pmdiff (1) - Linux Man Pages
pmdiff: compares archives and report significant differences
NAMEpmdiff - compares archives and report significant differences
SYNOPSISpmdiff [-d/--keep] [-z/--hostzone] [-p/--precision precision] [-q/--threshold thres] [-S/--start starttime] [-T/--finish endtime] [-B/--begin starttime] [-E/--end endtime] [-x metric] [-X file] [--skip-excluded] [--skip-missing] [-Z/--timezone timezone] archive1 [archive2]
DESCRIPTIONpmdiff compares the average values for every metric in either one or two sets of archives, in a given time window, for changes that are likely to be of interest when searching for performance regressions.
The archive specifiers archive1 and archive2 may be comma-separated lists of names, each of which may be the base name of an archive or the name of a directory containing one or more archives. Each archive in the resulting set of archives must have been previously created using pmlogger(1). The pmlogsummary(1) utility is used to obtain the average values used for comparison.
There are two sorts of invocation of the tool: with either one or two sets of archives.
In the first case, the only sensible command line requires use of all four time window arguments. These are specified using the same time window format described in PCPIntro(1), and are -S/--start and -T/--finish for the start and end times of the first time window of interest in the archive set, and -B/--before and -E/--end for the start and end times of the second time window of interest.
In the second case, with two sets of archives, the -B/--before and -E/--end options might be unnecessary. This might be the case, for example, when comparing the same time window of two consecutive days (usually two separate sets of archives), or a time window on the same day of different weeks.
In either case, pmdiff produces a sorted summary of those metrics in the specified window whose values have deviated the most from a minimal threshold. The level of deviation is calculated by dividing the average value of each metric in both logs, and then calculating whether the ratio falls outside of a range considered normal. This ratio can be adjusted using the -q/--threshold option, and by default it is 2 (i.e. report all metrics with average values that have more than doubled in the two time windows or more than halved in the two time windows).
If the baseline value is zero and the comparison value is non-zero, the ratio is reported as ``|+|'' (infinitely large). If the comparison value is zero and the baseline value is non-zero, the ratio is reported as ``|-|'' (infinitely small).
Reported metrics are sorted in ascending ratio order.
Should any metrics be present in one window but missing from the other, a diagnostic will be displayed listing each missing metric and the archive set from which it was missing.
The remaining options control the specific information to be reported. Metrics with counter semantics are converted to rates before being evaluated.
- Print all floating point numbers with precision digits after the decimal place.
- Cull the list of names of metrics being excluded from the output.
- By default, pmdiff will report the names of any metrics that are in one archive set but not the other. This option suppresses that reporting.
- Compare each metric in each archive set in the time windows specified to a given egrep(1) pattern, excluding those that match from the report output.
- Allows a file to be specified which containing egrep(1) patterns which are applied to the metric names to optionally exclude some from the report.
- Use the local timezone from the given archive set.
- Changes the timezone in the archive labels to timezone in the format of the environment variable TZ as described in environ(7).
- Default directory for PCP archives containing performance metric values collected from the host hostname.