Note: This class is a Rakudo-specific feature and not standard Raku.
On creation, a Telemetry
object contains a snapshot of various aspects of the current state of the virtual machine. This is in itself useful, but generally one needs two snapshots for the difference (which is a Telemetry::Period
object).
The Telemetry object is really a collection of snapshots taken by different "instruments". By default, the Telemetry::Instrument::Usage
and Telemetry::Instrument::ThreadPool
instruments are activated.
The Telemetry
(and Telemetry::Period
) object also Associative
. This means that you can treat a Telemetry object as a read-only Hash
, with all of the data values of the instruments as keys.
You can determine which instruments Telemetry
should use by setting the $*SAMPLER
dynamic variable, which is a Telemetry::Sampler
object.
Currently, the following instruments are supported by the Rakudo core:
Telemetry::Instrument::Usage
Provides (in alphabetical order): cpu
, cpu-sys
, cpu-user
, cpus
, id-rss
, inb
, invcsw
, is-rss
, ix-rss
, majf
, max-rss
, minf
, mrcv
, msnd
, nsig
, nswp
, volcsw
, outb
, util%
and wallclock
. For complete documentation of the meaning of these data values, see Telemetry::Instrument::Usage
.
Telemetry::Instrument::Thread
Provides (in alphabetical order): tad
, tcd
, thid
, tjd
, tsd
and tys
. For complete documentation of the meaning of these data values, see Telemetry::Instrument::Thread
.
Telemetry::Instrument::ThreadPool
Provides (in alphabetical order): atc
, atq
, aw
, gtc
, gtq
, gw
, s
, ttc
, ttq
and tw
. For complete documentation of the meaning of these data values, see Telemetry::Instrument::ThreadPool
.
Telemetry::Instrument::AdHoc
Does not provide any data by itself: one must indicate which variables are to be monitored, which will then become available as methods with the same name on the instrument.
routine T§
sub T()
Shortcut for Telemetry.new
. It is exported by default. Since the Telemetry
class also provides an Associative
interface, one can easily interpolate multiple values in a single statement:
use Telemetry;say "Used (KiB CPU) so far";
routine snap§
multi snap(--> Nil)multi snap(Str --> Nil)multi snap(Str = "taking heap snapshot...", :!)multi snap( --> Nil)
The snap
subroutine is shorthand for creating a new Telemetry
object and pushing it to an array for later processing. It is exported by default. From release 2021.12, it returns the filename it's storing the snapshots in the case it's provided with a :$heap
associative parameter.
use Telemetry;my ;for ^5
If no array is specified, it will use an internal array for convenience.
routine snapper§
sub snapper( = 0.1, :, : --> Nil)
The snapper
routine starts a separate thread that will call snap
repeatedly until the end of program. It is exported by default.
By default, it will call snap
every 0.1 second. The only positional parameter is taken to be the delay between snap
s.
Please see the snapper module for externally starting a snapper without having to change the code. Simply adding -Msnapper
as a command line parameter, will then start a snapper for you.
routine periods§
multi periods( --> Seq)multi periods( --> Seq)
The periods
subroutine processes an array of Telemetry
objects and generates a Seq
of Telemetry::Period
objects out of that. It is exported by default.
.<cpu wallclock>.say for periods();# OUTPUT:# ====================# (164 / 160)# (23 / 21)# (17 / 17)# (15 / 16)# (29 / 28)
If no array is specified, it will use the internal array of snap
without parameters and will reset that array upon completion (so that new snap
s can be added again).
use Telemetry;for ^5say .<cpu wallclock>.join(" / ") for periods;# OUTPUT:# ====================# 172 / 168# 24 / 21# 17 / 18# 17 / 16# 27 / 27
If only one snap
was done, another snap
will be done to create at least one Telemetry::Period
object.
routine report§
multi report(:, :, :, :, :)
The report
subroutine generates a report about an array of Telemetry
objects. It is exported by default. These can have been created by regularly calling snap
, or by having a snapper running. If no positional parameter is used, it will assume the internal array to which the parameterless snap
pushes.
Below are the additional named parameters of report
.
:columns
Specify the names of the columns to be included in the report. Names can be specified with the column name (e.g. gw
). If not specified, defaults to what is specified in the RAKUDO_REPORT_COLUMNS
environment variable. If that is not set either, defaults to:
wallclock util% max-rss gw gtc tw ttc aw atc
:header-repeat
Specifies after how many lines the header should be repeated in the report. If not specified, defaults to what is specified in the RAKUDO_REPORT_HEADER_REPEAT
environment variable. If that is not set either, defaults to 32.
:legend
Specifies whether a legend should be added to the report. If not specified, defaults to what is specified in the RAKUDO_REPORT_LEGEND
environment variable. If that is not set either, defaults to True.
If there are snap
s available in the internal array at the end of the program, then report
will be automatically generated and printed on STDERR
.
module snapper§
Start a thread taking repeated system state snapshots.
This module contains no subroutines or methods or anything. It is intended as a shortcut for starting the snapper subroutine of the Telemetry
module, allowing taking snapshots of the execution of a program without needing to change the program. Simple loading the module with -Msnapper
will do all that is needed to start the snapper, and have a report printed on STDERR upon completion of the program.
The RAKUDO_SNAPPER
environment variable can be set to indicate the time between snapshots. If not specified, it will default to 0.1 seconds.
The snapper
module assumes an orderly shutdown of the process. Killing the process (for instance by pressing Control-c) will not produce a report.
module safe-snapper§
Available as of the 2021.09 release of the Rakudo compiler.
This module provides a safe alternative to the snapper
module: killing a process by pressing Control-c will produce a report. It is able to do so by installing a signal handler, which may interfere with normal functioning of interactive programs.
Killing a process in any other way, will not produce a report.