Algorithm-Classifier-IsolationForest
view release on metacpan or search on metacpan
lib/Algorithm/Classifier/IsolationForest/App/Command/streamd.pm view on Meta::CPAN
package Algorithm::Classifier::IsolationForest::App::Command::streamd;
use strict;
use warnings;
use Algorithm::Classifier::IsolationForest ();
use Algorithm::Classifier::IsolationForest::Online ();
use Algorithm::Classifier::IsolationForest::App -command;
use File::Slurp qw(read_file write_file);
use File::Path qw(make_path);
use File::Basename qw(dirname);
use File::Spec ();
use Scalar::Util qw(looks_like_number);
use IO::Socket::UNIX ();
use IO::Select ();
use POSIX qw(setsid strftime);
use Errno qw();
# The daemon is a singleton per process, so its runtime state lives in
# file-scoped lexicals rather than being threaded through every helper.
my $JSON; # JSON::MaybeXS codec (required at runtime, see execute)
my $OIF; # the online model
my %OPT; # resolved options
my $LOG_FH; # log handle (STDERR in foreground without --log)
my %CONN; # fileno => { sock, inbuf, outbuf, mode }
my $DIRTY = 0; # learned anything since the last save?
my $RUN; # cleared by SIGTERM/SIGINT
my $SAVE_NOW; # set by SIGUSR1
my $REOPEN_LOG; # set by SIGHUP
# Sanity caps on per-connection buffers: a client is allowed big batch
# messages, but one that streams an endless line (or stops reading its
# replies) gets dropped instead of eating the daemon's memory.
use constant MAX_INBUF => 16 * 1024 * 1024;
use constant MAX_OUTBUF => 16 * 1024 * 1024;
sub opt_spec {
return (
[
'set=s',
'Named instance. Appended to --model-dir, and the socket/pid become <set>.sock / <set>.pid '
. 'under the run dir, so several daemons run side by side with no other flags. '
. 'Must match /\A[A-Za-z0-9+\-@_]+\z/.'
],
[
'socket=s',
'Unix domain socket to listen on; default /var/run/iforest_streamd/streamd.sock. With --set '
. 'this is instead the base run dir (default /var/run/iforest_streamd) the <set>.sock is created in.',
{ 'completion' => 'files' }
],
[
'pid=s',
'Where to write the daemon pid; default /var/run/iforest_streamd/streamd.pid. With --set '
. 'this is instead the base run dir (default /var/run/iforest_streamd) the <set>.pid is created in.',
{ 'completion' => 'files' }
],
[
'model-dir=s',
'Directory timestamped model saves land in; the symlink latest.json in it always points '
. 'at the newest, and the daemon resumes from it at startup when it exists. With --set '
. 'the set name is appended as a subdirectory.',
{ 'default' => '/var/db/iforest_streamd', 'completion' => 'files' }
],
[
'save-interval=i',
'Seconds between periodic model saves (only when learning happened).',
{ 'default' => 300 }
],
[ 'keep=i', 'Prune all but the newest N timestamped model files after each save.' ],
[ 'f|foreground', 'Do not daemonize; log to stderr unless --log is given.' ],
[
'log=s',
'Log file. Defaults to <model-dir>/streamd.log when daemonized; stderr in the foreground.',
{ 'completion' => 'files' }
],
[ 'socket-mode=s', 'Octal permissions to chmod the socket file to (e.g. 0660).' ],
[ 'threshold=f', 'Alternative decision threshold to use for the label field. 0 < $val < 1' ],
# creation knobs, used only when <model-dir>/latest.json does not exist yet
[ 'n=i', 'Number of isolation trees in the ensemble (new models only).' ],
[ 'window=i', 'Sliding window size; 0 disables forgetting (new models only).' ],
[ 'eta=i', 'max_leaf_samples: points a leaf accumulates before splitting (new models only).' ],
[ 'growth=s', "Leaf split-requirement growth, 'adaptive' or 'fixed' (new models only)." ],
[ 'subsample=f', 'Per-tree stream subsampling probability, in (0, 1] (new models only).' ],
[ 's=i', 'Seed int (new models only).' ],
[
'c=f',
'Contamination. Expected fraction of anomalies, in (0, 0.5]; the decision threshold is '
. 'relearned from the window before every save (new models only).'
],
[
't=s@',
'Feature name tag. Pass once per feature; enables the tagged (JSON object) row form '
. '(new models only).'
],
[
'mungers=s',
'JSON file of Algorithm::ToNumberMunger specs, keyed by feature tag (new models only; requires -t).',
{ 'completion' => 'files' }
],
[
'prototype=s',
'JSON prototype file to create the model from (new models only). May not be combined '
. 'with -t or --mungers. See PROTOTYPES in the module POD.',
{ 'completion' => 'files' }
],
);
} ## end sub opt_spec
sub abstract { 'Run an Online Isolation Forest scoring daemon on a Unix socket, speaking JSON lines' }
sub description {
'Runs a prequential scoring daemon around an Online Isolation Forest
model (Algorithm::Classifier::IsolationForest::Online): clients connect
to the Unix domain socket and exchange one JSON document per line.
At startup the daemon resumes from <model-dir>/latest.json when it
exists; otherwise it creates a new model from the creation knobs (-n,
--window, --eta, --growth, --subsample, -s, -c, -t, --mungers,
--prototype -- the same set `iforest stream` takes). The model is saved
to a timestamped file in --model-dir every --save-interval seconds
(only when something was learned), on SIGUSR1, on the save command, and
at shutdown; the symlink latest.json is atomically repointed at every
save, so a restart resumes the stream losing at most one interval.
Requests are JSON objects carrying exactly one of "row", "rows", or
"cmd", an optional "mode", and an optional "tag" (any JSON value,
echoed back verbatim in the reply -- a correlation tag, not to be
confused with feature tags):
{"row": [0.1, 0.7]} -> {"score": 0.41, "label": 0}
{"row": {"cpu": 0.1, "mem": 0.7}} -> {"score": 0.41, "label": 0}
{"rows": [[...], {...}], "tag": "b7"} -> {"scores": [[0.41,0], ...], "tag": "b7"}
{"rows": [[...]], "mode": "learn"} -> {"ok": {"learned": 1}}
{"cmd": "mode", "mode": "score"} -> {"ok": {"mode": "score"}}
{"cmd": "ping"} -> {"ok": "pong"}
{"cmd": "stats"} -> {"ok": {"seen": ..., ...}}
{"cmd": "save"} -> {"ok": {"saved": "oiforest-....json"}}
{"cmd": "relearn-threshold"} -> {"ok": {"threshold": 0.61}}
anything invalid -> {"error": "...", "tag": ...}
The array row form is positional (scalar mungers applied, like stream
CSV input); the object form is a tagged row and runs the full munger
plan, including expanding and combining mungers -- and, being JSON, the
raw values may safely contain commas, newlines, or any unicode.
A worked tagged example. Create the daemon around raw HTTP request
data, with mungers turning the raw values into numbers (mungers.json
here; a --prototype carrying the same schema works identically):
{ "method": { "munger": "http_method_enum", "default": -1 },
"path_len": { "munger": "length", "from": "path" },
"host_entropy": { "munger": "entropy", "from": "host" } }
iforest streamd --set web -t method -t path_len -t host_entropy \
--mungers mungers.json -c 0.05
Clients then send the raw values themselves -- note the input fields
are the munger SOURCES (method, path, host), not the feature tags,
because the plan derives path_len and host_entropy from them:
-> {"row": {"method": "GET", "path": "/index.html",
"host": "www.example.com"}, "tag": "r-1"}
<- {"score": 0.31, "label": 0, "tag": "r-1"}
-> {"row": {"method": "BREW", "path": "/aa,a\"a.php",
"host": "kq3xv9z2.biz"}, "tag": "r-2"}
<- {"score": 0.74, "label": 1, "tag": "r-2"}
The same rows work from the shell via
`iforest streamc --set web --jsonl -i rows.jsonl`.
Modes are prequential (score each row against the model as it stood, then
learn it -- the default), learn (learn only), and score (score only);
"mode" on a row/rows message overrides the connection default set by
the mode command for that message. A bad row gets an {"error": ...}
lib/Algorithm/Classifier/IsolationForest/App/Command/streamd.pm view on Meta::CPAN
$self->usage_error('--mungers requires feature tags (-t) to compile against');
}
}
if ( defined( $opt->{'prototype'} ) ) {
if ( !-f $opt->{'prototype'} ) {
$self->usage_error( '--prototype, "' . $opt->{'prototype'} . '", is not a file or does not exist' );
} elsif ( !-r $opt->{'prototype'} ) {
$self->usage_error( '--prototype, "' . $opt->{'prototype'} . '", is not readable' );
}
if ( defined( $opt->{'t'} ) || defined( $opt->{'mungers'} ) ) {
$self->usage_error(
'--prototype may not be combined with -t or --mungers; the schema comes only from the prototype');
}
} ## end if ( defined( $opt->{'prototype'} ) )
return 1;
} ## end sub validate
sub execute {
my ( $self, $opt, $args ) = @_;
# JSON::MaybeXS is required lazily so a box without it still has a
# working iforest CLI (App::Cmd loads every command module up front).
eval { require JSON::MaybeXS; 1 }
or die( 'iforest streamd requires JSON::MaybeXS for its wire protocol; install it: ' . $@ );
$JSON = JSON::MaybeXS->new( utf8 => 1, canonical => 1, allow_nonref => 0 );
%OPT = %$opt;
# --set turns --socket/--pid into base run dirs holding <set>.sock /
# <set>.pid and appends the set name to --model-dir, so several named
# daemons run side by side with no other flags. Without a set the
# flags are the socket/pid files themselves, defaulting as documented.
if ( defined $OPT{'set'} ) {
my $run = defined $OPT{'socket'} ? $OPT{'socket'} : '/var/run/iforest_streamd';
$OPT{'socket'} = File::Spec->catfile( $run, $OPT{'set'} . '.sock' );
my $prun = defined $OPT{'pid'} ? $OPT{'pid'} : '/var/run/iforest_streamd';
$OPT{'pid'} = File::Spec->catfile( $prun, $OPT{'set'} . '.pid' );
$OPT{'model_dir'} = File::Spec->catdir( $OPT{'model_dir'}, $OPT{'set'} );
} else {
$OPT{'socket'} = '/var/run/iforest_streamd/streamd.sock' unless defined $OPT{'socket'};
$OPT{'pid'} = '/var/run/iforest_streamd/streamd.pid' unless defined $OPT{'pid'};
}
# Daemonizing chdirs to /, so every path used after that point must be
# absolute -- including the socket and pid file, which are unlinked at
# shutdown.
for my $path_opt (qw(socket pid model_dir log)) {
$OPT{$path_opt} = File::Spec->rel2abs( $OPT{$path_opt} )
if defined $OPT{$path_opt};
}
# sun_path is 104 bytes on the BSDs and 108 on Linux (including the
# NUL); Socket.pm just warns and TRUNCATES an over-long path, which
# binds a socket nobody will ever find. Refuse loudly instead.
die( '--socket, "'
. $OPT{'socket'}
. '", is '
. length( $OPT{'socket'} )
. ' bytes; Unix socket paths are limited to ~104 bytes -- use a shorter path'
. "\n" )
if length( $OPT{'socket'} ) > 100;
# --- directories, before anything forks or binds -----------------------
_ensure_dir( $OPT{'model_dir'}, '--model-dir' );
_ensure_dir( dirname( $OPT{'socket'} ), '--socket' );
_ensure_dir( dirname( $OPT{'pid'} ), '--pid' );
# --- refuse to double-start ---------------------------------------------
if ( -e $OPT{'socket'} ) {
my $probe = IO::Socket::UNIX->new( Peer => $OPT{'socket'} );
die( 'another daemon is already listening on "' . $OPT{'socket'} . '"' . "\n" ) if $probe;
unlink $OPT{'socket'}; # stale socket from an unclean exit
}
if ( -f $OPT{'pid'} ) {
my $old = read_file( $OPT{'pid'} );
chomp $old if defined $old;
if ( defined $old && $old =~ /\A\d+\z/ && ( kill( 0, $old ) || $!{EPERM} ) ) {
die( 'another daemon appears to be running (pid ' . $old . ' from "' . $OPT{'pid'} . '")' . "\n" );
}
unlink $OPT{'pid'}; # stale pid file
}
# --- resume or create the model ----------------------------------------
my $latest = File::Spec->catfile( $OPT{'model_dir'}, 'latest.json' );
if ( -e $latest ) {
$OIF = Algorithm::Classifier::IsolationForest->load($latest);
die( '"' . $latest . '" is not an online model; streamd only works on those' . "\n" )
unless ref $OIF eq 'Algorithm::Classifier::IsolationForest::Online';
} elsif ( defined $OPT{'prototype'} ) {
my $proto = eval {
Algorithm::Classifier::IsolationForest->validate_prototype( scalar read_file( $OPT{'prototype'} ) );
};
die( '--prototype, "' . $OPT{'prototype'} . '", is not a valid prototype: ' . $@ ) if $@;
die( '--prototype, "' . $OPT{'prototype'} . '", is for a batch model; streamd needs an online one' . "\n" )
unless $proto->{class} eq 'online';
my %overrides;
$overrides{'n_trees'} = $OPT{'n'} if defined $OPT{'n'};
$overrides{'window_size'} = $OPT{'window'} if defined $OPT{'window'};
$overrides{'max_leaf_samples'} = $OPT{'eta'} if defined $OPT{'eta'};
$overrides{'growth'} = $OPT{'growth'} if defined $OPT{'growth'};
$overrides{'subsample'} = $OPT{'subsample'} if defined $OPT{'subsample'};
$overrides{'seed'} = $OPT{'s'} if defined $OPT{'s'};
$overrides{'contamination'} = $OPT{'c'} if defined $OPT{'c'};
$OIF = eval { Algorithm::Classifier::IsolationForest->new_from_prototype( $proto, %overrides ) };
die( '--prototype, "' . $OPT{'prototype'} . '", failed to create a model: ' . $@ ) if $@;
} else {
my $mungers;
if ( defined $OPT{'mungers'} ) {
$mungers = eval { $JSON->decode( scalar read_file( $OPT{'mungers'} ) ) };
die( '--mungers, "' . $OPT{'mungers'} . '", did not parse as JSON: ' . $@ ) if $@;
die( '--mungers, "' . $OPT{'mungers'} . '", must be a JSON object of tag => spec' )
unless ref $mungers eq 'HASH';
}
$OIF = Algorithm::Classifier::IsolationForest::Online->new(
'n_trees' => $OPT{'n'},
'window_size' => $OPT{'window'},
'max_leaf_samples' => $OPT{'eta'},
lib/Algorithm/Classifier/IsolationForest/App/Command/streamd.pm view on Meta::CPAN
# One row through the model. A JSON object is a tagged row (full munger
# plan, expanding/combining mungers included); a JSON array is positional
# (scalar mungers, like stream CSV input). Either way the final vector
# is validated numeric before it touches the model -- JSON delivers
# typed values, so anything non-numeric left after munging is a caller
# bug worth an explicit error rather than Perl's silent string-to-0.
#
# Args:
# $row :: one row, either a hashref of tagged raw values or an arrayref
# of positional ones. undef cells are left alone and defer to
# the model's missing policy.
# $mode :: 'learn', 'score' or 'prequential' -- learn only, score
# without learning, or score then learn.
#
# Returns: the anomaly score as a float, or undef in learn mode. Sets the
# dirty flag whenever the model actually learned, which is what makes the
# interval save skip a quiet period. Dies on a row that is neither shape,
# on a cell left non-numeric after munging, or on anything the model
# itself croaks about.
#
# Example:
# _apply_row( { method => 'GET', host => 'h' }, 'prequential' ); # 0.41
# _apply_row( [ 0.2, 0.7 ], 'learn' ); # undef
sub _apply_row {
my ( $row, $mode ) = @_;
my $vec;
if ( ref $row eq 'HASH' ) {
$vec = $OIF->tagged_row_to_array( $row, 'streamd' );
} elsif ( ref $row eq 'ARRAY' ) {
$vec = $row;
if ( ref $OIF->{mungers} eq 'HASH' && %{ $OIF->{mungers} } ) {
$vec = $OIF->munge_rows( [$row] )->[0];
}
} else {
die 'row must be a JSON array (positional) or object (tagged)' . "\n";
}
for my $col ( 0 .. $#$vec ) {
next if !defined $vec->[$col]; # undef defers to the model's missing policy
die 'column ' . ( $col + 1 ) . ' is not a number after munging' . "\n"
unless looks_like_number( $vec->[$col] );
}
if ( $mode eq 'learn' ) {
$OIF->learn( [$vec] );
$DIRTY = 1;
return undef;
}
if ( $mode eq 'score' ) {
return $OIF->score_samples( [$vec] )->[0];
}
my $score = $OIF->score_learn( [$vec] )->[0];
$DIRTY = 1;
return $score;
} ## end sub _apply_row
=head1 NAME
Algorithm::Classifier::IsolationForest::App::Command::streamd - Run an Online Isolation Forest scoring daemon on a Unix socket, speaking JSON lines
=head1 DESCRIPTION
Wraps the prequential loop of C<iforest stream> in a daemon: it listens
on a Unix domain socket, serves many concurrent connections from one
shared model, and exchanges one JSON document per line. Because values
travel as JSON rather than positionally, raw input headed for mungers may
safely contain commas, newlines or any unicode, and object rows run the
full munger plan -- expanding and combining mungers included, which
positional CSV cannot express.
An optional C<"tag"> on a request is echoed back verbatim. Errors are
always per-message: a bad row gets an C<{"error": ...}> reply and the
connection lives on.
Models are saved to C<--model-dir> as timestamped files every
C<--save-interval> seconds when learning has happened, plus on the C<save>
command, on SIGUSR1, and at shutdown. The C<latest.json> symlink is
repointed atomically at each save and resumed from at the next startup, so
a crash loses at most one interval of learning.
Several named instances can run side by side: C<--set> gives each its own
socket, pid file, model subdirectory and resume state.
C<iforest streamc> is the matching client.
Run it as C<iforest streamd>; C<iforest help streamd> lists every option.
=head1 METHODS
L<App::Cmd> calls these while dispatching the subcommand. Nothing else
should.
=head2 opt_spec
Returns this command's option specifications, as the list of arrayrefs
L<Getopt::Long::Descriptive> expects.
=head2 abstract
Returns the one-line summary C<iforest commands> prints beside the
command name.
=head2 description
Returns the long help text C<iforest help streamd> prints under the option
list.
=head2 validate
Checks the parsed options before anything is read or written, so a
mistake costs nothing.
Checks that C<--set> is a usable instance name, that C<--save-interval>,
C<--keep>, C<--threshold>, C<--growth> and C<--socket-mode> hold sane
values, and that a C<--mungers> spec is readable and accompanied by the
feature tags (C<-t>) it compiles against.
Takes the parsed options hashref and the arrayref of remaining
arguments. Calls C<usage_error>, which prints the usage and exits, on
the first problem it finds, and returns 1 when everything checks out.
=head2 execute
Prepares the runtime and model directories, daemonizes unless C<-f> was
( run in 1.963 second using v1.01-cache-2.11-cpan-64ef6c95b5d )