|
NAME | SYNOPSIS | DESCRIPTION | OPTIONS | REWRITING RULES SYNTAX | EXAMPLES | FILES | PCP ENVIRONMENT | SEE ALSO | DIAGNOSTICS | COLOPHON |
|
PMLOGREWRITE(1) General Commands Manual PMLOGREWRITE(1)
pmlogrewrite - rewrite Performance Co-Pilot archives
$PCP_BINADM_DIR/pmlogrewrite [-Cdiqsvw ] [-c config] inlog [outlog]
pmlogrewrite reads a set of Performance Co-Pilot (PCP) archive logs
identified by inlog and creates a PCP archive log in outlog. Under
normal usage, the -c option will be used to nominate a configuration
file or files that contains specifications (see the REWRITING RULES
SYNTAX section below) that describe how the data and metadata from
inlog should be transformed to produce outlog.
The typical uses for pmlogrewrite would be to accommodate the
evolution of Performance Metric Domain Agents (PMDAs) where the
names, metadata and semantics of metrics and their associated
instance domains may change over time, e.g. promoting the type of a
metric from a 32-bit to a 64-bit integer, or renaming a group of
metrics. Refer to the EXAMPLES section for some additional use
cases.
pmlogrewrite is most useful where PMDA changes, or errors in the
production environment, result in archives that cannot be combined
with pmlogextract(1). By pre-processing the archives with
pmlogrewrite the resulting archives may be able to be merged with
pmlogextract(1).
The input inlog must be a set of PCP archive logs created by
pmlogger(1), or possibly one of the tools that read and create PCP
archives, e.g. pmlogextract(1) and pmlogreduce(1). inlog is a
comma-separated list of names, each of which may be the base name of
an archive or the name of a directory containing one or more
archives.
If no -c option is specified, then the default behavior simply
creates outlog as a copy of inlog. This is a little more complicated
than cat(1), as each PCP archive is made up of several physical
files.
While pmlogrewrite may be used to repair some data consistency issues
in PCP archives, there is also a class of repair tasks that cannot be
handled by pmlogrewrite and pmloglabel(1) may be a useful tool in
these cases.
The command line options for pmlogrewrite are as follows:
-C Parse the rewriting rules and quit. outlog is not created.
When -C is specified, this also sets -v and -w so that all
warnings and verbose messages are displayed as config is
parsed.
-c config
If config is a file or symbolic link, read and parse rewriting
rules from there. If config is a directory, then all of the
files or symbolic links in that directory (excluding those
beginning with a period ``.'') will be used to provide the
rewriting rules. Multiple -c options are allowed.
-d Desperate mode. Normally if a fatal error occurs, all trace
of the partially written PCP archive outlog is removed. With
the -d option, the partially created outlog archive log is not
removed.
-i Rather than creating outlog, inlog is rewritten in place when
the -i option is used. A new archive is created using
temporary file names and then renamed to inlog in such a way
that if any errors (not warnings) are encountered, inlog
remains unaltered.
-q Quick mode, where if there are no rewriting actions to be
performed (none of the global data, instance domains or
metrics from inlog will be changed), then pmlogrewrite will
exit (with status 0, so success) immediately after parsing the
configuration file(s) and outlog is not created.
-s When the ``units'' of a metric are changed, if the dimension
in terms of space, time and count is unaltered, then the
scaling factor is being changed, e.g. BYTE to KBYTE, or MSEC-1
to USEC-1, or the composite MBYTE.SEC-1 to KBYTE.USEC-1. The
motivation may be (a) that the original metadata was wrong but
the values in inlog are correct, or (b) the metadata is
changing so the values need to change as well. The default
pmlogrewrite behaviour matches case (a). If case (b) applies,
then use the -s option and the values of all the metrics with
a scale factor change in each result will be rescaled. For
finer control over value rescaling refer to the RESCALE option
for the UNITS clause of the metric rewriting rule described
below.
-v Increase verbosity of diagnostic output.
-w Emit warnings. Normally pmlogrewrite remains silent for any
warning that is not fatal and it is expected that for a
particular archive, some (or indeed, all) of the rewriting
specifications may not apply. For example, changes to a PMDA
may be captured in a set of rewriting rules, but a single
archive may not contain all of the modified metrics nor all of
the modified instance domains and/or instances. Because these
cases are expected, they do not prevent pmlogrewrite
executing, and rules that do not apply to inlog are silently
ignored by default. Similarly, some rewriting rules may
involve no change because the metadata in inlog already
matches the intent of the rewriting rule to correct data from
a previous version of a PMDA. The -w flag forces warnings to
be emitted for all of these cases.
The argument outlog is required in all cases, except when -i is
specified.
A configuration file contains zero or more rewriting rules as defined
below.
Keywords and special punctuation characters are shown below in
bolditalic font and are case-insensitive, so METRIC, metric and
Metric are all equivalent in rewriting rules.
The character ``#'' introduces a comment and the remainder of the
line is ignored. Otherwise the input is relatively free format with
optional white space (spaces, tabs or newlines) between lexical items
in the rules.
A global rewriting rule has the form:
GLOBAL { globalspec ... }
where globalspec is zero or more of the following clauses:
HOSTNAME -> hostname
Modifies the label records in the outlog PCP archive, so that
the metrics will appear to have been collected from the host
hostname.
TIME -> delta
Both metric values and the instance domain metadata in a PCP
archive carry timestamps. This clause forces all the
timestamps to be adjusted by delta, where delta is an
optional sign ``+'' (the default) or ``-'', an optional
number of hours followed by a colon ``:'', an optional number
of minutes followed by a colon ``:'', a number of seconds, an
optional fraction of seconds following a period ``.''. The
simplest example would be ``30'' to increase the timestamps
by 30 seconds. A more complex example would be
``-23:59:59.999'' to move the timestamps backwards by one
millisecond less than one day.
TZ -> "timezone"
Modifies the label records in the outlog PCP archive, so that
the metrics will appear to have been collected from a host
with a local timezone of timezone. timezone must be enclosed
in quotes, and should conform to the valid timezone syntax
rules for the local platform.
An indom rewriting rule modifies an instance domain and has the form:
INDOM domain.serial { indomspec ... }
where domain and serial identify one or more existing instance
domains from inlog - typically domain would be an integer in the
range 1 to 510 and serial would be an integer in the range 0 to
4194304.
As a special case serial could be an asterisk ``*'' which means the
rule applies to every instance domain with a domain number of domain.
If a designated instance domain is not in inlog the rule has no
effect.
The indomspec is zero or more of the following clauses:
INAME "oldname" -> "newname"
The instance identified by the external instance name oldname
is renamed to newname. Both oldname and newname must be
enclosed in quotes.
As a special case, the new name may be the keyword DELETE
(with no quotes), and then the instance oldname will be
expunged from outlog which removes it from the instance
domain metadata and removes all values of this instance for
all the associated metrics.
If the instance names contain any embedded spaces then
special care needs to be taken in respect of the PCP instance
naming rule that treats the leading non-space part of the
instance name as the unique portion of the name for the
purposes of matching and ensuring uniqueness within an
instance domain, refer to pmdaInstance(3) for a discussion of
this issue.
As an illustration, consider the hypothetical instance domain
for a metric which contains 2 instances with the following
names:
red
eek urk
Then some possible INAME clauses might be:
"eek" -> "yellow like a flower"
Acceptable, oldname "eek" matches the "eek urk"
instance.
"red" -> "eek"
Error, newname "eek" matches the existing "eek urk"
instance.
"eek urk" -> "red of another hue"
Error, newname "red of another hue" matches the
existing "red" instance.
INDOM -> newdomain.newserial
Modifies the metadata for the instance domain and every
metric associated with the instance domain. As a special
case, newserial could be an asterisk ``*'' which means use
serial from the indom rewriting rule, although this is most
useful when serial is also an asterisk. So for example:
indom 29.* { indom -> 109.* }
will move all instance domains from domain 29 to domain 109.
INDOM -> DUPLICATE newdomain.newserial
A special case of the previous INDOM clause where the
instance domain is a duplicate copy of the domain.serial
instance domain from the indom rewriting rule, and then any
mapping rules are applied to the copied newdomain.newserial
instance domain. This is useful when a PMDA is split and the
same instance domain needs to be replicated for domain domain
and domain newdomain. So for example if the metrics foo.one
and foo.two are both defined over instance domain 12.34, and
foo.two is moved to another PMDA using domain 27, then the
following rewriting rules could be used:
indom 12.34 { indom -> duplicate 27.34 }
metric foo.two { indom -> 27.34 pmid -> 27.*.* }
INST oldid -> newid
The instance identified by the internal instance identifier
oldid is renumbered to newid. Both oldid and newid are
integers in the range 0 to 231-1.
As a special case, newid may be the keyword DELETE and then
the instance oldid will be expunged from outlog which removes
it from the instance domain metadata and removes all values
of this instance for all the associated metrics.
A metric rewriting rule has the form:
METRIC metricid { metricspec ... }
where metricid identifies one or more existing metrics from inlog
using either a metric name, or the internal encoding for a metric's
PMID as domain.cluster.item. In the latter case, typically domain
would be an integer in the range 1 to 510, cluster would be an
integer in the range 0 to 4095, and item would be an integer in the
range 0 to 1023.
As special cases item could be an asterisk ``*'' which means the rule
applies to every metric with a domain number of domain and a cluster
number of cluster, or cluster could be an asterisk which means the
rule applies to every metric with a domain number of domain and an
item number of item, or both cluster and item could be asterisks, and
rule applies to every metric with a domain number of domain.
If a designated metric is not in inlog the rule has no effect.
The metricspec is zero or more of the following clauses:
DELETE
The metric is completely removed from outlog, both the
metadata and all values in results are expunged.
INDOM -> newdomain.newserial [ pick ]
Modifies the metadata to change the instance domain for this
metric. The new instance domain must exist in outlog.
The optional pick clause may be used to select one input
value, or compute an aggregate value from the instances in an
input result, or assign an internal instance identifier to a
single output value. If no pick clause is specified, the
default behaviour is to copy all input values from each input
result to an output result, however if the input instance
domain is singular (indom PM_INDOM_NULL) then the one output
value must be assigned an internal instance identifier, which
is 0 by default, unless over-ridden by a INST or INAME clause
as defined below.
The choices for pick are as follows:
OUTPUT FIRST
choose the value of the first instance from each
input result
OUTPUT LAST choose the value of the last instance from each
input result
OUTPUT INST instid
choose the value of the instance with internal
instance identifier instid from each result; the
sequence of rewriting rules ensures the OUTPUT
processing happens before instance identifier
renumbering from any associated indom rule, so
instid should be one of the internal instance
identifiers that appears in inlog
OUTPUT INAME "name"
choose the value of the instance with name for
its external instance name from each result; the
sequence of rewriting rules ensures the OUTPUT
processing happens before instance renaming from
any associated indom rule, so name should be one
of the external instance names that appears in
inlog
OUTPUT MIN choose the smallest value in each result (metric
type must be numeric and output instance will be
0 for a non-singular instance domain)
OUTPUT MAX choose the largest value in each result (metric
type must be numeric and output instance will be
0 for a non-singular instance domain)
OUTPUT SUM choose the sum of all values in each result
(metric type must be numeric and output instance
will be 0 for a non-singular instance domain)
OUTPUT AVG choose the average of all values in each result
(metric type must be numeric and output instance
will be 0 for a non-singular instance domain)
If the input instance domain is singular (indom
PM_INDOM_NULL) then independent of any pick specifications,
there is at most one value in each input result and so FIRST,
LAST, MIN, MAX, SUM and AVG are all equivalent and the output
instance identifier will be 0.
In general it is an error to specify a rewriting action for
the same metadata or result values more than once, e.g. more
than one INDOM clause for the same instance domain. The one
exception is the possible interaction between the INDOM
clauses in the indom and metric rules. For example the
metric sample.bin is defined over the instance domain 29.2 in
inlog and the following is acceptable (albeit redundant):
indom 29.* { indom -> 109.* }
metric sample.bin { indom -> 109.2 }
However the following is an error, because the instance
domain for sample.bin has two conflicting definitions:
indom 29.* { indom -> 109.* }
metric sample.bin { indom -> 123.2 }
INDOM -> NULL[ pick ]
The metric (which must have been previously defined over an
instance domain) is being modified to be a singular metric.
This involves a metadata change and collapsing all results
for this metric so that multiple values become one value.
The optional pick part of the clause defines how the one
value for each result should be calculated and follows the
same rules as described for the non-NULL INDOM case above.
In the absence of pick, the default is OUTPUT FIRST.
NAME -> newname
Renames the metric in the PCP archive's metadata that
supports the Performance Metrics Name Space (PMNS). newname
should not match any existing name in the archive's PMNS and
must follow the syntactic rules for valid metric names as
outlined in pmns(5).
PMID -> newdomain.newcluster.newitem
Modifies the metadata and results to renumber the metric's
PMID. As special cases, newcluster could be an asterisk
``*'' which means use cluster from the metric rewriting rule
and/or item could be an asterisk which means use item from
the metric rewriting rule. This is most useful when cluster
and/or item is also an asterisk. So for example:
metric 30.*.* { pmid -> 123.*.* }
will move all metrics from domain 30 to domain 123.
SEM -> newsem
Change the semantics of the metric. newsem should be the XXX
part of the name of one of the PM_SEM_XXX macros defined in
<pcp/pmapi.h> or pmLookupDesc(3), e.g. COUNTER for
PM_TYPE_COUNTER.
No data value rewriting is performed as a result of the SEM
clause, so the usefulness is limited to cases where a version
of the associated PMDA was exporting incorrect semantics for
the metric. pmlogreduce(1) may provide an alternative in
cases where re-computation of result values is desired.
TYPE -> newtype
Change the type of the metric which alters the metadata and
may change the encoding of values in results. newtype should
be the XXX part of the name of one of the PM_TYPE_XXX macros
defined in <pcp/pmapi.h> or pmLookupDesc(3), e.g. FLOAT for
PM_TYPE_FLOAT.
Type conversion is only supported for cases where the old and
new metric type is numeric, so PM_TYPE_STRING,
PM_TYPE_AGGREGATE and PM_TYPE_EVENT are not allowed. Even
for the numeric cases, some conversions may produce run-time
errors, e.g. integer overflow, or attempting to rewrite a
negative value into an unsigned type.
TYPE IF oldtype -> newtype
The same as the preceding TYPE clause, except the type of the
metric is only changed to newtype if the type of the metric
in inlog is oldtype.
This useful in cases where the type of metricid in inlog may
be platform dependent and so more than one type rewriting
rule is required.
UNITS -> newunits [ RESCALE ]
newunits is six values separated by commas. The first 3
values describe the dimension of the metric along the
dimensions of space, time and count; these are integer
values, usually 0, 1 or -1. The remaining 3 values describe
the scale of the metric's values in the dimensions of space,
time and count. Space scale values should be 0 (if the space
dimension is 0), else the XXX part of the name of one of the
PM_SPACE_XXX macros, e.g. KBYTE for PM_TYPE_KBYTE. Time
scale values should be 0 (if the time dimension is 0), else
the XXX part of the name of one of the PM_TIME_XXX macros,
e.g. SEC for PM_TIME_SEC. Count scale values should be 0
(if the time dimension is 0), else ONE for PM_COUNT_ONE.
The PM_SPACE_XXX, PM_TIME_XXX and PM_COUNT_XXX macros are
defined in <pcp/pmapi.h> or pmLookupDesc(3).
When the scale is changed (but the dimension is unaltered)
the optional keyword RESCALE may be used to chose value
rescaling as per the -s command line option, but applied to
just this metric.
When changing the domain number for a metric or instance domain,
the new domain number will usually match an existing PMDA's
domain number. If this is not the case, then the new domain
number should not be randomly chosen; consult
$PCP_VAR_DIR/pmns/stdpmid for domain numbers that are already
assigned to PMDAs.
To promote the values of the per-disk IOPS metrics to 64-bit to allow
aggregation over a long time period for capacity planning, or because
the PMDA has changed to export 64-bit counters and we want to convert
old archives so they can be processed alongside new archives.
metric disk.dev.read { type -> U64 }
metric disk.dev.write { type -> U64 }
metric disk.dev.total { type -> U64 }
The instances associated with the load average metric kernel.all.load
could be renamed and renumbered by the rules below.
# for the Linux PMDA, the kernel.all.load metric is defined
# over instance domain 60.2
indom 60.2 {
inst 1 -> 60 iname "1 minute" -> "60 second"
inst 5 -> 300 iname "5 minute" -> "300 second"
inst 15 -> 900 iname "15 minute" -> "900 second"
}
If we decide to split the ``proc'' metrics out of the Linux PMDA,
this will involve changing the domain number for the PMID of these
metrics and the associated instance domains. The rules below would
rewrite an old archive to match the changes after the PMDA split.
# all Linux proc metrics are in 7 clusters
metric 60.8.* { pmid -> 123.*.* }
metric 60.9.* { pmid -> 123.*.* }
metric 60.13.* { pmid -> 123.*.* }
metric 60.24.* { pmid -> 123.*.* }
metric 60.31.* { pmid -> 123.*.* }
metric 60.32.* { pmid -> 123.*.* }
metric 60.51.* { pmid -> 123.*.* }
# only one instance domain for Linux proc metrics
indom 60.9 { indom -> 123.0 }
If the metric foo.count_em was exported as a native ``long'' then it
could be a 32-bit integer on some platforms and a 64-bit integer on
other platforms. Subsequent investigations show the value is in fact
unsigned, so the following rules could be used.
metric foo.count_em {
type if 32 -> U32
type if 64 -> U64
}
For each of the inlog and outlog archive logs, several physical files
are used.
archive.meta
metadata (metric descriptions, instance domains, etc.) for
the archive log
archive.0 initial volume of metrics values (subsequent volumes have
suffixes 1, 2, ...).
archive.index
temporal index to support rapid random access to the other
files in the archive log.
Environment variables with the prefix PCP_ are used to parameterize
the file and directory names used by PCP. On each installation, the
file /etc/pcp.conf contains the local values for these variables.
The $PCP_CONF variable may be used to specify an alternative
configuration file, as described in pcp.conf(5).
PCPIntro(1), pmdaInstance(3), pmdumplog(1), pmlogger(1),
pmlogextract(1), pmloglabel(1), pmlogreduce(1), pmLookupDesc(3),
pmns(5), pcp.conf(5) and pcp.env(5).
All error conditions detected by pmlogrewrite are reported on stderr
with textual (if sometimes terse) explanation.
Should the input archive log be corrupted (this can happen if the
pmlogger instance writing the log suddenly dies), then pmlogrewrite
will detect and report the position of the corruption in the file,
and any subsequent information from that archive log will not be
processed.
If any error is detected, pmlogrewrite will exit with a non-zero
status.
This page is part of the PCP (Performance Co-Pilot) project.
Information about the project can be found at ⟨http://www.pcp.io/⟩.
If you have a bug report for this manual page, send it to
pcp@groups.io. This page was obtained from the project's upstream
Git repository ⟨https://github.com/performancecopilot/pcp.git⟩ on
2018-02-02. (At that time, the date of the most recent commit that
was found in the repository was 2018-02-02.) If you discover any
rendering problems in this HTML version of the page, or you believe
there is a better or more up-to-date source for the page, or you have
corrections or improvements to the information in this COLOPHON
(which is not part of the original manual page), send a mail to
man-pages@man7.org
Performance Co-Pilot PMLOGREWRITE(1)
Pages that refer to this page: pmdaperfevent(1), pmlogcheck(1), pmlogextract(1), pmlogger_check(1), pmloglabel(1), pmdainit(3), LOGARCHIVE(5)
|
HTML rendering created 2018-02-02 by Michael Kerrisk, author of The Linux Programming Interface, maintainer of the Linux man-pages project. For details of in-depth Linux/UNIX system programming training courses that I teach, look here. Hosting by jambit GmbH.
|
|