step 300 \
463 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
464 RRA:AVERAGE:0.5:1:2016 \
465 RRA:HWPREDICT:1440:0.1:0.0035:288:3 \
466 RRA:SEASONAL:288:0.1:2 \
467 RRA:DEVPREDICT:1440:5 \
468 RRA:DEVSEASONAL:288:0.1:2 \
469 RRA:FAILURES:288:7:9:5
470 Of course, explicit creation need not replicate implicit create, a num‐
472 ber of arguments could be changed.
473
475 rrdtool create proxy.rrd --step 300 \
476 DS:Total:DERIVE:1800:0:U \
477 DS:Duration:DERIVE:1800:0:U \
478 DS:AvgReqDur:COMPUTE:Duration,Requests,0,EQ,1,Requests,IF,/ \
479 RRA:AVERAGE:0.5:1:2016
480 This example is monitoring the average request duration during each 300
482 sec interval for requests processed by a web proxy during the interval.
483 In this case, the proxy exposes two counters, the number of requests
484 processed since boot and the total cumulative duration of all processed
485 requests. Clearly these counters both have some rollover point, but
486 using the DERIVE data source also handles the reset that occurs when
487 the web proxy is stopped and restarted.
488 In the RRD, the first data source stores the requests per second rate
490 during the interval. The second data source stores the total duration
491 of all requests processed during the interval divided by 300. The COM‐
492 PUTE data source divides each PDP of the AccumDuration by the corre‐
493 sponding PDP of TotalRequests and stores the average request duration.
494 The remainder of the RPN expression handles the divide by zero case.
495
497 Tobias Oetiker <tobi@oetiker.ch>
4981.2.27 2008-02-17 RRDCREATE(1)