TIME_VALUES:
count_integer_value
interval_in_seconds_integer_value
Here is a sample TIME_VALUES stanza generated by pmc:
TIME_VALUES:
12 10 # h=my_hostname f=7 l=17 x='' d=''
METADATA:
class_1_name class_1_type class_1_data_set_start
metric1 metric2 ... metricN
...
The METADATA stanza for an SLES system looks like this:
METADATA:
VM V 2
r b swpd free buff cache si so bi bo in cs cpu_us cpu_sy cpu_id cpu_wa st
IO A 2
tps kBrdps kBwtps kBrd kBwt
NET A 1
rxpkps txpkps rxkBps txkBps rxcmps txcmps rxmctps
The class_type (V or A) indicates whether the class (i.e., 'VM', 'IO' or 'NET') is a Vector class (without objects) or an Array class (with objects). The class data set start integer after 'V' or 'A' indicates the first data set that pma will process.
Note: The first "report" (output) from both vmstat (one line) and iostat (one line per object) displays data "since the last reboot". This data is not useful for our purpose, so by setting class data set start integer to 2, pma will skip the first data set.
However with the 'NET' class (generated by sar -n DEV), all the data ("reports") are useful, so by setting class data set start integer to 1, pma will not skip any data - it will start with set 1.
DATE:
seconds_since_the_Epoch
Here is a sample DATE stanza generated by pmc:
DATE:
1462914019 # Wed May 11 07:00:19 2016 +1000
CLASS_NAME:
metric1 metric2 ... metricN
metric1 metric2 ... metricN
...
metric1 metric2 ... metricN
CLASS_NAME:
object1 metric1 metric2 ... metricN (first row of group 1)
object2 metric1 metric2 ... metricN
...
objectX metric1 metric2 ... metricN (last row of group 1)
object1 metric1 metric2 ... metricN (first row of group 2)
object2 metric1 metric2 ... metricN
...
objectX metric1 metric2 ... metricN (last row of group 2)
...
object1 metric1 metric2 ... metricN
(first row of group count)
object2 metric1 metric2 ... metricN
...
objectX metric1 metric2 ... metricN
(last row of group count)