rrdfetch - fetch data from an RRD.
rrdtool fetch filename CF [--resolution|-r resolution] [--start|-s start] [--end|-e end]
The fetch function is normally used internally by the graph function, to get data from RRDs. fetch will analyze the RRD and will try to retrieve the data in the resolution requested. The data fetched is printed to stdout. *UNKNOWN* data is often represented by the string ``NaN'' depending on your OSs printf function.
In order to get RRDTool to fetch anything other than the finest resolution RRA both the start and end time must be specified on boundaries that are multiples of the wanted resolution. Consider the following example:
rrdtool create subdata.rrd -s 10 DS:ds0:GAUGE:300:0:U \ RRA:AVERAGE:0.5:30:3600 \ RRA:AVERAGE:0.5:90:1200 \ RRA:AVERAGE:0.5:360:1200 \ RRA:MAX:0.5:360:1200 \ RRA:AVERAGE:0.5:8640:600 \ RRA:MAX:0.5:8640:600
This RRD collects data every 10 seconds and stores its averages over 5 minutes, 15 minutes, 1 hour and 1 day as well as the maxima for 1 hour and 1 day.
Consider now that you want too fetch the 15 minute average data for last hour. So you might try
rrdtool fetch subdata.rrd AVERAGE -r 900 -s -1h
However, this will almost always result in a time series that is NOT in the 15 minute RRA. Therefore the highest resolution RRA, i.e. 5 minute averages, will be chosen which, in this case, is not what you want.
Hence, make sure that
So, if time now is called ``t'',
do end time == int(t/900)*900, start time == end time -1hour, resolution == 900.
In e.g. bash this could look as:
TIME=$(date +%s); RRDRES=900; rrdtool fetch subdata.rrd AVERAGE -r $RRDRES \ -e $(echo $(($TIME/$RRDRES*$RRDRES))) -s e-1h
Or in Perl:
perl -e '$ctime = time; $rrdres = 900; system "rrdtool fetch subdata.rrd AVERAGE \
-r $rrdres -e @{[int($ctime/$rrdres)*$rrdres]} -s e-1h"'
Apart from the traditional Seconds since epoch, RRDTool does also
understand at-style time specification. The specification is called
``at-style'' after Unix command at(1) that has moderately complex ways
to specify time to run your job at. The at-style specification
consists of two parts: TIME REFERENCE specification and TIME
OFFSET specification.
Time reference specification is used, well,... to establish a reference moment in time (for time offset to be applied to). When present, it should come first, when omitted, it defaults to now. On its own part, time reference consists of time-of-day reference (which should come first, if present) and day reference.
Time-of-day can be specified as HH:MM, HH.MM, or just HH, you can suffix it with am or pm or use 24-hours clock. The few special times of day are understood as well, these include midnight (00:00), noon (12:00) and British teatime (16:00).
The day can be specified as month-name day-of-the-month and optional 2- or 4-digit year number (e.g. March 8 1999). Alternatively, you can use day-of-week-name (e.g. Monday), or one of the words: yesterday, today, tomorrow. You can also specify day as a full date in several numerical formats; these include: MM/DD/[YY]YY, DD.MM.[YY]YY, YYYYMMDD.
NOTE1: this is different from the original at(1) behavior,
which interprets a single-number date as MMDD[YY]YY.
NOTE2: if you specify day this way, the time-of-day is REQUIRED to be present.
Finally, you can use words now, start, or end as your time reference. Now refers to the current moment (and is also a default time reference). Start (end) can be used to specify time relative to the start (end) time for those tools that use these categories (rrdfetch, the rrdgraph manpage).
Month and weekday names can be used in their naturally abbreviated form (e.g., Dec for December, Sun for Sunday, etc.). The words now, start, end can be abbreviated to n, s, e.
Time offset specification is used to add (or subtract) certain time interval to (from) the time reference moment. It consists of sign (+ or -) and amount. The following time units can be used to specify the amount: years, months, weeks, days, hours, minutes, seconds, these can be used in singular or plural form, and abbreviated naturally or to a single letter (e.g. +3days, -1wk, -3y). Several time units can be combined together (e.g., -5mon1w2d), as well as several time offsets can be concatenated (e.g., -5h45min = -5h-45min = -6h+15min = -7h+1h30m-15min, etc.)
NOTE3: If you specify time offset in days, weeks, months, or years,
you will end with the time offset that may vary depending on you time
reference, because all those time units have no single well defined
time interval value (1 year contains either 365 or 366 days, 1 month
is 28 to 31 days long, and even 1 day may be not equal to 24 hours
twice a year, when DST-related clock adjustments take place).
To cope with this, when you use days, weeks, months, or years
as your time offset units your time reference date is adjusted
accordingly without taking too much further effort to ensure anything
about it (in the hope that mktime(3) will take care of this later).
This may lead to some surprising (or even invalid!) results,
e.g. 'May 31 -1month' = 'Apr 31' (meaningless) = 'May 1'
(after mktime(3) normalization); in the EET timezone
'3:30am Mar 29 1999 -1 day' yields '3:30am Mar 28 1999' (Sunday)
which is invalid time/date combination (because of 3am -> 4am DST
forward clock adjustment, see the below example).
On the other hand, hours, minutes, and seconds are well defined time
intervals, and these are guaranteed to always produce time offsets
exactly as specified (e.g. for EET timezone, '8:00 Mar 27 1999 +2 days' =
'8:00 Mar 29 1999', but since there is 1-hour DST forward clock adjustment
takes place around 3:00 Mar 28 1999, the actual time interval between
8:00 Mar 27 1999 and 8:00 Mar 29 1999 equals 47 hours; on the other hand,
'8:00 Mar 27 1999 +48 hours' = '9:00 Mar 29 1999', as expected)
NOTE4: The single-letter abbreviation for both months and minutes is m. To disambiguate, the parser tries to read your mind :) by applying the following two heuristics:
Final NOTES: Time specification is case-insensitive. Whitespace can be inserted freely or omitted altogether, there are, however, cases when whitespace is required (e.g., 'midnight Thu'). In this case you should either quote the whole phrase to prevent it from being taken apart by your shell or use '_' (underscore) or ',' (comma) which also count as whitespace (e.g., midnight_Thu or midnight,Thu)
Oct 12 -- October 12 this year
-1month or -1m -- current time of day, only a month before (may yield surprises, see the NOTE3 above)
noon yesterday -3hours -- yesterday morning; can be put also as 9am-1day
23:59 31.12.1999 -- 1 minute to the year 2000
12/31/99 11:59pm -- 1 minute to the year 2000 for imperialists
12am 01/01/01 -- start of the new millennium
end-3weeks or e-3w -- 3 weeks before end time (may be used as start time specification)
start+6hours or s+6h -- 6 hours after start time (may be used as end time specification)
931225537 -- 18:45 July 5th, 1999 (yes, seconds since 1970 are valid as well)
19970703 12:45 -- 12:45 July 3th, 1997 (my favorite, and its even got an ISO number (8601))
Tobias Oetiker <oetiker@ee.ethz.ch>