1 RRDCREATE(1) rrdtool RRDCREATE(1)
6 rrdcreate - Set up a new Round Robin Database
9 r\brr\brd\bdt\bto\boo\bol\bl c\bcr\bre\bea\bat\bte\be _\bf_\bi_\bl_\be_\bn_\ba_\bm_\be [-\b--\b-s\bst\bta\bar\brt\bt|-\b-b\bb _\bs_\bt_\ba_\br_\bt _\bt_\bi_\bm_\be] [-\b--\b-s\bst\bte\bep\bp|-\b-s\bs _\bs_\bt_\be_\bp]
10 [D\bDS\bS:\b:_\bd_\bs_\b-_\bn_\ba_\bm_\be:\b:_\bD_\bS_\bT:\b:_\bd_\bs_\bt _\ba_\br_\bg_\bu_\bm_\be_\bn_\bt_\bs] [R\bRR\bRA\bA:\b:_\bC_\bF:\b:_\bc_\bf _\ba_\br_\bg_\bu_\bm_\be_\bn_\bt_\bs]
13 The create function of RRDtool lets you set up new Round Robin Database
20 name.
23 Specifies the time in seconds since 1970-01-01 UTC when the
25 any data timed before or at the time specified.
28 documentation for other ways to specify time.
31 Specifies the base interval in seconds with which data will be
34 D\bDS\bS:\b:_\bd_\bs_\b-_\bn_\ba_\bm_\be:\b:_\bD_\bS_\bT:\b:_\bd_\bs_\bt _\ba_\br_\bg_\bu_\bm_\be_\bn_\bt_\bs
36 for example incoming and outgoing traffic on a specific commu-
38 some basic properties of each data source you want to store in
43 long in the characters [a-zA-Z0-9_].
46 data source entry depend on the data source type. For GAUGE,
47 COUNTER, DERIVE, and ABSOLUTE the format for a data source
48 entry is:
50 D\bDS\bS:\b:_\bd_\bs_\b-_\bn_\ba_\bm_\be:\b:_\bG_\bA_\bU_\bG_\bE _\b| _\bC_\bO_\bU_\bN_\bT_\bE_\bR _\b| _\bD_\bE_\bR_\bI_\bV_\bE _\b| _\bA_\bB_\bS_\bO_\bL_\bU_\bT_\bE:\b:_\bh_\be_\ba_\br_\bt_\b-
53 For COMPUTE data sources, the format is:
55 D\bDS\bS:\b:_\bd_\bs_\b-_\bn_\ba_\bm_\be:\b:_\bC_\bO_\bM_\bP_\bU_\bT_\bE:\b:_\br_\bp_\bn_\b-_\be_\bx_\bp_\br_\be_\bs_\bs_\bi_\bo_\bn
57 In order to decide which data source type to use, review the
58 definitions that follow. Also consult the section on "HOW TO
59 MEASURE" for further insight.
62 is for things like temperatures or number of people in a
63 room or the value of a RedHat share.
66 is for continuous incrementing counters like the ifInOctets
68 the counter never decreases, except when a counter over-
69 flows. The update function takes the overflow into
70 account. The counter is stored as a per-second rate. When
71 the counter overflows, RRDtool checks if the overflow hap-
72 pened at the 32bit or 64bit border and acts accordingly by
73 adding an appropriate value to the result.
76 will store the derivative of the line going from the last
77 to the current value of the data source. This can be useful
78 for gauges, for example, to measure the rate of people
79 entering or leaving a room. Internally, derive works
80 exactly like COUNTER but without overflow checks. So if
81 your counter does not reset at 32 or 64 bit you might want
82 to use DERIVE and combine it with a MIN value of 0.
84 NOTE on COUNTER vs DERIVE
85 by Don Baarda <don.baarda@baesystems.com>
87 If you cannot tolerate ever mistaking the occasional
88 counter reset for a legitimate counter wrap, and would
89 prefer "Unknowns" for all legitimate counter wraps and
90 resets, always use DERIVE with min=0. Otherwise, using
91 COUNTER with a suitable max will return correct values
92 for all legitimate counter wraps, mark some counter
93 resets as "Unknown", but can mistake some counter
94 resets for a legitimate counter wrap.
96 For a 5 minute step and 32-bit counter, the probability
97 of mistaking a counter reset for a legitimate wrap is
98 arguably about 0.8% per 1Mbps of maximum bandwidth.
99 Note that this equates to 80% for 100Mbps interfaces,
100 so for high bandwidth interfaces and a 32bit counter,
101 DERIVE with min=0 is probably preferable. If you are
102 using a 64bit counter, just about any max setting will
103 eliminate the possibility of mistaking a reset for a
104 counter wrap.
107 is for counters which get reset upon reading. This is used
108 for fast counters which tend to overflow. So instead of
109 reading them normally you reset them after every read to
110 make sure you have a maximum time available before the next
111 overflow. Another usage is for things you count like number
112 of messages since the last update.
115 is for storing the result of a formula applied to other
117 value on update, but rather its Primary Data Points (PDPs)
118 are computed from the PDPs of the data sources according to
119 the rpn-expression that defines the formula. Consolidation
120 functions are then applied normally to the PDPs of the COM-
121 PUTE data source (that is the rpn-expression is only
122 applied to generate PDPs). In database software, such data
123 sets are referred to as "virtual" or "computed" columns.
126 between two updates of this data source before the value of the
132 or care about min and max, set them to U for unknown. Note that
133 min and max always refer to the processed values of the DS. For
135 data-rate expected from the device.
137 _\bI_\bf _\bi_\bn_\bf_\bo_\br_\bm_\ba_\bt_\bi_\bo_\bn _\bo_\bn _\bm_\bi_\bn_\bi_\bm_\ba_\bl_\b/_\bm_\ba_\bx_\bi_\bm_\ba_\bl _\be_\bx_\bp_\be_\bc_\bt_\be_\bd _\bv_\ba_\bl_\bu_\be_\bs _\bi_\bs _\ba_\bv_\ba_\bi_\bl_\ba_\bb_\bl_\be_\b,
138 _\ba_\bl_\bw_\ba_\by_\bs _\bs_\be_\bt _\bt_\bh_\be _\bm_\bi_\bn _\ba_\bn_\bd_\b/_\bo_\br _\bm_\ba_\bx _\bp_\br_\bo_\bp_\be_\br_\bt_\bi_\be_\bs_\b. _\bT_\bh_\bi_\bs _\bw_\bi_\bl_\bl _\bh_\be_\bl_\bp _\bR_\bR_\bD_\b-
139 _\bt_\bo_\bo_\bl _\bi_\bn _\bd_\bo_\bi_\bn_\bg _\ba _\bs_\bi_\bm_\bp_\bl_\be _\bs_\ba_\bn_\bi_\bt_\by _\bc_\bh_\be_\bc_\bk _\bo_\bn _\bt_\bh_\be _\bd_\ba_\bt_\ba _\bs_\bu_\bp_\bp_\bl_\bi_\be_\bd _\bw_\bh_\be_\bn
142 _\br_\bp_\bn_\b-_\be_\bx_\bp_\br_\be_\bs_\bs_\bi_\bo_\bn defines the formula used to compute the PDPs of
143 a COMPUTE data source from other data sources in the same
145 command. Please refer to that manual page for a list and
146 description of RPN operations supported. For COMPUTE data
147 sources, the following RPN operations are not supported: COUNT,
148 PREV, TIME, and LTIME. In addition, in defining the RPN expres-
149 sion, the COMPUTE data source may only refer to the names of
150 data source listed previously in the create command. This is
165 of the archive. There are several consolidation functions that
166 consolidate primary data points via an aggregate function:
167 A\bAV\bVE\bER\bRA\bAG\bGE\bE, M\bMI\bIN\bN, M\bMA\bAX\bX, L\bLA\bAS\bST\bT. The format of R\bRR\bRA\bA line for these con-
168 solidation functions is:
170 R\bRR\bRA\bA:\b:_\bA_\bV_\bE_\bR_\bA_\bG_\bE _\b| _\bM_\bI_\bN _\b| _\bM_\bA_\bX _\b| _\bL_\bA_\bS_\bT:\b:_\bx_\bf_\bf:\b:_\bs_\bt_\be_\bp_\bs:\b:_\br_\bo_\bw_\bs
174 dated value is still regarded as known. It is given as the
176 interval. Thus, it ranges from 0 to 1 (exclusive).
178 _\bs_\bt_\be_\bp_\bs defines how many of these _\bp_\br_\bi_\bm_\ba_\br_\by _\bd_\ba_\bt_\ba _\bp_\bo_\bi_\bn_\bt_\bs are used to
179 build a _\bc_\bo_\bn_\bs_\bo_\bl_\bi_\bd_\ba_\bt_\be_\bd _\bd_\ba_\bt_\ba _\bp_\bo_\bi_\bn_\bt which then goes into the
180 archive.
185 A\bAb\bbe\ber\brr\bra\ban\bnt\bt B\bBe\beh\bha\bav\bvi\bio\bor\br D\bDe\bet\bte\bec\bct\bti\bio\bon\bn w\bwi\bit\bth\bh H\bHo\bol\blt\bt-\b-W\bWi\bin\bnt\bte\ber\brs\bs F\bFo\bor\bre\bec\bca\bas\bst\bti\bin\bng\bg
186 In addition to the aggregate functions, there are a set of specialized
188 Winters forecasting algorithm), confidence bands, and the flagging
189 aberrant behavior in the data source time series:
191 · R\bRR\bRA\bA:\b:_\bH_\bW_\bP_\bR_\bE_\bD_\bI_\bC_\bT:\b:_\br_\bo_\bw_\bs:\b:_\ba_\bl_\bp_\bh_\ba:\b:_\bb_\be_\bt_\ba:\b:_\bs_\be_\ba_\bs_\bo_\bn_\ba_\bl _\bp_\be_\br_\bi_\bo_\bd[:\b:_\br_\br_\ba_\b-_\bn_\bu_\bm]
193 · R\bRR\bRA\bA:\b:_\bS_\bE_\bA_\bS_\bO_\bN_\bA_\bL:\b:_\bs_\be_\ba_\bs_\bo_\bn_\ba_\bl _\bp_\be_\br_\bi_\bo_\bd:\b:_\bg_\ba_\bm_\bm_\ba:\b:_\br_\br_\ba_\b-_\bn_\bu_\bm
195 · R\bRR\bRA\bA:\b:_\bD_\bE_\bV_\bS_\bE_\bA_\bS_\bO_\bN_\bA_\bL:\b:_\bs_\be_\ba_\bs_\bo_\bn_\ba_\bl _\bp_\be_\br_\bi_\bo_\bd:\b:_\bg_\ba_\bm_\bm_\ba:\b:_\br_\br_\ba_\b-_\bn_\bu_\bm
197 · R\bRR\bRA\bA:\b:_\bD_\bE_\bV_\bP_\bR_\bE_\bD_\bI_\bC_\bT:\b:_\br_\bo_\bw_\bs:\b:_\br_\br_\ba_\b-_\bn_\bu_\bm
199 · R\bRR\bRA\bA:\b:_\bF_\bA_\bI_\bL_\bU_\bR_\bE_\bS:\b:_\br_\bo_\bw_\bs:\b:_\bt_\bh_\br_\be_\bs_\bh_\bo_\bl_\bd:\b:_\bw_\bi_\bn_\bd_\bo_\bw _\bl_\be_\bn_\bg_\bt_\bh:\b:_\br_\br_\ba_\b-_\bn_\bu_\bm
204 confidence bounds, a matched set of HWPREDICT, SEASONAL, DEVSEASONAL,
205 and DEVPREDICT must exist. Generating smoothed values of the primary
207 behavior detection requires FAILURES, HWPREDICT, DEVSEASONAL, and SEA-
208 SONAL.
210 The actual predicted, or smoothed, values are stored in the HWPREDICT
212 dard deviation which can be scaled to yield a confidence band). The
214 tion as failure; that is, the number of confidence bounds violations in
215 the preceding window of observations met or exceeded a specified
217 and failures appears in rrdgraph.
220 the Holt-Winters forecasting algorithm and the seasonal deviations,
221 respectively. There is one entry per observation time point in the
222 seasonal cycle. For example, if primary data points are generated every
223 five minutes and the seasonal cycle is 1 day, both SEASONAL and DEVSEA-
224 SONAL will have 288 rows.
226 In order to simplify the creation for the novice user, in addition to
227 supporting explicit creation of the HWPREDICT, SEASONAL, DEVPREDICT,
228 DEVSEASONAL, and FAILURES R\bRR\bRA\bAs\bs, the R\bRR\bRD\bDt\bto\boo\bol\bl create command supports
229 implicit creation of the other four when HWPREDICT is specified alone
233 that there is a one-to-one correspondence between primary data points
235 than the _\bs_\be_\ba_\bs_\bo_\bn_\ba_\bl _\bp_\be_\br_\bi_\bo_\bd. If the DEVPREDICT R\bRR\bRA\bA is implicitly created,
237 If the FAILURES R\bRR\bRA\bA is implicitly created, _\br_\bo_\bw_\bs will be set to the _\bs_\be_\ba_\b-
238 _\bs_\bo_\bn_\ba_\bl _\bp_\be_\br_\bi_\bo_\bd argument of the HWPREDICT R\bRR\bRA\bA. Of course, the R\bRR\bRD\bDt\bto\boo\bol\bl
240 the creator wishes to avoid explicit creations of the other specialized
243 _\bs_\be_\ba_\bs_\bo_\bn_\ba_\bl _\bp_\be_\br_\bi_\bo_\bd specifies the number of primary data points in a sea-
244 sonal cycle. If SEASONAL and DEVSEASONAL are implicitly created, this
246 HWPREDICT. If they are explicitly created, the creator should verify
250 cient in the Holt-Winters forecasting algorithm. See rrdtool for a
252 closer to 1 means that more recent observations carry greater weight in
253 predicting the baseline component of the forecast. A value closer to 0
254 means that past history carries greater weight in predicting the base-
255 line component.
260 linear trend.
263 Holt-Winters forecasting algorithm (HWPREDICT) or the adaption parame-
264 ter in the exponential smoothing update of the seasonal deviations. It
268 there is one seasonal coefficient (or deviation) for each time point
269 during the seasonal cycle, the adaptation rate is much slower than the
270 baseline. Each seasonal coefficient is only updated (or adapts) when
271 the observed value occurs at the offset in the seasonal cycle corre-
272 sponding to that coefficient.
274 If SEASONAL and DEVSEASONAL R\bRR\bRA\bAs\bs are created explicitly, _\bg_\ba_\bm_\bm_\ba need not
275 be the same for both. Note that _\bg_\ba_\bm_\bm_\ba can also be changed via the R\bRR\bRD\bD-\b-
278 _\br_\br_\ba_\b-_\bn_\bu_\bm provides the links between related R\bRR\bRA\bAs\bs. If HWPREDICT is speci-
283 The _\br_\br_\ba_\b-_\bn_\bu_\bm argument is the 1-based index in the order of R\bRR\bRA\bA creation
285 R\bRR\bRA\bA for each R\bRR\bRA\bA requiring the _\br_\br_\ba_\b-_\bn_\bu_\bm argument is listed here:
297 _\bt_\bh_\br_\be_\bs_\bh_\bo_\bl_\bd is the minimum number of violations (observed values outside
298 the confidence bounds) within a window that constitutes a failure. If
301 _\bw_\bi_\bn_\bd_\bo_\bw _\bl_\be_\bn_\bg_\bt_\bh is the number of time points in the window. Specify an
302 integer greater than or equal to the threshold and less than or equal
303 to 28. The time interval this window represents depends on the inter-
305 ated, the default value is 9.
308 Here is an explanation by Don Baarda on the inner workings of RRDtool.
309 It may help you to sort out why all this *UNKNOWN* data is popping up
310 in your databases:
312 RRDtool gets fed samples at arbitrary times. From these it builds Pri-
313 mary Data Points (PDPs) at exact times on every "step" interval. The
314 PDPs are then accumulated into RRAs.
316 The "heartbeat" defines the maximum acceptable interval between sam-
317 ples. If the interval between samples is less than "heartbeat", then an
318 average rate is calculated and applied for that interval. If the inter-
319 val between samples is longer than "heartbeat", then that entire inter-
320 val is considered "unknown". Note that there are other things that can
321 make a sample interval "unknown", such as the rate exceeding limits, or
322 even an "unknown" input sample.
324 The known rates during a PDP's "step" interval are used to calculate an
325 average rate for that PDP. Also, if the total "unknown" time during the
326 "step" interval exceeds the "heartbeat", the entire PDP is marked as
327 "unknown". This means that a mixture of known and "unknown" sample
328 times in a single PDP "step" may or may not add up to enough "unknown"
329 time to exceed "heartbeat" and hence mark the whole PDP "unknown". So
330 "heartbeat" is not only the maximum acceptable interval between sam-
331 ples, but also the maximum acceptable amount of "unknown" time per PDP
332 (obviously this is only significant if you have "heartbeat" less than
333 "step").
335 The "heartbeat" can be short (unusual) or long (typical) relative to
336 the "step" interval between PDPs. A short "heartbeat" means you require
337 multiple samples per PDP, and if you don't get them mark the PDP
338 unknown. A long heartbeat can span multiple "steps", which means it is
339 acceptable to have multiple PDPs calculated from a single sample. An
340 extreme example of this might be a "step" of 5 minutes and a "heart-
341 beat" of one day, in which case a single sample every day will result
342 in all the PDPs for that entire day period being set to the same aver-
343 age rate. _\b-_\b- _\bD_\bo_\bn _\bB_\ba_\ba_\br_\bd_\ba _\b<_\bd_\bo_\bn_\b._\bb_\ba_\ba_\br_\bd_\ba_\b@_\bb_\ba_\be_\bs_\by_\bs_\bt_\be_\bm_\bs_\b._\bc_\bo_\bm_\b>
345 time|
346 axis|
347 begin__|00|
348 |01|
349 u|02|----* sample1, restart "hb"-timer
350 u|03| /
351 u|04| /
352 u|05| /
353 u|06|/ "hbt" expired
354 u|07|
355 |08|----* sample2, restart "hb"
356 |09| /
357 |10| /
358 u|11|----* sample3, restart "hb"
359 u|12| /
360 u|13| /
361 step1_u|14| /
362 u|15|/ "swt" expired
363 u|16|
364 |17|----* sample4, restart "hb", create "pdp" for step1 =
365 |18| / = unknown due to 10 "u" labled secs > "hb"
366 |19| /
367 |20| /
368 |21|----* sample5, restart "hb"
369 |22| /
370 |23| /
371 |24|----* sample6, restart "hb"
372 |25| /
373 |26| /
374 |27|----* sample7, restart "hb"
375 step2__|28| /
376 |22| /
377 |23|----* sample8, restart "hb", create "pdp" for step1, create "cdp"
378 |24| /
379 |25| /
381 graphics by _\bv_\bl_\ba_\bd_\bi_\bm_\bi_\br_\b._\bl_\ba_\bv_\br_\bo_\bv_\b@_\bd_\be_\bs_\by_\b._\bd_\be.
384 Here are a few hints on how to measure:
386 Temperature
387 Usually you have some type of meter you can read to get the temper-
388 ature. The temperature is not really connected with a time. The
389 only connection is that the temperature reading happened at a cer-
391 will then record your reading together with the time.
393 Mail Messages
394 Assume you have a method to count the number of messages trans-
395 ported by your mailserver in a certain amount of time, giving you
396 data like '5 messages in the last 65 seconds'. If you look at the
398 with the number 5 and the end time of your monitoring period. RRD-
399 tool will then record the number of messages per second. If at some
400 later stage you want to know the number of messages transported in
401 a day, you can get the average messages per second from RRDtool for
402 the day in question and multiply this number with the number of
403 seconds in a day. Because all math is run with Doubles, the preci-
404 sion should be acceptable.
406 It's always a Rate
407 RRDtool stores rates in amount/second for COUNTER, DERIVE and ABSO-
408 LUTE data. When you plot the data, you will get on the y axis
409 amount/second which you might be tempted to convert to an absolute
410 amount by multiplying by the delta-time between the points. RRDtool
411 plots continuous data, and as such is not appropriate for plotting
412 absolute amounts as for example "total bytes" sent and received in
413 a router. What you probably want is plot rates that you can scale
414 to bytes/hour, for example, or plot absolute amounts with another
415 tool that draws bar-plots, where the delta-time is clear on the
416 plot for each point (such that when you read the graph you see for
417 example GB on the y axis, days on the x axis and one bar for each
418 day).
421 rrdtool create temperature.rrd --step 300 \
422 DS:temp:GAUGE:600:-273:5000 \
423 RRA:AVERAGE:0.5:1:1200 \
424 RRA:MIN:0.5:12:2400 \
425 RRA:MAX:0.5:12:2400 \
426 RRA:AVERAGE:0.5:12:2400
428 This sets up an R\bRR\bRD\bD called _\bt_\be_\bm_\bp_\be_\br_\ba_\bt_\bu_\br_\be_\b._\br_\br_\bd which accepts one tempera-
429 ture value every 300 seconds. If no new data is supplied for more than
430 600 seconds, the temperature becomes _\b*_\bU_\bN_\bK_\bN_\bO_\bW_\bN_\b*. The minimum acceptable
431 value is -273 and the maximum is 5'000.
433 A few archive areas are also defined. The first stores the temperatures
434 supplied for 100 hours (1'200 * 300 seconds = 100 hours). The second
435 RRA stores the minimum temperature recorded over every hour (12 * 300
436 seconds = 1 hour), for 100 days (2'400 hours). The third and the fourth
437 RRA's do the same for the maximum and average temperature, respec-
438 tively.
441 rrdtool create monitor.rrd --step 300 \
442 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
443 RRA:AVERAGE:0.5:1:2016 \
444 RRA:HWPREDICT:1440:0.1:0.0035:288
448 functions R\bRR\bRA\bAs\bs for aberrant behavior detection. Note that the _\br_\br_\ba_\b-_\bn_\bu_\bm
450 created with default parameter values. In this example, the forecasting
451 algorithm baseline adapts quickly; in fact the most recent one hour of
452 observations (each at 5 minute intervals) accounts for 75% of the base-
453 line prediction. The linear trend forecast adapts much more slowly.
454 Observations made during the last day (at 288 observations per day)
455 account for only 65% of the predicted linear trend. Note: these compu-
456 tations rely on an exponential smoothing formula described in the LISA
457 2000 paper.
459 The seasonal cycle is one day (288 data points at 300 second inter-
460 vals), and the seasonal adaption parameter will be set to 0.1. The RRD
461 file will store 5 days (1'440 data points) of forecasts and deviation
462 predictions before wrap around. The file will store 1 day (a seasonal
468 rrdtool create monitor.rrd --step 300 \
469 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
470 RRA:AVERAGE:0.5:1:2016 \
471 RRA:HWPREDICT:1440:0.1:0.0035:288:3 \
472 RRA:SEASONAL:288:0.1:2 \
473 RRA:DEVPREDICT:1440:5 \
474 RRA:DEVSEASONAL:288:0.1:2 \
475 RRA:FAILURES:288:7:9:5
477 Of course, explicit creation need not replicate implicit create, a num-
478 ber of arguments could be changed.
481 rrdtool create proxy.rrd --step 300 \
482 DS:Total:DERIVE:1800:0:U \
483 DS:Duration:DERIVE:1800:0:U \
484 DS:AvgReqDur:COMPUTE:Duration,Requests,0,EQ,1,Requests,IF,/ \
485 RRA:AVERAGE:0.5:1:2016
487 This example is monitoring the average request duration during each 300
488 sec interval for requests processed by a web proxy during the interval.
489 In this case, the proxy exposes two counters, the number of requests
490 processed since boot and the total cumulative duration of all processed
491 requests. Clearly these counters both have some rollover point, but
492 using the DERIVE data source also handles the reset that occurs when
493 the web proxy is stopped and restarted.
496 during the interval. The second data source stores the total duration
497 of all requests processed during the interval divided by 300. The COM-
498 PUTE data source divides each PDP of the AccumDuration by the corre-
499 sponding PDP of TotalRequests and stores the average request duration.
500 The remainder of the RPN expression handles the divide by zero case.
503 Tobias Oetiker <tobi@oetiker.ch>
507 1.2.26 2007-11-20 RRDCREATE(1)